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_resume">'<tt>resume</tt>' Instruction</a></li>
128 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
131 <li><a href="#binaryops">Binary Operations</a>
133 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
134 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
135 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
136 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
137 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
138 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
139 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
140 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
141 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
142 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
143 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
144 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
147 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
149 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
150 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
151 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
152 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
153 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
154 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
157 <li><a href="#vectorops">Vector Operations</a>
159 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
160 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
161 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
164 <li><a href="#aggregateops">Aggregate Operations</a>
166 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
167 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
170 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
172 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
173 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
174 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
175 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
176 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
177 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
178 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
181 <li><a href="#convertops">Conversion Operations</a>
183 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
184 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
185 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
186 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
187 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
188 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
189 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
190 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
191 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
192 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
193 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
194 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
197 <li><a href="#otherops">Other Operations</a>
199 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
200 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
201 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
202 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
203 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
204 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
205 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
210 <li><a href="#intrinsics">Intrinsic Functions</a>
212 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
214 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
215 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
216 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
219 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
221 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
222 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
223 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
226 <li><a href="#int_codegen">Code Generator Intrinsics</a>
228 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
229 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
230 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
231 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
232 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
233 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
234 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
237 <li><a href="#int_libc">Standard C Library Intrinsics</a>
239 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
240 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
241 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
242 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
243 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
244 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
245 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
246 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
247 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
248 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
249 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
252 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
254 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
255 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
256 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
257 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
260 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
262 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
263 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
264 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
265 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
266 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
267 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
270 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
272 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
273 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
276 <li><a href="#int_debugger">Debugger intrinsics</a></li>
277 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
278 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
280 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
283 <li><a href="#int_atomics">Atomic intrinsics</a>
285 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
286 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
287 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
288 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
289 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
290 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
291 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
292 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
293 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
294 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
295 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
296 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
297 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
300 <li><a href="#int_memorymarkers">Memory Use Markers</a>
302 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
303 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
304 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
305 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
308 <li><a href="#int_general">General intrinsics</a>
310 <li><a href="#int_var_annotation">
311 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
312 <li><a href="#int_annotation">
313 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
314 <li><a href="#int_trap">
315 '<tt>llvm.trap</tt>' Intrinsic</a></li>
316 <li><a href="#int_stackprotector">
317 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
318 <li><a href="#int_objectsize">
319 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
326 <div class="doc_author">
327 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
328 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
331 <!-- *********************************************************************** -->
332 <h2><a name="abstract">Abstract</a></h2>
333 <!-- *********************************************************************** -->
337 <p>This document is a reference manual for the LLVM assembly language. LLVM is
338 a Static Single Assignment (SSA) based representation that provides type
339 safety, low-level operations, flexibility, and the capability of representing
340 'all' high-level languages cleanly. It is the common code representation
341 used throughout all phases of the LLVM compilation strategy.</p>
345 <!-- *********************************************************************** -->
346 <h2><a name="introduction">Introduction</a></h2>
347 <!-- *********************************************************************** -->
351 <p>The LLVM code representation is designed to be used in three different forms:
352 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
353 for fast loading by a Just-In-Time compiler), and as a human readable
354 assembly language representation. This allows LLVM to provide a powerful
355 intermediate representation for efficient compiler transformations and
356 analysis, while providing a natural means to debug and visualize the
357 transformations. The three different forms of LLVM are all equivalent. This
358 document describes the human readable representation and notation.</p>
360 <p>The LLVM representation aims to be light-weight and low-level while being
361 expressive, typed, and extensible at the same time. It aims to be a
362 "universal IR" of sorts, by being at a low enough level that high-level ideas
363 may be cleanly mapped to it (similar to how microprocessors are "universal
364 IR's", allowing many source languages to be mapped to them). By providing
365 type information, LLVM can be used as the target of optimizations: for
366 example, through pointer analysis, it can be proven that a C automatic
367 variable is never accessed outside of the current function, allowing it to
368 be promoted to a simple SSA value instead of a memory location.</p>
370 <!-- _______________________________________________________________________ -->
372 <a name="wellformed">Well-Formedness</a>
377 <p>It is important to note that this document describes 'well formed' LLVM
378 assembly language. There is a difference between what the parser accepts and
379 what is considered 'well formed'. For example, the following instruction is
380 syntactically okay, but not well formed:</p>
382 <pre class="doc_code">
383 %x = <a href="#i_add">add</a> i32 1, %x
386 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
387 LLVM infrastructure provides a verification pass that may be used to verify
388 that an LLVM module is well formed. This pass is automatically run by the
389 parser after parsing input assembly and by the optimizer before it outputs
390 bitcode. The violations pointed out by the verifier pass indicate bugs in
391 transformation passes or input to the parser.</p>
397 <!-- Describe the typesetting conventions here. -->
399 <!-- *********************************************************************** -->
400 <h2><a name="identifiers">Identifiers</a></h2>
401 <!-- *********************************************************************** -->
405 <p>LLVM identifiers come in two basic types: global and local. Global
406 identifiers (functions, global variables) begin with the <tt>'@'</tt>
407 character. Local identifiers (register names, types) begin with
408 the <tt>'%'</tt> character. Additionally, there are three different formats
409 for identifiers, for different purposes:</p>
412 <li>Named values are represented as a string of characters with their prefix.
413 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
414 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
415 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
416 other characters in their names can be surrounded with quotes. Special
417 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
418 ASCII code for the character in hexadecimal. In this way, any character
419 can be used in a name value, even quotes themselves.</li>
421 <li>Unnamed values are represented as an unsigned numeric value with their
422 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
424 <li>Constants, which are described in a <a href="#constants">section about
425 constants</a>, below.</li>
428 <p>LLVM requires that values start with a prefix for two reasons: Compilers
429 don't need to worry about name clashes with reserved words, and the set of
430 reserved words may be expanded in the future without penalty. Additionally,
431 unnamed identifiers allow a compiler to quickly come up with a temporary
432 variable without having to avoid symbol table conflicts.</p>
434 <p>Reserved words in LLVM are very similar to reserved words in other
435 languages. There are keywords for different opcodes
436 ('<tt><a href="#i_add">add</a></tt>',
437 '<tt><a href="#i_bitcast">bitcast</a></tt>',
438 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
439 ('<tt><a href="#t_void">void</a></tt>',
440 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
441 reserved words cannot conflict with variable names, because none of them
442 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
444 <p>Here is an example of LLVM code to multiply the integer variable
445 '<tt>%X</tt>' by 8:</p>
449 <pre class="doc_code">
450 %result = <a href="#i_mul">mul</a> i32 %X, 8
453 <p>After strength reduction:</p>
455 <pre class="doc_code">
456 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
459 <p>And the hard way:</p>
461 <pre class="doc_code">
462 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
463 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
464 %result = <a href="#i_add">add</a> i32 %1, %1
467 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
468 lexical features of LLVM:</p>
471 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
474 <li>Unnamed temporaries are created when the result of a computation is not
475 assigned to a named value.</li>
477 <li>Unnamed temporaries are numbered sequentially</li>
480 <p>It also shows a convention that we follow in this document. When
481 demonstrating instructions, we will follow an instruction with a comment that
482 defines the type and name of value produced. Comments are shown in italic
487 <!-- *********************************************************************** -->
488 <h2><a name="highlevel">High Level Structure</a></h2>
489 <!-- *********************************************************************** -->
491 <!-- ======================================================================= -->
493 <a name="modulestructure">Module Structure</a>
498 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
499 of the input programs. Each module consists of functions, global variables,
500 and symbol table entries. Modules may be combined together with the LLVM
501 linker, which merges function (and global variable) definitions, resolves
502 forward declarations, and merges symbol table entries. Here is an example of
503 the "hello world" module:</p>
505 <pre class="doc_code">
506 <i>; Declare the string constant as a global constant.</i>
507 <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>
509 <i>; External declaration of the puts function</i>
510 <a href="#functionstructure">declare</a> i32 @puts(i8*) <i>; i32 (i8*)* </i>
512 <i>; Definition of main function</i>
513 define i32 @main() { <i>; i32()* </i>
514 <i>; Convert [13 x i8]* to i8 *...</i>
515 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8*</i>
517 <i>; Call puts function to write out the string to stdout.</i>
518 <a href="#i_call">call</a> i32 @puts(i8* %cast210) <i>; i32</i>
519 <a href="#i_ret">ret</a> i32 0
522 <i>; Named metadata</i>
523 !1 = metadata !{i32 41}
527 <p>This example is made up of a <a href="#globalvars">global variable</a> named
528 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
529 a <a href="#functionstructure">function definition</a> for
530 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
533 <p>In general, a module is made up of a list of global values, where both
534 functions and global variables are global values. Global values are
535 represented by a pointer to a memory location (in this case, a pointer to an
536 array of char, and a pointer to a function), and have one of the
537 following <a href="#linkage">linkage types</a>.</p>
541 <!-- ======================================================================= -->
543 <a name="linkage">Linkage Types</a>
548 <p>All Global Variables and Functions have one of the following types of
552 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
553 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
554 by objects in the current module. In particular, linking code into a
555 module with an private global value may cause the private to be renamed as
556 necessary to avoid collisions. Because the symbol is private to the
557 module, all references can be updated. This doesn't show up in any symbol
558 table in the object file.</dd>
560 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
561 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
562 assembler and evaluated by the linker. Unlike normal strong symbols, they
563 are removed by the linker from the final linked image (executable or
564 dynamic library).</dd>
566 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
567 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
568 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
569 linker. The symbols are removed by the linker from the final linked image
570 (executable or dynamic library).</dd>
572 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
573 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
574 of the object is not taken. For instance, functions that had an inline
575 definition, but the compiler decided not to inline it. Note,
576 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
577 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
578 visibility. The symbols are removed by the linker from the final linked
579 image (executable or dynamic library).</dd>
581 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
582 <dd>Similar to private, but the value shows as a local symbol
583 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
584 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
586 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
587 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
588 into the object file corresponding to the LLVM module. They exist to
589 allow inlining and other optimizations to take place given knowledge of
590 the definition of the global, which is known to be somewhere outside the
591 module. Globals with <tt>available_externally</tt> linkage are allowed to
592 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
593 This linkage type is only allowed on definitions, not declarations.</dd>
595 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
596 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
597 the same name when linkage occurs. This can be used to implement
598 some forms of inline functions, templates, or other code which must be
599 generated in each translation unit that uses it, but where the body may
600 be overridden with a more definitive definition later. Unreferenced
601 <tt>linkonce</tt> globals are allowed to be discarded. Note that
602 <tt>linkonce</tt> linkage does not actually allow the optimizer to
603 inline the body of this function into callers because it doesn't know if
604 this definition of the function is the definitive definition within the
605 program or whether it will be overridden by a stronger definition.
606 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
609 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
610 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
611 <tt>linkonce</tt> linkage, except that unreferenced globals with
612 <tt>weak</tt> linkage may not be discarded. This is used for globals that
613 are declared "weak" in C source code.</dd>
615 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
616 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
617 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
619 Symbols with "<tt>common</tt>" linkage are merged in the same way as
620 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
621 <tt>common</tt> symbols may not have an explicit section,
622 must have a zero initializer, and may not be marked '<a
623 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
624 have common linkage.</dd>
627 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
628 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
629 pointer to array type. When two global variables with appending linkage
630 are linked together, the two global arrays are appended together. This is
631 the LLVM, typesafe, equivalent of having the system linker append together
632 "sections" with identical names when .o files are linked.</dd>
634 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
635 <dd>The semantics of this linkage follow the ELF object file model: the symbol
636 is weak until linked, if not linked, the symbol becomes null instead of
637 being an undefined reference.</dd>
639 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
640 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
641 <dd>Some languages allow differing globals to be merged, such as two functions
642 with different semantics. Other languages, such as <tt>C++</tt>, ensure
643 that only equivalent globals are ever merged (the "one definition rule"
644 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
645 and <tt>weak_odr</tt> linkage types to indicate that the global will only
646 be merged with equivalent globals. These linkage types are otherwise the
647 same as their non-<tt>odr</tt> versions.</dd>
649 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
650 <dd>If none of the above identifiers are used, the global is externally
651 visible, meaning that it participates in linkage and can be used to
652 resolve external symbol references.</dd>
655 <p>The next two types of linkage are targeted for Microsoft Windows platform
656 only. They are designed to support importing (exporting) symbols from (to)
657 DLLs (Dynamic Link Libraries).</p>
660 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
661 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
662 or variable via a global pointer to a pointer that is set up by the DLL
663 exporting the symbol. On Microsoft Windows targets, the pointer name is
664 formed by combining <code>__imp_</code> and the function or variable
667 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
668 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
669 pointer to a pointer in a DLL, so that it can be referenced with the
670 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
671 name is formed by combining <code>__imp_</code> and the function or
675 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
676 another module defined a "<tt>.LC0</tt>" variable and was linked with this
677 one, one of the two would be renamed, preventing a collision. Since
678 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
679 declarations), they are accessible outside of the current module.</p>
681 <p>It is illegal for a function <i>declaration</i> to have any linkage type
682 other than "externally visible", <tt>dllimport</tt>
683 or <tt>extern_weak</tt>.</p>
685 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
686 or <tt>weak_odr</tt> linkages.</p>
690 <!-- ======================================================================= -->
692 <a name="callingconv">Calling Conventions</a>
697 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
698 and <a href="#i_invoke">invokes</a> can all have an optional calling
699 convention specified for the call. The calling convention of any pair of
700 dynamic caller/callee must match, or the behavior of the program is
701 undefined. The following calling conventions are supported by LLVM, and more
702 may be added in the future:</p>
705 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
706 <dd>This calling convention (the default if no other calling convention is
707 specified) matches the target C calling conventions. This calling
708 convention supports varargs function calls and tolerates some mismatch in
709 the declared prototype and implemented declaration of the function (as
712 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
713 <dd>This calling convention attempts to make calls as fast as possible
714 (e.g. by passing things in registers). This calling convention allows the
715 target to use whatever tricks it wants to produce fast code for the
716 target, without having to conform to an externally specified ABI
717 (Application Binary Interface).
718 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
719 when this or the GHC convention is used.</a> This calling convention
720 does not support varargs and requires the prototype of all callees to
721 exactly match the prototype of the function definition.</dd>
723 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
724 <dd>This calling convention attempts to make code in the caller as efficient
725 as possible under the assumption that the call is not commonly executed.
726 As such, these calls often preserve all registers so that the call does
727 not break any live ranges in the caller side. This calling convention
728 does not support varargs and requires the prototype of all callees to
729 exactly match the prototype of the function definition.</dd>
731 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
732 <dd>This calling convention has been implemented specifically for use by the
733 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
734 It passes everything in registers, going to extremes to achieve this by
735 disabling callee save registers. This calling convention should not be
736 used lightly but only for specific situations such as an alternative to
737 the <em>register pinning</em> performance technique often used when
738 implementing functional programming languages.At the moment only X86
739 supports this convention and it has the following limitations:
741 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
742 floating point types are supported.</li>
743 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
744 6 floating point parameters.</li>
746 This calling convention supports
747 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
748 requires both the caller and callee are using it.
751 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
752 <dd>Any calling convention may be specified by number, allowing
753 target-specific calling conventions to be used. Target specific calling
754 conventions start at 64.</dd>
757 <p>More calling conventions can be added/defined on an as-needed basis, to
758 support Pascal conventions or any other well-known target-independent
763 <!-- ======================================================================= -->
765 <a name="visibility">Visibility Styles</a>
770 <p>All Global Variables and Functions have one of the following visibility
774 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
775 <dd>On targets that use the ELF object file format, default visibility means
776 that the declaration is visible to other modules and, in shared libraries,
777 means that the declared entity may be overridden. On Darwin, default
778 visibility means that the declaration is visible to other modules. Default
779 visibility corresponds to "external linkage" in the language.</dd>
781 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
782 <dd>Two declarations of an object with hidden visibility refer to the same
783 object if they are in the same shared object. Usually, hidden visibility
784 indicates that the symbol will not be placed into the dynamic symbol
785 table, so no other module (executable or shared library) can reference it
788 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
789 <dd>On ELF, protected visibility indicates that the symbol will be placed in
790 the dynamic symbol table, but that references within the defining module
791 will bind to the local symbol. That is, the symbol cannot be overridden by
797 <!-- ======================================================================= -->
799 <a name="namedtypes">Named Types</a>
804 <p>LLVM IR allows you to specify name aliases for certain types. This can make
805 it easier to read the IR and make the IR more condensed (particularly when
806 recursive types are involved). An example of a name specification is:</p>
808 <pre class="doc_code">
809 %mytype = type { %mytype*, i32 }
812 <p>You may give a name to any <a href="#typesystem">type</a> except
813 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
814 is expected with the syntax "%mytype".</p>
816 <p>Note that type names are aliases for the structural type that they indicate,
817 and that you can therefore specify multiple names for the same type. This
818 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
819 uses structural typing, the name is not part of the type. When printing out
820 LLVM IR, the printer will pick <em>one name</em> to render all types of a
821 particular shape. This means that if you have code where two different
822 source types end up having the same LLVM type, that the dumper will sometimes
823 print the "wrong" or unexpected type. This is an important design point and
824 isn't going to change.</p>
828 <!-- ======================================================================= -->
830 <a name="globalvars">Global Variables</a>
835 <p>Global variables define regions of memory allocated at compilation time
836 instead of run-time. Global variables may optionally be initialized, may
837 have an explicit section to be placed in, and may have an optional explicit
838 alignment specified. A variable may be defined as "thread_local", which
839 means that it will not be shared by threads (each thread will have a
840 separated copy of the variable). A variable may be defined as a global
841 "constant," which indicates that the contents of the variable
842 will <b>never</b> be modified (enabling better optimization, allowing the
843 global data to be placed in the read-only section of an executable, etc).
844 Note that variables that need runtime initialization cannot be marked
845 "constant" as there is a store to the variable.</p>
847 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
848 constant, even if the final definition of the global is not. This capability
849 can be used to enable slightly better optimization of the program, but
850 requires the language definition to guarantee that optimizations based on the
851 'constantness' are valid for the translation units that do not include the
854 <p>As SSA values, global variables define pointer values that are in scope
855 (i.e. they dominate) all basic blocks in the program. Global variables
856 always define a pointer to their "content" type because they describe a
857 region of memory, and all memory objects in LLVM are accessed through
860 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
861 that the address is not significant, only the content. Constants marked
862 like this can be merged with other constants if they have the same
863 initializer. Note that a constant with significant address <em>can</em>
864 be merged with a <tt>unnamed_addr</tt> constant, the result being a
865 constant whose address is significant.</p>
867 <p>A global variable may be declared to reside in a target-specific numbered
868 address space. For targets that support them, address spaces may affect how
869 optimizations are performed and/or what target instructions are used to
870 access the variable. The default address space is zero. The address space
871 qualifier must precede any other attributes.</p>
873 <p>LLVM allows an explicit section to be specified for globals. If the target
874 supports it, it will emit globals to the section specified.</p>
876 <p>An explicit alignment may be specified for a global, which must be a power
877 of 2. If not present, or if the alignment is set to zero, the alignment of
878 the global is set by the target to whatever it feels convenient. If an
879 explicit alignment is specified, the global is forced to have exactly that
880 alignment. Targets and optimizers are not allowed to over-align the global
881 if the global has an assigned section. In this case, the extra alignment
882 could be observable: for example, code could assume that the globals are
883 densely packed in their section and try to iterate over them as an array,
884 alignment padding would break this iteration.</p>
886 <p>For example, the following defines a global in a numbered address space with
887 an initializer, section, and alignment:</p>
889 <pre class="doc_code">
890 @G = addrspace(5) constant float 1.0, section "foo", align 4
896 <!-- ======================================================================= -->
898 <a name="functionstructure">Functions</a>
903 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
904 optional <a href="#linkage">linkage type</a>, an optional
905 <a href="#visibility">visibility style</a>, an optional
906 <a href="#callingconv">calling convention</a>,
907 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
908 <a href="#paramattrs">parameter attribute</a> for the return type, a function
909 name, a (possibly empty) argument list (each with optional
910 <a href="#paramattrs">parameter attributes</a>), optional
911 <a href="#fnattrs">function attributes</a>, an optional section, an optional
912 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
913 curly brace, a list of basic blocks, and a closing curly brace.</p>
915 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
916 optional <a href="#linkage">linkage type</a>, an optional
917 <a href="#visibility">visibility style</a>, an optional
918 <a href="#callingconv">calling convention</a>,
919 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
920 <a href="#paramattrs">parameter attribute</a> for the return type, a function
921 name, a possibly empty list of arguments, an optional alignment, and an
922 optional <a href="#gc">garbage collector name</a>.</p>
924 <p>A function definition contains a list of basic blocks, forming the CFG
925 (Control Flow Graph) for the function. Each basic block may optionally start
926 with a label (giving the basic block a symbol table entry), contains a list
927 of instructions, and ends with a <a href="#terminators">terminator</a>
928 instruction (such as a branch or function return).</p>
930 <p>The first basic block in a function is special in two ways: it is immediately
931 executed on entrance to the function, and it is not allowed to have
932 predecessor basic blocks (i.e. there can not be any branches to the entry
933 block of a function). Because the block can have no predecessors, it also
934 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
936 <p>LLVM allows an explicit section to be specified for functions. If the target
937 supports it, it will emit functions to the section specified.</p>
939 <p>An explicit alignment may be specified for a function. If not present, or if
940 the alignment is set to zero, the alignment of the function is set by the
941 target to whatever it feels convenient. If an explicit alignment is
942 specified, the function is forced to have at least that much alignment. All
943 alignments must be a power of 2.</p>
945 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
946 be significant and two identical functions can be merged</p>.
949 <pre class="doc_code">
950 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
951 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
952 <ResultType> @<FunctionName> ([argument list])
953 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
954 [<a href="#gc">gc</a>] { ... }
959 <!-- ======================================================================= -->
961 <a name="aliasstructure">Aliases</a>
966 <p>Aliases act as "second name" for the aliasee value (which can be either
967 function, global variable, another alias or bitcast of global value). Aliases
968 may have an optional <a href="#linkage">linkage type</a>, and an
969 optional <a href="#visibility">visibility style</a>.</p>
972 <pre class="doc_code">
973 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
978 <!-- ======================================================================= -->
980 <a name="namedmetadatastructure">Named Metadata</a>
985 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
986 nodes</a> (but not metadata strings) are the only valid operands for
987 a named metadata.</p>
990 <pre class="doc_code">
991 ; Some unnamed metadata nodes, which are referenced by the named metadata.
992 !0 = metadata !{metadata !"zero"}
993 !1 = metadata !{metadata !"one"}
994 !2 = metadata !{metadata !"two"}
996 !name = !{!0, !1, !2}
1001 <!-- ======================================================================= -->
1003 <a name="paramattrs">Parameter Attributes</a>
1008 <p>The return type and each parameter of a function type may have a set of
1009 <i>parameter attributes</i> associated with them. Parameter attributes are
1010 used to communicate additional information about the result or parameters of
1011 a function. Parameter attributes are considered to be part of the function,
1012 not of the function type, so functions with different parameter attributes
1013 can have the same function type.</p>
1015 <p>Parameter attributes are simple keywords that follow the type specified. If
1016 multiple parameter attributes are needed, they are space separated. For
1019 <pre class="doc_code">
1020 declare i32 @printf(i8* noalias nocapture, ...)
1021 declare i32 @atoi(i8 zeroext)
1022 declare signext i8 @returns_signed_char()
1025 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1026 <tt>readonly</tt>) come immediately after the argument list.</p>
1028 <p>Currently, only the following parameter attributes are defined:</p>
1031 <dt><tt><b>zeroext</b></tt></dt>
1032 <dd>This indicates to the code generator that the parameter or return value
1033 should be zero-extended to the extent required by the target's ABI (which
1034 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1035 parameter) or the callee (for a return value).</dd>
1037 <dt><tt><b>signext</b></tt></dt>
1038 <dd>This indicates to the code generator that the parameter or return value
1039 should be sign-extended to the extent required by the target's ABI (which
1040 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1043 <dt><tt><b>inreg</b></tt></dt>
1044 <dd>This indicates that this parameter or return value should be treated in a
1045 special target-dependent fashion during while emitting code for a function
1046 call or return (usually, by putting it in a register as opposed to memory,
1047 though some targets use it to distinguish between two different kinds of
1048 registers). Use of this attribute is target-specific.</dd>
1050 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1051 <dd><p>This indicates that the pointer parameter should really be passed by
1052 value to the function. The attribute implies that a hidden copy of the
1054 is made between the caller and the callee, so the callee is unable to
1055 modify the value in the callee. This attribute is only valid on LLVM
1056 pointer arguments. It is generally used to pass structs and arrays by
1057 value, but is also valid on pointers to scalars. The copy is considered
1058 to belong to the caller not the callee (for example,
1059 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1060 <tt>byval</tt> parameters). This is not a valid attribute for return
1063 <p>The byval attribute also supports specifying an alignment with
1064 the align attribute. It indicates the alignment of the stack slot to
1065 form and the known alignment of the pointer specified to the call site. If
1066 the alignment is not specified, then the code generator makes a
1067 target-specific assumption.</p></dd>
1069 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1070 <dd>This indicates that the pointer parameter specifies the address of a
1071 structure that is the return value of the function in the source program.
1072 This pointer must be guaranteed by the caller to be valid: loads and
1073 stores to the structure may be assumed by the callee to not to trap. This
1074 may only be applied to the first parameter. This is not a valid attribute
1075 for return values. </dd>
1077 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1078 <dd>This indicates that pointer values
1079 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1080 value do not alias pointer values which are not <i>based</i> on it,
1081 ignoring certain "irrelevant" dependencies.
1082 For a call to the parent function, dependencies between memory
1083 references from before or after the call and from those during the call
1084 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1085 return value used in that call.
1086 The caller shares the responsibility with the callee for ensuring that
1087 these requirements are met.
1088 For further details, please see the discussion of the NoAlias response in
1089 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1091 Note that this definition of <tt>noalias</tt> is intentionally
1092 similar to the definition of <tt>restrict</tt> in C99 for function
1093 arguments, though it is slightly weaker.
1095 For function return values, C99's <tt>restrict</tt> is not meaningful,
1096 while LLVM's <tt>noalias</tt> is.
1099 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1100 <dd>This indicates that the callee does not make any copies of the pointer
1101 that outlive the callee itself. This is not a valid attribute for return
1104 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1105 <dd>This indicates that the pointer parameter can be excised using the
1106 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1107 attribute for return values.</dd>
1112 <!-- ======================================================================= -->
1114 <a name="gc">Garbage Collector Names</a>
1119 <p>Each function may specify a garbage collector name, which is simply a
1122 <pre class="doc_code">
1123 define void @f() gc "name" { ... }
1126 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1127 collector which will cause the compiler to alter its output in order to
1128 support the named garbage collection algorithm.</p>
1132 <!-- ======================================================================= -->
1134 <a name="fnattrs">Function Attributes</a>
1139 <p>Function attributes are set to communicate additional information about a
1140 function. Function attributes are considered to be part of the function, not
1141 of the function type, so functions with different parameter attributes can
1142 have the same function type.</p>
1144 <p>Function attributes are simple keywords that follow the type specified. If
1145 multiple attributes are needed, they are space separated. For example:</p>
1147 <pre class="doc_code">
1148 define void @f() noinline { ... }
1149 define void @f() alwaysinline { ... }
1150 define void @f() alwaysinline optsize { ... }
1151 define void @f() optsize { ... }
1155 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1156 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1157 the backend should forcibly align the stack pointer. Specify the
1158 desired alignment, which must be a power of two, in parentheses.
1160 <dt><tt><b>alwaysinline</b></tt></dt>
1161 <dd>This attribute indicates that the inliner should attempt to inline this
1162 function into callers whenever possible, ignoring any active inlining size
1163 threshold for this caller.</dd>
1165 <dt><tt><b>hotpatch</b></tt></dt>
1166 <dd>This attribute indicates that the function should be 'hotpatchable',
1167 meaning the function can be patched and/or hooked even while it is
1168 loaded into memory. On x86, the function prologue will be preceded
1169 by six bytes of padding and will begin with a two-byte instruction.
1170 Most of the functions in the Windows system DLLs in Windows XP SP2 or
1171 higher were compiled in this fashion.</dd>
1173 <dt><tt><b>nonlazybind</b></tt></dt>
1174 <dd>This attribute suppresses lazy symbol binding for the function. This
1175 may make calls to the function faster, at the cost of extra program
1176 startup time if the function is not called during program startup.</dd>
1178 <dt><tt><b>inlinehint</b></tt></dt>
1179 <dd>This attribute indicates that the source code contained a hint that inlining
1180 this function is desirable (such as the "inline" keyword in C/C++). It
1181 is just a hint; it imposes no requirements on the inliner.</dd>
1183 <dt><tt><b>naked</b></tt></dt>
1184 <dd>This attribute disables prologue / epilogue emission for the function.
1185 This can have very system-specific consequences.</dd>
1187 <dt><tt><b>noimplicitfloat</b></tt></dt>
1188 <dd>This attributes disables implicit floating point instructions.</dd>
1190 <dt><tt><b>noinline</b></tt></dt>
1191 <dd>This attribute indicates that the inliner should never inline this
1192 function in any situation. This attribute may not be used together with
1193 the <tt>alwaysinline</tt> attribute.</dd>
1195 <dt><tt><b>noredzone</b></tt></dt>
1196 <dd>This attribute indicates that the code generator should not use a red
1197 zone, even if the target-specific ABI normally permits it.</dd>
1199 <dt><tt><b>noreturn</b></tt></dt>
1200 <dd>This function attribute indicates that the function never returns
1201 normally. This produces undefined behavior at runtime if the function
1202 ever does dynamically return.</dd>
1204 <dt><tt><b>nounwind</b></tt></dt>
1205 <dd>This function attribute indicates that the function never returns with an
1206 unwind or exceptional control flow. If the function does unwind, its
1207 runtime behavior is undefined.</dd>
1209 <dt><tt><b>optsize</b></tt></dt>
1210 <dd>This attribute suggests that optimization passes and code generator passes
1211 make choices that keep the code size of this function low, and otherwise
1212 do optimizations specifically to reduce code size.</dd>
1214 <dt><tt><b>readnone</b></tt></dt>
1215 <dd>This attribute indicates that the function computes its result (or decides
1216 to unwind an exception) based strictly on its arguments, without
1217 dereferencing any pointer arguments or otherwise accessing any mutable
1218 state (e.g. memory, control registers, etc) visible to caller functions.
1219 It does not write through any pointer arguments
1220 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1221 changes any state visible to callers. This means that it cannot unwind
1222 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1223 could use the <tt>unwind</tt> instruction.</dd>
1225 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1226 <dd>This attribute indicates that the function does not write through any
1227 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1228 arguments) or otherwise modify any state (e.g. memory, control registers,
1229 etc) visible to caller functions. It may dereference pointer arguments
1230 and read state that may be set in the caller. A readonly function always
1231 returns the same value (or unwinds an exception identically) when called
1232 with the same set of arguments and global state. It cannot unwind an
1233 exception by calling the <tt>C++</tt> exception throwing methods, but may
1234 use the <tt>unwind</tt> instruction.</dd>
1236 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1237 <dd>This attribute indicates that the function should emit a stack smashing
1238 protector. It is in the form of a "canary"—a random value placed on
1239 the stack before the local variables that's checked upon return from the
1240 function to see if it has been overwritten. A heuristic is used to
1241 determine if a function needs stack protectors or not.<br>
1243 If a function that has an <tt>ssp</tt> attribute is inlined into a
1244 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1245 function will have an <tt>ssp</tt> attribute.</dd>
1247 <dt><tt><b>sspreq</b></tt></dt>
1248 <dd>This attribute indicates that the function should <em>always</em> emit a
1249 stack smashing protector. This overrides
1250 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1252 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1253 function that doesn't have an <tt>sspreq</tt> attribute or which has
1254 an <tt>ssp</tt> attribute, then the resulting function will have
1255 an <tt>sspreq</tt> attribute.</dd>
1257 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1258 <dd>This attribute indicates that the ABI being targeted requires that
1259 an unwind table entry be produce for this function even if we can
1260 show that no exceptions passes by it. This is normally the case for
1261 the ELF x86-64 abi, but it can be disabled for some compilation
1268 <!-- ======================================================================= -->
1270 <a name="moduleasm">Module-Level Inline Assembly</a>
1275 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1276 the GCC "file scope inline asm" blocks. These blocks are internally
1277 concatenated by LLVM and treated as a single unit, but may be separated in
1278 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1280 <pre class="doc_code">
1281 module asm "inline asm code goes here"
1282 module asm "more can go here"
1285 <p>The strings can contain any character by escaping non-printable characters.
1286 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1289 <p>The inline asm code is simply printed to the machine code .s file when
1290 assembly code is generated.</p>
1294 <!-- ======================================================================= -->
1296 <a name="datalayout">Data Layout</a>
1301 <p>A module may specify a target specific data layout string that specifies how
1302 data is to be laid out in memory. The syntax for the data layout is
1305 <pre class="doc_code">
1306 target datalayout = "<i>layout specification</i>"
1309 <p>The <i>layout specification</i> consists of a list of specifications
1310 separated by the minus sign character ('-'). Each specification starts with
1311 a letter and may include other information after the letter to define some
1312 aspect of the data layout. The specifications accepted are as follows:</p>
1316 <dd>Specifies that the target lays out data in big-endian form. That is, the
1317 bits with the most significance have the lowest address location.</dd>
1320 <dd>Specifies that the target lays out data in little-endian form. That is,
1321 the bits with the least significance have the lowest address
1324 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1325 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1326 <i>preferred</i> alignments. All sizes are in bits. Specifying
1327 the <i>pref</i> alignment is optional. If omitted, the
1328 preceding <tt>:</tt> should be omitted too.</dd>
1330 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1331 <dd>This specifies the alignment for an integer type of a given bit
1332 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1334 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1335 <dd>This specifies the alignment for a vector type of a given bit
1338 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1339 <dd>This specifies the alignment for a floating point type of a given bit
1340 <i>size</i>. Only values of <i>size</i> that are supported by the target
1341 will work. 32 (float) and 64 (double) are supported on all targets;
1342 80 or 128 (different flavors of long double) are also supported on some
1345 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1346 <dd>This specifies the alignment for an aggregate type of a given bit
1349 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1350 <dd>This specifies the alignment for a stack object of a given bit
1353 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1354 <dd>This specifies a set of native integer widths for the target CPU
1355 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1356 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1357 this set are considered to support most general arithmetic
1358 operations efficiently.</dd>
1361 <p>When constructing the data layout for a given target, LLVM starts with a
1362 default set of specifications which are then (possibly) overridden by the
1363 specifications in the <tt>datalayout</tt> keyword. The default specifications
1364 are given in this list:</p>
1367 <li><tt>E</tt> - big endian</li>
1368 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1369 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1370 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1371 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1372 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1373 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1374 alignment of 64-bits</li>
1375 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1376 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1377 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1378 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1379 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1380 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1383 <p>When LLVM is determining the alignment for a given type, it uses the
1384 following rules:</p>
1387 <li>If the type sought is an exact match for one of the specifications, that
1388 specification is used.</li>
1390 <li>If no match is found, and the type sought is an integer type, then the
1391 smallest integer type that is larger than the bitwidth of the sought type
1392 is used. If none of the specifications are larger than the bitwidth then
1393 the the largest integer type is used. For example, given the default
1394 specifications above, the i7 type will use the alignment of i8 (next
1395 largest) while both i65 and i256 will use the alignment of i64 (largest
1398 <li>If no match is found, and the type sought is a vector type, then the
1399 largest vector type that is smaller than the sought vector type will be
1400 used as a fall back. This happens because <128 x double> can be
1401 implemented in terms of 64 <2 x double>, for example.</li>
1406 <!-- ======================================================================= -->
1408 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1413 <p>Any memory access must be done through a pointer value associated
1414 with an address range of the memory access, otherwise the behavior
1415 is undefined. Pointer values are associated with address ranges
1416 according to the following rules:</p>
1419 <li>A pointer value is associated with the addresses associated with
1420 any value it is <i>based</i> on.
1421 <li>An address of a global variable is associated with the address
1422 range of the variable's storage.</li>
1423 <li>The result value of an allocation instruction is associated with
1424 the address range of the allocated storage.</li>
1425 <li>A null pointer in the default address-space is associated with
1427 <li>An integer constant other than zero or a pointer value returned
1428 from a function not defined within LLVM may be associated with address
1429 ranges allocated through mechanisms other than those provided by
1430 LLVM. Such ranges shall not overlap with any ranges of addresses
1431 allocated by mechanisms provided by LLVM.</li>
1434 <p>A pointer value is <i>based</i> on another pointer value according
1435 to the following rules:</p>
1438 <li>A pointer value formed from a
1439 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1440 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1441 <li>The result value of a
1442 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1443 of the <tt>bitcast</tt>.</li>
1444 <li>A pointer value formed by an
1445 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1446 pointer values that contribute (directly or indirectly) to the
1447 computation of the pointer's value.</li>
1448 <li>The "<i>based</i> on" relationship is transitive.</li>
1451 <p>Note that this definition of <i>"based"</i> is intentionally
1452 similar to the definition of <i>"based"</i> in C99, though it is
1453 slightly weaker.</p>
1455 <p>LLVM IR does not associate types with memory. The result type of a
1456 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1457 alignment of the memory from which to load, as well as the
1458 interpretation of the value. The first operand type of a
1459 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1460 and alignment of the store.</p>
1462 <p>Consequently, type-based alias analysis, aka TBAA, aka
1463 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1464 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1465 additional information which specialized optimization passes may use
1466 to implement type-based alias analysis.</p>
1470 <!-- ======================================================================= -->
1472 <a name="volatile">Volatile Memory Accesses</a>
1477 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1478 href="#i_store"><tt>store</tt></a>s, and <a
1479 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1480 The optimizers must not change the number of volatile operations or change their
1481 order of execution relative to other volatile operations. The optimizers
1482 <i>may</i> change the order of volatile operations relative to non-volatile
1483 operations. This is not Java's "volatile" and has no cross-thread
1484 synchronization behavior.</p>
1488 <!-- ======================================================================= -->
1490 <a name="memmodel">Memory Model for Concurrent Operations</a>
1495 <p>The LLVM IR does not define any way to start parallel threads of execution
1496 or to register signal handlers. Nonetheless, there are platform-specific
1497 ways to create them, and we define LLVM IR's behavior in their presence. This
1498 model is inspired by the C++0x memory model.</p>
1500 <p>We define a <i>happens-before</i> partial order as the least partial order
1503 <li>Is a superset of single-thread program order, and</li>
1504 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1505 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1506 by platform-specific techniques, like pthread locks, thread
1507 creation, thread joining, etc., and by atomic instructions.
1508 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1512 <p>Note that program order does not introduce <i>happens-before</i> edges
1513 between a thread and signals executing inside that thread.</p>
1515 <p>Every (defined) read operation (load instructions, memcpy, atomic
1516 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1517 (defined) write operations (store instructions, atomic
1518 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1519 initialized globals are considered to have a write of the initializer which is
1520 atomic and happens before any other read or write of the memory in question.
1521 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1522 any write to the same byte, except:</p>
1525 <li>If <var>write<sub>1</sub></var> happens before
1526 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1527 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1528 does not see <var>write<sub>1</sub></var>.
1529 <li>If <var>R<sub>byte</sub></var> happens before
1530 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1531 see <var>write<sub>3</sub></var>.
1534 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1536 <li>If there is no write to the same byte that happens before
1537 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1538 <tt>undef</tt> for that byte.
1539 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1540 <var>R<sub>byte</sub></var> returns the value written by that
1542 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1543 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1544 values written. See the <a href="#ordering">Atomic Memory Ordering
1545 Constraints</a> section for additional constraints on how the choice
1547 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1550 <p><var>R</var> returns the value composed of the series of bytes it read.
1551 This implies that some bytes within the value may be <tt>undef</tt>
1552 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1553 defines the semantics of the operation; it doesn't mean that targets will
1554 emit more than one instruction to read the series of bytes.</p>
1556 <p>Note that in cases where none of the atomic intrinsics are used, this model
1557 places only one restriction on IR transformations on top of what is required
1558 for single-threaded execution: introducing a store to a byte which might not
1559 otherwise be stored is not allowed in general. (Specifically, in the case
1560 where another thread might write to and read from an address, introducing a
1561 store can change a load that may see exactly one write into a load that may
1562 see multiple writes.)</p>
1564 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1565 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1566 none of the backends currently in the tree fall into this category; however,
1567 there might be targets which care. If there are, we want a paragraph
1570 Targets may specify that stores narrower than a certain width are not
1571 available; on such a target, for the purposes of this model, treat any
1572 non-atomic write with an alignment or width less than the minimum width
1573 as if it writes to the relevant surrounding bytes.
1578 <!-- ======================================================================= -->
1580 <a name="ordering">Atomic Memory Ordering Constraints</a>
1585 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1586 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1587 <a href="#i_fence"><code>fence</code></a>,
1588 <a href="#i_load"><code>atomic load</code></a>, and
1589 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1590 that determines which other atomic instructions on the same address they
1591 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1592 but are somewhat more colloquial. If these descriptions aren't precise enough,
1593 check those specs. <a href="#i_fence"><code>fence</code></a> instructions
1594 treat these orderings somewhat differently since they don't take an address.
1595 See that instruction's documentation for details.</p>
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 inserted
2035 as defined by the TargetData string in the module, which is required to match
2036 what the underlying processor expects.</p>
2038 <p>Structures can either be "literal" or "identified". A literal structure is
2039 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2040 types are always defined at the top level with a name. Literal types are
2041 uniqued by their contents and can never be recursive or opaque since there is
2042 no way to write one. Identified types can be recursive, can be opaqued, and are
2048 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2049 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2053 <table class="layout">
2055 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2056 <td class="left">A triple of three <tt>i32</tt> values</td>
2059 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2060 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2061 second element is a <a href="#t_pointer">pointer</a> to a
2062 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2063 an <tt>i32</tt>.</td>
2066 <td class="left"><tt><{ i8, i32 }></tt></td>
2067 <td class="left">A packed struct known to be 5 bytes in size.</td>
2073 <!-- _______________________________________________________________________ -->
2075 <a name="t_opaque">Opaque Structure Types</a>
2081 <p>Opaque structure types are used to represent named structure types that do
2082 not have a body specified. This corresponds (for example) to the C notion of
2083 a forward declared structure.</p>
2092 <table class="layout">
2094 <td class="left"><tt>opaque</tt></td>
2095 <td class="left">An opaque type.</td>
2103 <!-- _______________________________________________________________________ -->
2105 <a name="t_pointer">Pointer Type</a>
2111 <p>The pointer type is used to specify memory locations.
2112 Pointers are commonly used to reference objects in memory.</p>
2114 <p>Pointer types may have an optional address space attribute defining the
2115 numbered address space where the pointed-to object resides. The default
2116 address space is number zero. The semantics of non-zero address
2117 spaces are target-specific.</p>
2119 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2120 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2128 <table class="layout">
2130 <td class="left"><tt>[4 x i32]*</tt></td>
2131 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2132 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2135 <td class="left"><tt>i32 (i32*) *</tt></td>
2136 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2137 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2141 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2142 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2143 that resides in address space #5.</td>
2149 <!-- _______________________________________________________________________ -->
2151 <a name="t_vector">Vector Type</a>
2157 <p>A vector type is a simple derived type that represents a vector of elements.
2158 Vector types are used when multiple primitive data are operated in parallel
2159 using a single instruction (SIMD). A vector type requires a size (number of
2160 elements) and an underlying primitive data type. Vector types are considered
2161 <a href="#t_firstclass">first class</a>.</p>
2165 < <# elements> x <elementtype> >
2168 <p>The number of elements is a constant integer value larger than 0; elementtype
2169 may be any integer or floating point type. Vectors of size zero are not
2170 allowed, and pointers are not allowed as the element type.</p>
2173 <table class="layout">
2175 <td class="left"><tt><4 x i32></tt></td>
2176 <td class="left">Vector of 4 32-bit integer values.</td>
2179 <td class="left"><tt><8 x float></tt></td>
2180 <td class="left">Vector of 8 32-bit floating-point values.</td>
2183 <td class="left"><tt><2 x i64></tt></td>
2184 <td class="left">Vector of 2 64-bit integer values.</td>
2192 <!-- *********************************************************************** -->
2193 <h2><a name="constants">Constants</a></h2>
2194 <!-- *********************************************************************** -->
2198 <p>LLVM has several different basic types of constants. This section describes
2199 them all and their syntax.</p>
2201 <!-- ======================================================================= -->
2203 <a name="simpleconstants">Simple Constants</a>
2209 <dt><b>Boolean constants</b></dt>
2210 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2211 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2213 <dt><b>Integer constants</b></dt>
2214 <dd>Standard integers (such as '4') are constants of
2215 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2216 with integer types.</dd>
2218 <dt><b>Floating point constants</b></dt>
2219 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2220 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2221 notation (see below). The assembler requires the exact decimal value of a
2222 floating-point constant. For example, the assembler accepts 1.25 but
2223 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2224 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2226 <dt><b>Null pointer constants</b></dt>
2227 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2228 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2231 <p>The one non-intuitive notation for constants is the hexadecimal form of
2232 floating point constants. For example, the form '<tt>double
2233 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2234 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2235 constants are required (and the only time that they are generated by the
2236 disassembler) is when a floating point constant must be emitted but it cannot
2237 be represented as a decimal floating point number in a reasonable number of
2238 digits. For example, NaN's, infinities, and other special values are
2239 represented in their IEEE hexadecimal format so that assembly and disassembly
2240 do not cause any bits to change in the constants.</p>
2242 <p>When using the hexadecimal form, constants of types float and double are
2243 represented using the 16-digit form shown above (which matches the IEEE754
2244 representation for double); float values must, however, be exactly
2245 representable as IEE754 single precision. Hexadecimal format is always used
2246 for long double, and there are three forms of long double. The 80-bit format
2247 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2248 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2249 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2250 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2251 currently supported target uses this format. Long doubles will only work if
2252 they match the long double format on your target. All hexadecimal formats
2253 are big-endian (sign bit at the left).</p>
2255 <p>There are no constants of type x86mmx.</p>
2258 <!-- ======================================================================= -->
2260 <a name="aggregateconstants"></a> <!-- old anchor -->
2261 <a name="complexconstants">Complex Constants</a>
2266 <p>Complex constants are a (potentially recursive) combination of simple
2267 constants and smaller complex constants.</p>
2270 <dt><b>Structure constants</b></dt>
2271 <dd>Structure constants are represented with notation similar to structure
2272 type definitions (a comma separated list of elements, surrounded by braces
2273 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2274 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2275 Structure constants must have <a href="#t_struct">structure type</a>, and
2276 the number and types of elements must match those specified by the
2279 <dt><b>Array constants</b></dt>
2280 <dd>Array constants are represented with notation similar to array type
2281 definitions (a comma separated list of elements, surrounded by square
2282 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2283 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2284 the number and types of elements must match those specified by the
2287 <dt><b>Vector constants</b></dt>
2288 <dd>Vector constants are represented with notation similar to vector type
2289 definitions (a comma separated list of elements, surrounded by
2290 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2291 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2292 have <a href="#t_vector">vector type</a>, and the number and types of
2293 elements must match those specified by the type.</dd>
2295 <dt><b>Zero initialization</b></dt>
2296 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2297 value to zero of <em>any</em> type, including scalar and
2298 <a href="#t_aggregate">aggregate</a> types.
2299 This is often used to avoid having to print large zero initializers
2300 (e.g. for large arrays) and is always exactly equivalent to using explicit
2301 zero initializers.</dd>
2303 <dt><b>Metadata node</b></dt>
2304 <dd>A metadata node is a structure-like constant with
2305 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2306 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2307 be interpreted as part of the instruction stream, metadata is a place to
2308 attach additional information such as debug info.</dd>
2313 <!-- ======================================================================= -->
2315 <a name="globalconstants">Global Variable and Function Addresses</a>
2320 <p>The addresses of <a href="#globalvars">global variables</a>
2321 and <a href="#functionstructure">functions</a> are always implicitly valid
2322 (link-time) constants. These constants are explicitly referenced when
2323 the <a href="#identifiers">identifier for the global</a> is used and always
2324 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2325 legal LLVM file:</p>
2327 <pre class="doc_code">
2330 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2335 <!-- ======================================================================= -->
2337 <a name="undefvalues">Undefined Values</a>
2342 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2343 indicates that the user of the value may receive an unspecified bit-pattern.
2344 Undefined values may be of any type (other than '<tt>label</tt>'
2345 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2347 <p>Undefined values are useful because they indicate to the compiler that the
2348 program is well defined no matter what value is used. This gives the
2349 compiler more freedom to optimize. Here are some examples of (potentially
2350 surprising) transformations that are valid (in pseudo IR):</p>
2353 <pre class="doc_code">
2363 <p>This is safe because all of the output bits are affected by the undef bits.
2364 Any output bit can have a zero or one depending on the input bits.</p>
2366 <pre class="doc_code">
2377 <p>These logical operations have bits that are not always affected by the input.
2378 For example, if <tt>%X</tt> has a zero bit, then the output of the
2379 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2380 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2381 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2382 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2383 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2384 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2385 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2387 <pre class="doc_code">
2388 %A = select undef, %X, %Y
2389 %B = select undef, 42, %Y
2390 %C = select %X, %Y, undef
2401 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2402 branch) conditions can go <em>either way</em>, but they have to come from one
2403 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2404 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2405 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2406 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2407 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2410 <pre class="doc_code">
2411 %A = xor undef, undef
2429 <p>This example points out that two '<tt>undef</tt>' operands are not
2430 necessarily the same. This can be surprising to people (and also matches C
2431 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2432 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2433 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2434 its value over its "live range". This is true because the variable doesn't
2435 actually <em>have a live range</em>. Instead, the value is logically read
2436 from arbitrary registers that happen to be around when needed, so the value
2437 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2438 need to have the same semantics or the core LLVM "replace all uses with"
2439 concept would not hold.</p>
2441 <pre class="doc_code">
2449 <p>These examples show the crucial difference between an <em>undefined
2450 value</em> and <em>undefined behavior</em>. An undefined value (like
2451 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2452 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2453 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2454 defined on SNaN's. However, in the second example, we can make a more
2455 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2456 arbitrary value, we are allowed to assume that it could be zero. Since a
2457 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2458 the operation does not execute at all. This allows us to delete the divide and
2459 all code after it. Because the undefined operation "can't happen", the
2460 optimizer can assume that it occurs in dead code.</p>
2462 <pre class="doc_code">
2463 a: store undef -> %X
2464 b: store %X -> undef
2470 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2471 undefined value can be assumed to not have any effect; we can assume that the
2472 value is overwritten with bits that happen to match what was already there.
2473 However, a store <em>to</em> an undefined location could clobber arbitrary
2474 memory, therefore, it has undefined behavior.</p>
2478 <!-- ======================================================================= -->
2480 <a name="trapvalues">Trap Values</a>
2485 <p>Trap values are similar to <a href="#undefvalues">undef values</a>, however
2486 instead of representing an unspecified bit pattern, they represent the
2487 fact that an instruction or constant expression which cannot evoke side
2488 effects has nevertheless detected a condition which results in undefined
2491 <p>There is currently no way of representing a trap value in the IR; they
2492 only exist when produced by operations such as
2493 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2495 <p>Trap value behavior is defined in terms of value <i>dependence</i>:</p>
2498 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2499 their operands.</li>
2501 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2502 to their dynamic predecessor basic block.</li>
2504 <li>Function arguments depend on the corresponding actual argument values in
2505 the dynamic callers of their functions.</li>
2507 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2508 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2509 control back to them.</li>
2511 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2512 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_unwind"><tt>unwind</tt></a>,
2513 or exception-throwing call instructions that dynamically transfer control
2516 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2517 referenced memory addresses, following the order in the IR
2518 (including loads and stores implied by intrinsics such as
2519 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2521 <!-- TODO: In the case of multiple threads, this only applies if the store
2522 "happens-before" the load or store. -->
2524 <!-- TODO: floating-point exception state -->
2526 <li>An instruction with externally visible side effects depends on the most
2527 recent preceding instruction with externally visible side effects, following
2528 the order in the IR. (This includes
2529 <a href="#volatile">volatile operations</a>.)</li>
2531 <li>An instruction <i>control-depends</i> on a
2532 <a href="#terminators">terminator instruction</a>
2533 if the terminator instruction has multiple successors and the instruction
2534 is always executed when control transfers to one of the successors, and
2535 may not be executed when control is transferred to another.</li>
2537 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2538 instruction if the set of instructions it otherwise depends on would be
2539 different if the terminator had transferred control to a different
2542 <li>Dependence is transitive.</li>
2546 <p>Whenever a trap value is generated, all values which depend on it evaluate
2547 to trap. If they have side effects, the evoke their side effects as if each
2548 operand with a trap value were undef. If they have externally-visible side
2549 effects, the behavior is undefined.</p>
2551 <p>Here are some examples:</p>
2553 <pre class="doc_code">
2555 %trap = sub nuw i32 0, 1 ; Results in a trap value.
2556 %still_trap = and i32 %trap, 0 ; Whereas (and i32 undef, 0) would return 0.
2557 %trap_yet_again = getelementptr i32* @h, i32 %still_trap
2558 store i32 0, i32* %trap_yet_again ; undefined behavior
2560 store i32 %trap, i32* @g ; Trap value conceptually stored to memory.
2561 %trap2 = load i32* @g ; Returns a trap value, not just undef.
2563 volatile store i32 %trap, i32* @g ; External observation; undefined behavior.
2565 %narrowaddr = bitcast i32* @g to i16*
2566 %wideaddr = bitcast i32* @g to i64*
2567 %trap3 = load i16* %narrowaddr ; Returns a trap value.
2568 %trap4 = load i64* %wideaddr ; Returns a trap value.
2570 %cmp = icmp slt i32 %trap, 0 ; Returns a trap value.
2571 br i1 %cmp, label %true, label %end ; Branch to either destination.
2574 volatile store i32 0, i32* @g ; This is control-dependent on %cmp, so
2575 ; it has undefined behavior.
2579 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2580 ; Both edges into this PHI are
2581 ; control-dependent on %cmp, so this
2582 ; always results in a trap value.
2584 volatile store i32 0, i32* @g ; This would depend on the store in %true
2585 ; if %cmp is true, or the store in %entry
2586 ; otherwise, so this is undefined behavior.
2588 br i1 %cmp, label %second_true, label %second_end
2589 ; The same branch again, but this time the
2590 ; true block doesn't have side effects.
2597 volatile store i32 0, i32* @g ; This time, the instruction always depends
2598 ; on the store in %end. Also, it is
2599 ; control-equivalent to %end, so this is
2600 ; well-defined (again, ignoring earlier
2601 ; undefined behavior in this example).
2606 <!-- ======================================================================= -->
2608 <a name="blockaddress">Addresses of Basic Blocks</a>
2613 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2615 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2616 basic block in the specified function, and always has an i8* type. Taking
2617 the address of the entry block is illegal.</p>
2619 <p>This value only has defined behavior when used as an operand to the
2620 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2621 comparisons against null. Pointer equality tests between labels addresses
2622 results in undefined behavior — though, again, comparison against null
2623 is ok, and no label is equal to the null pointer. This may be passed around
2624 as an opaque pointer sized value as long as the bits are not inspected. This
2625 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2626 long as the original value is reconstituted before the <tt>indirectbr</tt>
2629 <p>Finally, some targets may provide defined semantics when using the value as
2630 the operand to an inline assembly, but that is target specific.</p>
2635 <!-- ======================================================================= -->
2637 <a name="constantexprs">Constant Expressions</a>
2642 <p>Constant expressions are used to allow expressions involving other constants
2643 to be used as constants. Constant expressions may be of
2644 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2645 operation that does not have side effects (e.g. load and call are not
2646 supported). The following is the syntax for constant expressions:</p>
2649 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2650 <dd>Truncate a constant to another type. The bit size of CST must be larger
2651 than the bit size of TYPE. Both types must be integers.</dd>
2653 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2654 <dd>Zero 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>sext (CST to TYPE)</tt></b></dt>
2658 <dd>Sign extend a constant to another type. The bit size of CST must be
2659 smaller than the bit size of TYPE. Both types must be integers.</dd>
2661 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2662 <dd>Truncate a floating point constant to another floating point type. The
2663 size of CST must be larger than the size of TYPE. Both types must be
2664 floating point.</dd>
2666 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2667 <dd>Floating point extend a constant to another type. The size of CST must be
2668 smaller or equal to the size of TYPE. Both types must be floating
2671 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2672 <dd>Convert a floating point constant to the corresponding unsigned integer
2673 constant. TYPE must be a scalar or vector integer type. CST must be of
2674 scalar or vector floating point type. Both CST and TYPE must be scalars,
2675 or vectors of the same number of elements. If the value won't fit in the
2676 integer type, the results are undefined.</dd>
2678 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2679 <dd>Convert a floating point constant to the corresponding signed integer
2680 constant. TYPE must be a scalar or vector integer type. CST must be of
2681 scalar or vector floating point type. Both CST and TYPE must be scalars,
2682 or vectors of the same number of elements. If the value won't fit in the
2683 integer type, the results are undefined.</dd>
2685 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2686 <dd>Convert an unsigned integer constant to the corresponding floating point
2687 constant. TYPE must be a scalar or vector floating point type. CST must be
2688 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2689 vectors of the same number of elements. If the value won't fit in the
2690 floating point type, the results are undefined.</dd>
2692 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2693 <dd>Convert a signed integer constant to the corresponding floating point
2694 constant. TYPE must be a scalar or vector floating point type. CST must be
2695 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2696 vectors of the same number of elements. If the value won't fit in the
2697 floating point type, the results are undefined.</dd>
2699 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2700 <dd>Convert a pointer typed constant to the corresponding integer constant
2701 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2702 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2703 make it fit in <tt>TYPE</tt>.</dd>
2705 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2706 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2707 type. CST must be of integer type. The CST value is zero extended,
2708 truncated, or unchanged to make it fit in a pointer size. This one is
2709 <i>really</i> dangerous!</dd>
2711 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2712 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2713 are the same as those for the <a href="#i_bitcast">bitcast
2714 instruction</a>.</dd>
2716 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2717 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2718 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2719 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2720 instruction, the index list may have zero or more indexes, which are
2721 required to make sense for the type of "CSTPTR".</dd>
2723 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2724 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2726 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2727 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2729 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2730 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2732 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2733 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2736 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2737 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2740 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2741 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2744 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2745 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2746 constants. The index list is interpreted in a similar manner as indices in
2747 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2748 index value must be specified.</dd>
2750 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2751 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2752 constants. The index list is interpreted in a similar manner as indices in
2753 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2754 index value must be specified.</dd>
2756 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2757 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2758 be any of the <a href="#binaryops">binary</a>
2759 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2760 on operands are the same as those for the corresponding instruction
2761 (e.g. no bitwise operations on floating point values are allowed).</dd>
2768 <!-- *********************************************************************** -->
2769 <h2><a name="othervalues">Other Values</a></h2>
2770 <!-- *********************************************************************** -->
2772 <!-- ======================================================================= -->
2774 <a name="inlineasm">Inline Assembler Expressions</a>
2779 <p>LLVM supports inline assembler expressions (as opposed
2780 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2781 a special value. This value represents the inline assembler as a string
2782 (containing the instructions to emit), a list of operand constraints (stored
2783 as a string), a flag that indicates whether or not the inline asm
2784 expression has side effects, and a flag indicating whether the function
2785 containing the asm needs to align its stack conservatively. An example
2786 inline assembler expression is:</p>
2788 <pre class="doc_code">
2789 i32 (i32) asm "bswap $0", "=r,r"
2792 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2793 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2796 <pre class="doc_code">
2797 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2800 <p>Inline asms with side effects not visible in the constraint list must be
2801 marked as having side effects. This is done through the use of the
2802 '<tt>sideeffect</tt>' keyword, like so:</p>
2804 <pre class="doc_code">
2805 call void asm sideeffect "eieio", ""()
2808 <p>In some cases inline asms will contain code that will not work unless the
2809 stack is aligned in some way, such as calls or SSE instructions on x86,
2810 yet will not contain code that does that alignment within the asm.
2811 The compiler should make conservative assumptions about what the asm might
2812 contain and should generate its usual stack alignment code in the prologue
2813 if the '<tt>alignstack</tt>' keyword is present:</p>
2815 <pre class="doc_code">
2816 call void asm alignstack "eieio", ""()
2819 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2822 <p>TODO: The format of the asm and constraints string still need to be
2823 documented here. Constraints on what can be done (e.g. duplication, moving,
2824 etc need to be documented). This is probably best done by reference to
2825 another document that covers inline asm from a holistic perspective.</p>
2828 <a name="inlineasm_md">Inline Asm Metadata</a>
2833 <p>The call instructions that wrap inline asm nodes may have a "!srcloc" MDNode
2834 attached to it that contains a list of constant integers. If present, the
2835 code generator will use the integer as the location cookie value when report
2836 errors through the LLVMContext error reporting mechanisms. This allows a
2837 front-end to correlate backend errors that occur with inline asm back to the
2838 source code that produced it. For example:</p>
2840 <pre class="doc_code">
2841 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2843 !42 = !{ i32 1234567 }
2846 <p>It is up to the front-end to make sense of the magic numbers it places in the
2847 IR. If the MDNode contains multiple constants, the code generator will use
2848 the one that corresponds to the line of the asm that the error occurs on.</p>
2854 <!-- ======================================================================= -->
2856 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2861 <p>LLVM IR allows metadata to be attached to instructions in the program that
2862 can convey extra information about the code to the optimizers and code
2863 generator. One example application of metadata is source-level debug
2864 information. There are two metadata primitives: strings and nodes. All
2865 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2866 preceding exclamation point ('<tt>!</tt>').</p>
2868 <p>A metadata string is a string surrounded by double quotes. It can contain
2869 any character by escaping non-printable characters with "\xx" where "xx" is
2870 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2872 <p>Metadata nodes are represented with notation similar to structure constants
2873 (a comma separated list of elements, surrounded by braces and preceded by an
2874 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2875 10}</tt>". Metadata nodes can have any values as their operand.</p>
2877 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2878 metadata nodes, which can be looked up in the module symbol table. For
2879 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2881 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2882 function is using two metadata arguments.</p>
2884 <div class="doc_code">
2886 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2890 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2891 attached with <tt>add</tt> instruction using <tt>!dbg</tt> identifier.</p>
2893 <div class="doc_code">
2895 %indvar.next = add i64 %indvar, 1, !dbg !21
2903 <!-- *********************************************************************** -->
2905 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2907 <!-- *********************************************************************** -->
2909 <p>LLVM has a number of "magic" global variables that contain data that affect
2910 code generation or other IR semantics. These are documented here. All globals
2911 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2912 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2915 <!-- ======================================================================= -->
2917 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2922 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2923 href="#linkage_appending">appending linkage</a>. This array contains a list of
2924 pointers to global variables and functions which may optionally have a pointer
2925 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2931 @llvm.used = appending global [2 x i8*] [
2933 i8* bitcast (i32* @Y to i8*)
2934 ], section "llvm.metadata"
2937 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2938 compiler, assembler, and linker are required to treat the symbol as if there is
2939 a reference to the global that it cannot see. For example, if a variable has
2940 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2941 list, it cannot be deleted. This is commonly used to represent references from
2942 inline asms and other things the compiler cannot "see", and corresponds to
2943 "attribute((used))" in GNU C.</p>
2945 <p>On some targets, the code generator must emit a directive to the assembler or
2946 object file to prevent the assembler and linker from molesting the symbol.</p>
2950 <!-- ======================================================================= -->
2952 <a name="intg_compiler_used">
2953 The '<tt>llvm.compiler.used</tt>' Global Variable
2959 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2960 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2961 touching the symbol. On targets that support it, this allows an intelligent
2962 linker to optimize references to the symbol without being impeded as it would be
2963 by <tt>@llvm.used</tt>.</p>
2965 <p>This is a rare construct that should only be used in rare circumstances, and
2966 should not be exposed to source languages.</p>
2970 <!-- ======================================================================= -->
2972 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2977 %0 = type { i32, void ()* }
2978 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2980 <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.
2985 <!-- ======================================================================= -->
2987 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2992 %0 = type { i32, void ()* }
2993 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2996 <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.
3003 <!-- *********************************************************************** -->
3004 <h2><a name="instref">Instruction Reference</a></h2>
3005 <!-- *********************************************************************** -->
3009 <p>The LLVM instruction set consists of several different classifications of
3010 instructions: <a href="#terminators">terminator
3011 instructions</a>, <a href="#binaryops">binary instructions</a>,
3012 <a href="#bitwiseops">bitwise binary instructions</a>,
3013 <a href="#memoryops">memory instructions</a>, and
3014 <a href="#otherops">other instructions</a>.</p>
3016 <!-- ======================================================================= -->
3018 <a name="terminators">Terminator Instructions</a>
3023 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3024 in a program ends with a "Terminator" instruction, which indicates which
3025 block should be executed after the current block is finished. These
3026 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3027 control flow, not values (the one exception being the
3028 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3030 <p>The terminator instructions are:
3031 '<a href="#i_ret"><tt>ret</tt></a>',
3032 '<a href="#i_br"><tt>br</tt></a>',
3033 '<a href="#i_switch"><tt>switch</tt></a>',
3034 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3035 '<a href="#i_invoke"><tt>invoke</tt></a>',
3036 '<a href="#i_unwind"><tt>unwind</tt></a>',
3037 '<a href="#i_resume"><tt>resume</tt></a>', and
3038 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3040 <!-- _______________________________________________________________________ -->
3042 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3049 ret <type> <value> <i>; Return a value from a non-void function</i>
3050 ret void <i>; Return from void function</i>
3054 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3055 a value) from a function back to the caller.</p>
3057 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3058 value and then causes control flow, and one that just causes control flow to
3062 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3063 return value. The type of the return value must be a
3064 '<a href="#t_firstclass">first class</a>' type.</p>
3066 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3067 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3068 value or a return value with a type that does not match its type, or if it
3069 has a void return type and contains a '<tt>ret</tt>' instruction with a
3073 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3074 the calling function's context. If the caller is a
3075 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3076 instruction after the call. If the caller was an
3077 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3078 the beginning of the "normal" destination block. If the instruction returns
3079 a value, that value shall set the call or invoke instruction's return
3084 ret i32 5 <i>; Return an integer value of 5</i>
3085 ret void <i>; Return from a void function</i>
3086 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3090 <!-- _______________________________________________________________________ -->
3092 <a name="i_br">'<tt>br</tt>' Instruction</a>
3099 br i1 <cond>, label <iftrue>, label <iffalse>
3100 br label <dest> <i>; Unconditional branch</i>
3104 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3105 different basic block in the current function. There are two forms of this
3106 instruction, corresponding to a conditional branch and an unconditional
3110 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3111 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3112 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3116 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3117 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3118 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3119 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3124 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3125 br i1 %cond, label %IfEqual, label %IfUnequal
3127 <a href="#i_ret">ret</a> i32 1
3129 <a href="#i_ret">ret</a> i32 0
3134 <!-- _______________________________________________________________________ -->
3136 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3143 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3147 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3148 several different places. It is a generalization of the '<tt>br</tt>'
3149 instruction, allowing a branch to occur to one of many possible
3153 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3154 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3155 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3156 The table is not allowed to contain duplicate constant entries.</p>
3159 <p>The <tt>switch</tt> instruction specifies a table of values and
3160 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3161 is searched for the given value. If the value is found, control flow is
3162 transferred to the corresponding destination; otherwise, control flow is
3163 transferred to the default destination.</p>
3165 <h5>Implementation:</h5>
3166 <p>Depending on properties of the target machine and the particular
3167 <tt>switch</tt> instruction, this instruction may be code generated in
3168 different ways. For example, it could be generated as a series of chained
3169 conditional branches or with a lookup table.</p>
3173 <i>; Emulate a conditional br instruction</i>
3174 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3175 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3177 <i>; Emulate an unconditional br instruction</i>
3178 switch i32 0, label %dest [ ]
3180 <i>; Implement a jump table:</i>
3181 switch i32 %val, label %otherwise [ i32 0, label %onzero
3183 i32 2, label %ontwo ]
3189 <!-- _______________________________________________________________________ -->
3191 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3198 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3203 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3204 within the current function, whose address is specified by
3205 "<tt>address</tt>". Address must be derived from a <a
3206 href="#blockaddress">blockaddress</a> constant.</p>
3210 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3211 rest of the arguments indicate the full set of possible destinations that the
3212 address may point to. Blocks are allowed to occur multiple times in the
3213 destination list, though this isn't particularly useful.</p>
3215 <p>This destination list is required so that dataflow analysis has an accurate
3216 understanding of the CFG.</p>
3220 <p>Control transfers to the block specified in the address argument. All
3221 possible destination blocks must be listed in the label list, otherwise this
3222 instruction has undefined behavior. This implies that jumps to labels
3223 defined in other functions have undefined behavior as well.</p>
3225 <h5>Implementation:</h5>
3227 <p>This is typically implemented with a jump through a register.</p>
3231 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3237 <!-- _______________________________________________________________________ -->
3239 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3246 <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>]
3247 to label <normal label> unwind label <exception label>
3251 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3252 function, with the possibility of control flow transfer to either the
3253 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3254 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3255 control flow will return to the "normal" label. If the callee (or any
3256 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
3257 instruction, control is interrupted and continued at the dynamically nearest
3258 "exception" label.</p>
3260 <p>The '<tt>exception</tt>' label is a
3261 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3262 exception. As such, '<tt>exception</tt>' label is required to have the
3263 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3264 the information about about the behavior of the program after unwinding
3265 happens, as its first non-PHI instruction. The restrictions on the
3266 "<tt>landingpad</tt>" instruction's tightly couples it to the
3267 "<tt>invoke</tt>" instruction, so that the important information contained
3268 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3272 <p>This instruction requires several arguments:</p>
3275 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3276 convention</a> the call should use. If none is specified, the call
3277 defaults to using C calling conventions.</li>
3279 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3280 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3281 '<tt>inreg</tt>' attributes are valid here.</li>
3283 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3284 function value being invoked. In most cases, this is a direct function
3285 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3286 off an arbitrary pointer to function value.</li>
3288 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3289 function to be invoked. </li>
3291 <li>'<tt>function args</tt>': argument list whose types match the function
3292 signature argument types and parameter attributes. All arguments must be
3293 of <a href="#t_firstclass">first class</a> type. If the function
3294 signature indicates the function accepts a variable number of arguments,
3295 the extra arguments can be specified.</li>
3297 <li>'<tt>normal label</tt>': the label reached when the called function
3298 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3300 <li>'<tt>exception label</tt>': the label reached when a callee returns with
3301 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
3303 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3304 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3305 '<tt>readnone</tt>' attributes are valid here.</li>
3309 <p>This instruction is designed to operate as a standard
3310 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3311 primary difference is that it establishes an association with a label, which
3312 is used by the runtime library to unwind the stack.</p>
3314 <p>This instruction is used in languages with destructors to ensure that proper
3315 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3316 exception. Additionally, this is important for implementation of
3317 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3319 <p>For the purposes of the SSA form, the definition of the value returned by the
3320 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3321 block to the "normal" label. If the callee unwinds then no return value is
3324 <p>Note that the code generator does not yet completely support unwind, and
3325 that the invoke/unwind semantics are likely to change in future versions.</p>
3329 %retval = invoke i32 @Test(i32 15) to label %Continue
3330 unwind label %TestCleanup <i>; {i32}:retval set</i>
3331 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3332 unwind label %TestCleanup <i>; {i32}:retval set</i>
3337 <!-- _______________________________________________________________________ -->
3340 <a name="i_unwind">'<tt>unwind</tt>' Instruction</a>
3351 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
3352 at the first callee in the dynamic call stack which used
3353 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
3354 This is primarily used to implement exception handling.</p>
3357 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
3358 immediately halt. The dynamic call stack is then searched for the
3359 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
3360 Once found, execution continues at the "exceptional" destination block
3361 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
3362 instruction in the dynamic call chain, undefined behavior results.</p>
3364 <p>Note that the code generator does not yet completely support unwind, and
3365 that the invoke/unwind semantics are likely to change in future versions.</p>
3369 <!-- _______________________________________________________________________ -->
3372 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3379 resume <type> <value>
3383 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3387 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3388 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3392 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3393 (in-flight) exception whose unwinding was interrupted with
3394 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3398 resume { i8*, i32 } %exn
3403 <!-- _______________________________________________________________________ -->
3406 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3417 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3418 instruction is used to inform the optimizer that a particular portion of the
3419 code is not reachable. This can be used to indicate that the code after a
3420 no-return function cannot be reached, and other facts.</p>
3423 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3429 <!-- ======================================================================= -->
3431 <a name="binaryops">Binary Operations</a>
3436 <p>Binary operators are used to do most of the computation in a program. They
3437 require two operands of the same type, execute an operation on them, and
3438 produce a single value. The operands might represent multiple data, as is
3439 the case with the <a href="#t_vector">vector</a> data type. The result value
3440 has the same type as its operands.</p>
3442 <p>There are several different binary operators:</p>
3444 <!-- _______________________________________________________________________ -->
3446 <a name="i_add">'<tt>add</tt>' Instruction</a>
3453 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3454 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3455 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3456 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3460 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3463 <p>The two arguments to the '<tt>add</tt>' instruction must
3464 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3465 integer values. Both arguments must have identical types.</p>
3468 <p>The value produced is the integer sum of the two operands.</p>
3470 <p>If the sum has unsigned overflow, the result returned is the mathematical
3471 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3473 <p>Because LLVM integers use a two's complement representation, this instruction
3474 is appropriate for both signed and unsigned integers.</p>
3476 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3477 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3478 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3479 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3480 respectively, occurs.</p>
3484 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3489 <!-- _______________________________________________________________________ -->
3491 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3498 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3502 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3505 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3506 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3507 floating point values. Both arguments must have identical types.</p>
3510 <p>The value produced is the floating point sum of the two operands.</p>
3514 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3519 <!-- _______________________________________________________________________ -->
3521 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3528 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3529 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3530 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3531 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3535 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3538 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3539 '<tt>neg</tt>' instruction present in most other intermediate
3540 representations.</p>
3543 <p>The two arguments to the '<tt>sub</tt>' instruction must
3544 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3545 integer values. Both arguments must have identical types.</p>
3548 <p>The value produced is the integer difference of the two operands.</p>
3550 <p>If the difference has unsigned overflow, the result returned is the
3551 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3554 <p>Because LLVM integers use a two's complement representation, this instruction
3555 is appropriate for both signed and unsigned integers.</p>
3557 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3558 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3559 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3560 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3561 respectively, occurs.</p>
3565 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3566 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3571 <!-- _______________________________________________________________________ -->
3573 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3580 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3584 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3587 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3588 '<tt>fneg</tt>' instruction present in most other intermediate
3589 representations.</p>
3592 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3593 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3594 floating point values. Both arguments must have identical types.</p>
3597 <p>The value produced is the floating point difference of the two operands.</p>
3601 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3602 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3607 <!-- _______________________________________________________________________ -->
3609 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3616 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3617 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3618 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3619 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3623 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3626 <p>The two arguments to the '<tt>mul</tt>' instruction must
3627 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3628 integer values. Both arguments must have identical types.</p>
3631 <p>The value produced is the integer product of the two operands.</p>
3633 <p>If the result of the multiplication has unsigned overflow, the result
3634 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3635 width of the result.</p>
3637 <p>Because LLVM integers use a two's complement representation, and the result
3638 is the same width as the operands, this instruction returns the correct
3639 result for both signed and unsigned integers. If a full product
3640 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3641 be sign-extended or zero-extended as appropriate to the width of the full
3644 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3645 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3646 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3647 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3648 respectively, occurs.</p>
3652 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3657 <!-- _______________________________________________________________________ -->
3659 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3666 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3670 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3673 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3674 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3675 floating point values. Both arguments must have identical types.</p>
3678 <p>The value produced is the floating point product of the two operands.</p>
3682 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3687 <!-- _______________________________________________________________________ -->
3689 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
3696 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3697 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3701 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3704 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3705 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3706 values. Both arguments must have identical types.</p>
3709 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3711 <p>Note that unsigned integer division and signed integer division are distinct
3712 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3714 <p>Division by zero leads to undefined behavior.</p>
3716 <p>If the <tt>exact</tt> keyword is present, the result value of the
3717 <tt>udiv</tt> is a <a href="#trapvalues">trap value</a> if %op1 is not a
3718 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
3723 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3728 <!-- _______________________________________________________________________ -->
3730 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
3737 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3738 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3742 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3745 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3746 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3747 values. Both arguments must have identical types.</p>
3750 <p>The value produced is the signed integer quotient of the two operands rounded
3753 <p>Note that signed integer division and unsigned integer division are distinct
3754 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3756 <p>Division by zero leads to undefined behavior. Overflow also leads to
3757 undefined behavior; this is a rare case, but can occur, for example, by doing
3758 a 32-bit division of -2147483648 by -1.</p>
3760 <p>If the <tt>exact</tt> keyword is present, the result value of the
3761 <tt>sdiv</tt> is a <a href="#trapvalues">trap value</a> if the result would
3766 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3771 <!-- _______________________________________________________________________ -->
3773 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
3780 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3784 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3787 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3788 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3789 floating point values. Both arguments must have identical types.</p>
3792 <p>The value produced is the floating point quotient of the two operands.</p>
3796 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3801 <!-- _______________________________________________________________________ -->
3803 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3810 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3814 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3815 division of its two arguments.</p>
3818 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3819 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3820 values. Both arguments must have identical types.</p>
3823 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3824 This instruction always performs an unsigned division to get the
3827 <p>Note that unsigned integer remainder and signed integer remainder are
3828 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3830 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3834 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3839 <!-- _______________________________________________________________________ -->
3841 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3848 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3852 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3853 division of its two operands. This instruction can also take
3854 <a href="#t_vector">vector</a> versions of the values in which case the
3855 elements must be integers.</p>
3858 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3859 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3860 values. Both arguments must have identical types.</p>
3863 <p>This instruction returns the <i>remainder</i> of a division (where the result
3864 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
3865 <i>modulo</i> operator (where the result is either zero or has the same sign
3866 as the divisor, <tt>op2</tt>) of a value.
3867 For more information about the difference,
3868 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3869 Math Forum</a>. For a table of how this is implemented in various languages,
3870 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3871 Wikipedia: modulo operation</a>.</p>
3873 <p>Note that signed integer remainder and unsigned integer remainder are
3874 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3876 <p>Taking the remainder of a division by zero leads to undefined behavior.
3877 Overflow also leads to undefined behavior; this is a rare case, but can
3878 occur, for example, by taking the remainder of a 32-bit division of
3879 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3880 lets srem be implemented using instructions that return both the result of
3881 the division and the remainder.)</p>
3885 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3890 <!-- _______________________________________________________________________ -->
3892 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
3899 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3903 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3904 its two operands.</p>
3907 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3908 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3909 floating point values. Both arguments must have identical types.</p>
3912 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3913 has the same sign as the dividend.</p>
3917 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3924 <!-- ======================================================================= -->
3926 <a name="bitwiseops">Bitwise Binary Operations</a>
3931 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3932 program. They are generally very efficient instructions and can commonly be
3933 strength reduced from other instructions. They require two operands of the
3934 same type, execute an operation on them, and produce a single value. The
3935 resulting value is the same type as its operands.</p>
3937 <!-- _______________________________________________________________________ -->
3939 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
3946 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3947 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3948 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3949 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3953 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3954 a specified number of bits.</p>
3957 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3958 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3959 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3962 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3963 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3964 is (statically or dynamically) negative or equal to or larger than the number
3965 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3966 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3967 shift amount in <tt>op2</tt>.</p>
3969 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
3970 <a href="#trapvalues">trap value</a> if it shifts out any non-zero bits. If
3971 the <tt>nsw</tt> keyword is present, then the shift produces a
3972 <a href="#trapvalues">trap value</a> if it shifts out any bits that disagree
3973 with the resultant sign bit. As such, NUW/NSW have the same semantics as
3974 they would if the shift were expressed as a mul instruction with the same
3975 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
3979 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3980 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3981 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3982 <result> = shl i32 1, 32 <i>; undefined</i>
3983 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3988 <!-- _______________________________________________________________________ -->
3990 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
3997 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3998 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4002 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4003 operand shifted to the right a specified number of bits with zero fill.</p>
4006 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4007 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4008 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4011 <p>This instruction always performs a logical shift right operation. The most
4012 significant bits of the result will be filled with zero bits after the shift.
4013 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4014 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4015 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4016 shift amount in <tt>op2</tt>.</p>
4018 <p>If the <tt>exact</tt> keyword is present, the result value of the
4019 <tt>lshr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
4020 shifted out are non-zero.</p>
4025 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4026 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4027 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4028 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4029 <result> = lshr i32 1, 32 <i>; undefined</i>
4030 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4035 <!-- _______________________________________________________________________ -->
4037 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4044 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4045 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4049 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4050 operand shifted to the right a specified number of bits with sign
4054 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4055 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4056 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4059 <p>This instruction always performs an arithmetic shift right operation, The
4060 most significant bits of the result will be filled with the sign bit
4061 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4062 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4063 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4064 the corresponding shift amount in <tt>op2</tt>.</p>
4066 <p>If the <tt>exact</tt> keyword is present, the result value of the
4067 <tt>ashr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
4068 shifted out are non-zero.</p>
4072 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4073 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4074 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4075 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4076 <result> = ashr i32 1, 32 <i>; undefined</i>
4077 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4082 <!-- _______________________________________________________________________ -->
4084 <a name="i_and">'<tt>and</tt>' Instruction</a>
4091 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4095 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4099 <p>The two arguments to the '<tt>and</tt>' instruction must be
4100 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4101 values. Both arguments must have identical types.</p>
4104 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4106 <table border="1" cellspacing="0" cellpadding="4">
4138 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4139 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4140 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4143 <!-- _______________________________________________________________________ -->
4145 <a name="i_or">'<tt>or</tt>' Instruction</a>
4152 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4156 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4160 <p>The two arguments to the '<tt>or</tt>' instruction must be
4161 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4162 values. Both arguments must have identical types.</p>
4165 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4167 <table border="1" cellspacing="0" cellpadding="4">
4199 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4200 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4201 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4206 <!-- _______________________________________________________________________ -->
4208 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4215 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4219 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4220 its two operands. The <tt>xor</tt> is used to implement the "one's
4221 complement" operation, which is the "~" operator in C.</p>
4224 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4225 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4226 values. Both arguments must have identical types.</p>
4229 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4231 <table border="1" cellspacing="0" cellpadding="4">
4263 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4264 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4265 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4266 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4273 <!-- ======================================================================= -->
4275 <a name="vectorops">Vector Operations</a>
4280 <p>LLVM supports several instructions to represent vector operations in a
4281 target-independent manner. These instructions cover the element-access and
4282 vector-specific operations needed to process vectors effectively. While LLVM
4283 does directly support these vector operations, many sophisticated algorithms
4284 will want to use target-specific intrinsics to take full advantage of a
4285 specific target.</p>
4287 <!-- _______________________________________________________________________ -->
4289 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4296 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4300 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4301 from a vector at a specified index.</p>
4305 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4306 of <a href="#t_vector">vector</a> type. The second operand is an index
4307 indicating the position from which to extract the element. The index may be
4311 <p>The result is a scalar of the same type as the element type of
4312 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4313 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4314 results are undefined.</p>
4318 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4323 <!-- _______________________________________________________________________ -->
4325 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4332 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4336 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4337 vector at a specified index.</p>
4340 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4341 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4342 whose type must equal the element type of the first operand. The third
4343 operand is an index indicating the position at which to insert the value.
4344 The index may be a variable.</p>
4347 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4348 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4349 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4350 results are undefined.</p>
4354 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4359 <!-- _______________________________________________________________________ -->
4361 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4368 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4372 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4373 from two input vectors, returning a vector with the same element type as the
4374 input and length that is the same as the shuffle mask.</p>
4377 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4378 with types that match each other. The third argument is a shuffle mask whose
4379 element type is always 'i32'. The result of the instruction is a vector
4380 whose length is the same as the shuffle mask and whose element type is the
4381 same as the element type of the first two operands.</p>
4383 <p>The shuffle mask operand is required to be a constant vector with either
4384 constant integer or undef values.</p>
4387 <p>The elements of the two input vectors are numbered from left to right across
4388 both of the vectors. The shuffle mask operand specifies, for each element of
4389 the result vector, which element of the two input vectors the result element
4390 gets. The element selector may be undef (meaning "don't care") and the
4391 second operand may be undef if performing a shuffle from only one vector.</p>
4395 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4396 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4397 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4398 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4399 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4400 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4401 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4402 <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>
4409 <!-- ======================================================================= -->
4411 <a name="aggregateops">Aggregate Operations</a>
4416 <p>LLVM supports several instructions for working with
4417 <a href="#t_aggregate">aggregate</a> values.</p>
4419 <!-- _______________________________________________________________________ -->
4421 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4428 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4432 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4433 from an <a href="#t_aggregate">aggregate</a> value.</p>
4436 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4437 of <a href="#t_struct">struct</a> or
4438 <a href="#t_array">array</a> type. The operands are constant indices to
4439 specify which value to extract in a similar manner as indices in a
4440 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4441 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4443 <li>Since the value being indexed is not a pointer, the first index is
4444 omitted and assumed to be zero.</li>
4445 <li>At least one index must be specified.</li>
4446 <li>Not only struct indices but also array indices must be in
4451 <p>The result is the value at the position in the aggregate specified by the
4456 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4461 <!-- _______________________________________________________________________ -->
4463 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4470 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4474 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4475 in an <a href="#t_aggregate">aggregate</a> value.</p>
4478 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4479 of <a href="#t_struct">struct</a> or
4480 <a href="#t_array">array</a> type. The second operand is a first-class
4481 value to insert. The following operands are constant indices indicating
4482 the position at which to insert the value in a similar manner as indices in a
4483 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4484 value to insert must have the same type as the value identified by the
4488 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4489 that of <tt>val</tt> except that the value at the position specified by the
4490 indices is that of <tt>elt</tt>.</p>
4494 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4495 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4496 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4503 <!-- ======================================================================= -->
4505 <a name="memoryops">Memory Access and Addressing Operations</a>
4510 <p>A key design point of an SSA-based representation is how it represents
4511 memory. In LLVM, no memory locations are in SSA form, which makes things
4512 very simple. This section describes how to read, write, and allocate
4515 <!-- _______________________________________________________________________ -->
4517 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4524 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4528 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4529 currently executing function, to be automatically released when this function
4530 returns to its caller. The object is always allocated in the generic address
4531 space (address space zero).</p>
4534 <p>The '<tt>alloca</tt>' instruction
4535 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4536 runtime stack, returning a pointer of the appropriate type to the program.
4537 If "NumElements" is specified, it is the number of elements allocated,
4538 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4539 specified, the value result of the allocation is guaranteed to be aligned to
4540 at least that boundary. If not specified, or if zero, the target can choose
4541 to align the allocation on any convenient boundary compatible with the
4544 <p>'<tt>type</tt>' may be any sized type.</p>
4547 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4548 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4549 memory is automatically released when the function returns. The
4550 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4551 variables that must have an address available. When the function returns
4552 (either with the <tt><a href="#i_ret">ret</a></tt>
4553 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
4554 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4558 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4559 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4560 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4561 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4566 <!-- _______________________________________________________________________ -->
4568 <a name="i_load">'<tt>load</tt>' Instruction</a>
4575 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4576 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4577 !<index> = !{ i32 1 }
4581 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4584 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4585 from which to load. The pointer must point to
4586 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4587 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4588 number or order of execution of this <tt>load</tt> with other <a
4589 href="#volatile">volatile operations</a>.</p>
4591 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
4592 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4593 argument. The <code>release</code> and <code>acq_rel</code> orderings are
4594 not valid on <code>load</code> instructions. Atomic loads produce <a
4595 href="#memorymodel">defined</a> results when they may see multiple atomic
4596 stores. The type of the pointee must be an integer type whose bit width
4597 is a power of two greater than or equal to eight and less than or equal
4598 to a target-specific size limit. <code>align</code> must be explicitly
4599 specified on atomic loads, and the load has undefined behavior if the
4600 alignment is not set to a value which is at least the size in bytes of
4601 the pointee. <code>!nontemporal</code> does not have any defined semantics
4602 for atomic loads.</p>
4604 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4605 operation (that is, the alignment of the memory address). A value of 0 or an
4606 omitted <tt>align</tt> argument means that the operation has the preferential
4607 alignment for the target. It is the responsibility of the code emitter to
4608 ensure that the alignment information is correct. Overestimating the
4609 alignment results in undefined behavior. Underestimating the alignment may
4610 produce less efficient code. An alignment of 1 is always safe.</p>
4612 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4613 metatadata name <index> corresponding to a metadata node with
4614 one <tt>i32</tt> entry of value 1. The existence of
4615 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4616 and code generator that this load is not expected to be reused in the cache.
4617 The code generator may select special instructions to save cache bandwidth,
4618 such as the <tt>MOVNT</tt> instruction on x86.</p>
4621 <p>The location of memory pointed to is loaded. If the value being loaded is of
4622 scalar type then the number of bytes read does not exceed the minimum number
4623 of bytes needed to hold all bits of the type. For example, loading an
4624 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4625 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4626 is undefined if the value was not originally written using a store of the
4631 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4632 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4633 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4638 <!-- _______________________________________________________________________ -->
4640 <a name="i_store">'<tt>store</tt>' Instruction</a>
4647 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4648 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
4652 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4655 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4656 and an address at which to store it. The type of the
4657 '<tt><pointer></tt>' operand must be a pointer to
4658 the <a href="#t_firstclass">first class</a> type of the
4659 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
4660 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
4661 order of execution of this <tt>store</tt> with other <a
4662 href="#volatile">volatile operations</a>.</p>
4664 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
4665 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4666 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
4667 valid on <code>store</code> instructions. Atomic loads produce <a
4668 href="#memorymodel">defined</a> results when they may see multiple atomic
4669 stores. The type of the pointee must be an integer type whose bit width
4670 is a power of two greater than or equal to eight and less than or equal
4671 to a target-specific size limit. <code>align</code> must be explicitly
4672 specified on atomic stores, and the store has undefined behavior if the
4673 alignment is not set to a value which is at least the size in bytes of
4674 the pointee. <code>!nontemporal</code> does not have any defined semantics
4675 for atomic stores.</p>
4677 <p>The optional constant "align" argument specifies the alignment of the
4678 operation (that is, the alignment of the memory address). A value of 0 or an
4679 omitted "align" argument means that the operation has the preferential
4680 alignment for the target. It is the responsibility of the code emitter to
4681 ensure that the alignment information is correct. Overestimating the
4682 alignment results in an undefined behavior. Underestimating the alignment may
4683 produce less efficient code. An alignment of 1 is always safe.</p>
4685 <p>The optional !nontemporal metadata must reference a single metatadata
4686 name <index> corresponding to a metadata node with one i32 entry of
4687 value 1. The existence of the !nontemporal metatadata on the
4688 instruction tells the optimizer and code generator that this load is
4689 not expected to be reused in the cache. The code generator may
4690 select special instructions to save cache bandwidth, such as the
4691 MOVNT instruction on x86.</p>
4695 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4696 location specified by the '<tt><pointer></tt>' operand. If
4697 '<tt><value></tt>' is of scalar type then the number of bytes written
4698 does not exceed the minimum number of bytes needed to hold all bits of the
4699 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4700 writing a value of a type like <tt>i20</tt> with a size that is not an
4701 integral number of bytes, it is unspecified what happens to the extra bits
4702 that do not belong to the type, but they will typically be overwritten.</p>
4706 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4707 store i32 3, i32* %ptr <i>; yields {void}</i>
4708 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4713 <!-- _______________________________________________________________________ -->
4715 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
4722 fence [singlethread] <ordering> <i>; yields {void}</i>
4726 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
4727 between operations.</p>
4729 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
4730 href="#ordering">ordering</a> argument which defines what
4731 <i>synchronizes-with</i> edges they add. They can only be given
4732 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
4733 <code>seq_cst</code> orderings.</p>
4736 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
4737 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
4738 <code>acquire</code> ordering semantics if and only if there exist atomic
4739 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
4740 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
4741 <var>X</var> modifies <var>M</var> (either directly or through some side effect
4742 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
4743 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
4744 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
4745 than an explicit <code>fence</code>, one (but not both) of the atomic operations
4746 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
4747 <code>acquire</code> (resp.) ordering constraint and still
4748 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
4749 <i>happens-before</i> edge.</p>
4751 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
4752 having both <code>acquire</code> and <code>release</code> semantics specified
4753 above, participates in the global program order of other <code>seq_cst</code>
4754 operations and/or fences.</p>
4756 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
4757 specifies that the fence only synchronizes with other fences in the same
4758 thread. (This is useful for interacting with signal handlers.)</p>
4762 fence acquire <i>; yields {void}</i>
4763 fence singlethread seq_cst <i>; yields {void}</i>
4768 <!-- _______________________________________________________________________ -->
4770 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
4777 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
4781 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
4782 It loads a value in memory and compares it to a given value. If they are
4783 equal, it stores a new value into the memory.</p>
4786 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
4787 address to operate on, a value to compare to the value currently be at that
4788 address, and a new value to place at that address if the compared values are
4789 equal. The type of '<var><cmp></var>' must be an integer type whose
4790 bit width is a power of two greater than or equal to eight and less than
4791 or equal to a target-specific size limit. '<var><cmp></var>' and
4792 '<var><new></var>' must have the same type, and the type of
4793 '<var><pointer></var>' must be a pointer to that type. If the
4794 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
4795 optimizer is not allowed to modify the number or order of execution
4796 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
4799 <!-- FIXME: Extend allowed types. -->
4801 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
4802 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
4804 <p>The optional "<code>singlethread</code>" argument declares that the
4805 <code>cmpxchg</code> is only atomic with respect to code (usually signal
4806 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
4807 cmpxchg is atomic with respect to all other code in the system.</p>
4809 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
4810 the size in memory of the operand.
4813 <p>The contents of memory at the location specified by the
4814 '<tt><pointer></tt>' operand is read and compared to
4815 '<tt><cmp></tt>'; if the read value is the equal,
4816 '<tt><new></tt>' is written. The original value at the location
4819 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
4820 purpose of identifying <a href="#release_sequence">release sequences</a>. A
4821 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
4822 parameter determined by dropping any <code>release</code> part of the
4823 <code>cmpxchg</code>'s ordering.</p>
4826 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
4827 optimization work on ARM.)
4829 FIXME: Is a weaker ordering constraint on failure helpful in practice?
4835 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
4836 <a href="#i_br">br</a> label %loop
4839 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
4840 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
4841 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
4842 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
4843 <a href="#i_br">br</a> i1 %success, label %done, label %loop
4851 <!-- _______________________________________________________________________ -->
4853 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
4860 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
4864 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
4867 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
4868 operation to apply, an address whose value to modify, an argument to the
4869 operation. The operation must be one of the following keywords:</p>
4884 <p>The type of '<var><value></var>' must be an integer type whose
4885 bit width is a power of two greater than or equal to eight and less than
4886 or equal to a target-specific size limit. The type of the
4887 '<code><pointer></code>' operand must be a pointer to that type.
4888 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
4889 optimizer is not allowed to modify the number or order of execution of this
4890 <code>atomicrmw</code> with other <a href="#volatile">volatile
4893 <!-- FIXME: Extend allowed types. -->
4896 <p>The contents of memory at the location specified by the
4897 '<tt><pointer></tt>' operand are atomically read, modified, and written
4898 back. The original value at the location is returned. The modification is
4899 specified by the <var>operation</var> argument:</p>
4902 <li>xchg: <code>*ptr = val</code></li>
4903 <li>add: <code>*ptr = *ptr + val</code></li>
4904 <li>sub: <code>*ptr = *ptr - val</code></li>
4905 <li>and: <code>*ptr = *ptr & val</code></li>
4906 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
4907 <li>or: <code>*ptr = *ptr | val</code></li>
4908 <li>xor: <code>*ptr = *ptr ^ val</code></li>
4909 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
4910 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
4911 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
4912 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
4917 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
4922 <!-- _______________________________________________________________________ -->
4924 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
4931 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4932 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4936 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4937 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
4938 It performs address calculation only and does not access memory.</p>
4941 <p>The first argument is always a pointer, and forms the basis of the
4942 calculation. The remaining arguments are indices that indicate which of the
4943 elements of the aggregate object are indexed. The interpretation of each
4944 index is dependent on the type being indexed into. The first index always
4945 indexes the pointer value given as the first argument, the second index
4946 indexes a value of the type pointed to (not necessarily the value directly
4947 pointed to, since the first index can be non-zero), etc. The first type
4948 indexed into must be a pointer value, subsequent types can be arrays,
4949 vectors, and structs. Note that subsequent types being indexed into
4950 can never be pointers, since that would require loading the pointer before
4951 continuing calculation.</p>
4953 <p>The type of each index argument depends on the type it is indexing into.
4954 When indexing into a (optionally packed) structure, only <tt>i32</tt>
4955 integer <b>constants</b> are allowed. When indexing into an array, pointer
4956 or vector, integers of any width are allowed, and they are not required to be
4957 constant. These integers are treated as signed values where relevant.</p>
4959 <p>For example, let's consider a C code fragment and how it gets compiled to
4962 <pre class="doc_code">
4974 int *foo(struct ST *s) {
4975 return &s[1].Z.B[5][13];
4979 <p>The LLVM code generated by the GCC frontend is:</p>
4981 <pre class="doc_code">
4982 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4983 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
4985 define i32* @foo(%ST* %s) {
4987 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
4993 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
4994 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
4995 }</tt>' type, a structure. The second index indexes into the third element
4996 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
4997 i8 }</tt>' type, another structure. The third index indexes into the second
4998 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
4999 array. The two dimensions of the array are subscripted into, yielding an
5000 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
5001 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
5003 <p>Note that it is perfectly legal to index partially through a structure,
5004 returning a pointer to an inner element. Because of this, the LLVM code for
5005 the given testcase is equivalent to:</p>
5008 define i32* @foo(%ST* %s) {
5009 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
5010 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
5011 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5012 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5013 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5018 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5019 <tt>getelementptr</tt> is a <a href="#trapvalues">trap value</a> if the
5020 base pointer is not an <i>in bounds</i> address of an allocated object,
5021 or if any of the addresses that would be formed by successive addition of
5022 the offsets implied by the indices to the base address with infinitely
5023 precise signed arithmetic are not an <i>in bounds</i> address of that
5024 allocated object. The <i>in bounds</i> addresses for an allocated object
5025 are all the addresses that point into the object, plus the address one
5026 byte past the end.</p>
5028 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5029 the base address with silently-wrapping two's complement arithmetic. If the
5030 offsets have a different width from the pointer, they are sign-extended or
5031 truncated to the width of the pointer. The result value of the
5032 <tt>getelementptr</tt> may be outside the object pointed to by the base
5033 pointer. The result value may not necessarily be used to access memory
5034 though, even if it happens to point into allocated storage. See the
5035 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5038 <p>The getelementptr instruction is often confusing. For some more insight into
5039 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5043 <i>; yields [12 x i8]*:aptr</i>
5044 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5045 <i>; yields i8*:vptr</i>
5046 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5047 <i>; yields i8*:eptr</i>
5048 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5049 <i>; yields i32*:iptr</i>
5050 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5057 <!-- ======================================================================= -->
5059 <a name="convertops">Conversion Operations</a>
5064 <p>The instructions in this category are the conversion instructions (casting)
5065 which all take a single operand and a type. They perform various bit
5066 conversions on the operand.</p>
5068 <!-- _______________________________________________________________________ -->
5070 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5077 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5081 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5082 type <tt>ty2</tt>.</p>
5085 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5086 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5087 of the same number of integers.
5088 The bit size of the <tt>value</tt> must be larger than
5089 the bit size of the destination type, <tt>ty2</tt>.
5090 Equal sized types are not allowed.</p>
5093 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5094 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5095 source size must be larger than the destination size, <tt>trunc</tt> cannot
5096 be a <i>no-op cast</i>. It will always truncate bits.</p>
5100 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5101 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5102 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5103 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5108 <!-- _______________________________________________________________________ -->
5110 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5117 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5121 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5126 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5127 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5128 of the same number of integers.
5129 The bit size of the <tt>value</tt> must be smaller than
5130 the bit size of the destination type,
5134 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5135 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5137 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5141 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5142 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5143 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5148 <!-- _______________________________________________________________________ -->
5150 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5157 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5161 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5164 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5165 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5166 of the same number of integers.
5167 The bit size of the <tt>value</tt> must be smaller than
5168 the bit size of the destination type,
5172 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5173 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5174 of the type <tt>ty2</tt>.</p>
5176 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5180 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5181 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5182 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5187 <!-- _______________________________________________________________________ -->
5189 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5196 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5200 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5204 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5205 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5206 to cast it to. The size of <tt>value</tt> must be larger than the size of
5207 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5208 <i>no-op cast</i>.</p>
5211 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5212 <a href="#t_floating">floating point</a> type to a smaller
5213 <a href="#t_floating">floating point</a> type. If the value cannot fit
5214 within the destination type, <tt>ty2</tt>, then the results are
5219 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5220 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5225 <!-- _______________________________________________________________________ -->
5227 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5234 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5238 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5239 floating point value.</p>
5242 <p>The '<tt>fpext</tt>' instruction takes a
5243 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5244 a <a href="#t_floating">floating point</a> type to cast it to. The source
5245 type must be smaller than the destination type.</p>
5248 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5249 <a href="#t_floating">floating point</a> type to a larger
5250 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5251 used to make a <i>no-op cast</i> because it always changes bits. Use
5252 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5256 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5257 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5262 <!-- _______________________________________________________________________ -->
5264 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5271 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5275 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5276 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5279 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5280 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5281 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5282 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5283 vector integer type with the same number of elements as <tt>ty</tt></p>
5286 <p>The '<tt>fptoui</tt>' instruction converts its
5287 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5288 towards zero) unsigned integer value. If the value cannot fit
5289 in <tt>ty2</tt>, the results are undefined.</p>
5293 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5294 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5295 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5300 <!-- _______________________________________________________________________ -->
5302 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5309 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5313 <p>The '<tt>fptosi</tt>' instruction converts
5314 <a href="#t_floating">floating point</a> <tt>value</tt> to
5315 type <tt>ty2</tt>.</p>
5318 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5319 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5320 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5321 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5322 vector integer type with the same number of elements as <tt>ty</tt></p>
5325 <p>The '<tt>fptosi</tt>' instruction converts its
5326 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5327 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5328 the results are undefined.</p>
5332 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5333 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5334 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5339 <!-- _______________________________________________________________________ -->
5341 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5348 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5352 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5353 integer and converts that value to the <tt>ty2</tt> type.</p>
5356 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5357 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5358 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5359 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5360 floating point type with the same number of elements as <tt>ty</tt></p>
5363 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5364 integer quantity and converts it to the corresponding floating point
5365 value. If the value cannot fit in the floating point value, the results are
5370 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5371 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5376 <!-- _______________________________________________________________________ -->
5378 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5385 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5389 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5390 and converts that value to the <tt>ty2</tt> type.</p>
5393 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5394 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5395 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5396 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5397 floating point type with the same number of elements as <tt>ty</tt></p>
5400 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5401 quantity and converts it to the corresponding floating point value. If the
5402 value cannot fit in the floating point value, the results are undefined.</p>
5406 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5407 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5412 <!-- _______________________________________________________________________ -->
5414 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5421 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5425 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
5426 the integer type <tt>ty2</tt>.</p>
5429 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5430 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
5431 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
5434 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5435 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5436 truncating or zero extending that value to the size of the integer type. If
5437 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5438 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5439 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5444 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
5445 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
5450 <!-- _______________________________________________________________________ -->
5452 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5459 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5463 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5464 pointer type, <tt>ty2</tt>.</p>
5467 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5468 value to cast, and a type to cast it to, which must be a
5469 <a href="#t_pointer">pointer</a> type.</p>
5472 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5473 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5474 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5475 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5476 than the size of a pointer then a zero extension is done. If they are the
5477 same size, nothing is done (<i>no-op cast</i>).</p>
5481 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5482 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5483 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5488 <!-- _______________________________________________________________________ -->
5490 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5497 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5501 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5502 <tt>ty2</tt> without changing any bits.</p>
5505 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5506 non-aggregate first class value, and a type to cast it to, which must also be
5507 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5508 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5509 identical. If the source type is a pointer, the destination type must also be
5510 a pointer. This instruction supports bitwise conversion of vectors to
5511 integers and to vectors of other types (as long as they have the same
5515 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5516 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5517 this conversion. The conversion is done as if the <tt>value</tt> had been
5518 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
5519 be converted to other pointer types with this instruction. To convert
5520 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5521 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5525 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5526 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5527 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5534 <!-- ======================================================================= -->
5536 <a name="otherops">Other Operations</a>
5541 <p>The instructions in this category are the "miscellaneous" instructions, which
5542 defy better classification.</p>
5544 <!-- _______________________________________________________________________ -->
5546 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5553 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5557 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5558 boolean values based on comparison of its two integer, integer vector, or
5559 pointer operands.</p>
5562 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5563 the condition code indicating the kind of comparison to perform. It is not a
5564 value, just a keyword. The possible condition code are:</p>
5567 <li><tt>eq</tt>: equal</li>
5568 <li><tt>ne</tt>: not equal </li>
5569 <li><tt>ugt</tt>: unsigned greater than</li>
5570 <li><tt>uge</tt>: unsigned greater or equal</li>
5571 <li><tt>ult</tt>: unsigned less than</li>
5572 <li><tt>ule</tt>: unsigned less or equal</li>
5573 <li><tt>sgt</tt>: signed greater than</li>
5574 <li><tt>sge</tt>: signed greater or equal</li>
5575 <li><tt>slt</tt>: signed less than</li>
5576 <li><tt>sle</tt>: signed less or equal</li>
5579 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
5580 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5581 typed. They must also be identical types.</p>
5584 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5585 condition code given as <tt>cond</tt>. The comparison performed always yields
5586 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
5587 result, as follows:</p>
5590 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
5591 <tt>false</tt> otherwise. No sign interpretation is necessary or
5594 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
5595 <tt>false</tt> otherwise. No sign interpretation is necessary or
5598 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
5599 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5601 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
5602 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5603 to <tt>op2</tt>.</li>
5605 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
5606 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5608 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
5609 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5611 <li><tt>sgt</tt>: interprets the operands as signed values and yields
5612 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5614 <li><tt>sge</tt>: interprets the operands as signed values and yields
5615 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5616 to <tt>op2</tt>.</li>
5618 <li><tt>slt</tt>: interprets the operands as signed values and yields
5619 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5621 <li><tt>sle</tt>: interprets the operands as signed values and yields
5622 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5625 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
5626 values are compared as if they were integers.</p>
5628 <p>If the operands are integer vectors, then they are compared element by
5629 element. The result is an <tt>i1</tt> vector with the same number of elements
5630 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
5634 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
5635 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
5636 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
5637 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
5638 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
5639 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
5642 <p>Note that the code generator does not yet support vector types with
5643 the <tt>icmp</tt> instruction.</p>
5647 <!-- _______________________________________________________________________ -->
5649 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
5656 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5660 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
5661 values based on comparison of its operands.</p>
5663 <p>If the operands are floating point scalars, then the result type is a boolean
5664 (<a href="#t_integer"><tt>i1</tt></a>).</p>
5666 <p>If the operands are floating point vectors, then the result type is a vector
5667 of boolean with the same number of elements as the operands being
5671 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
5672 the condition code indicating the kind of comparison to perform. It is not a
5673 value, just a keyword. The possible condition code are:</p>
5676 <li><tt>false</tt>: no comparison, always returns false</li>
5677 <li><tt>oeq</tt>: ordered and equal</li>
5678 <li><tt>ogt</tt>: ordered and greater than </li>
5679 <li><tt>oge</tt>: ordered and greater than or equal</li>
5680 <li><tt>olt</tt>: ordered and less than </li>
5681 <li><tt>ole</tt>: ordered and less than or equal</li>
5682 <li><tt>one</tt>: ordered and not equal</li>
5683 <li><tt>ord</tt>: ordered (no nans)</li>
5684 <li><tt>ueq</tt>: unordered or equal</li>
5685 <li><tt>ugt</tt>: unordered or greater than </li>
5686 <li><tt>uge</tt>: unordered or greater than or equal</li>
5687 <li><tt>ult</tt>: unordered or less than </li>
5688 <li><tt>ule</tt>: unordered or less than or equal</li>
5689 <li><tt>une</tt>: unordered or not equal</li>
5690 <li><tt>uno</tt>: unordered (either nans)</li>
5691 <li><tt>true</tt>: no comparison, always returns true</li>
5694 <p><i>Ordered</i> means that neither operand is a QNAN while
5695 <i>unordered</i> means that either operand may be a QNAN.</p>
5697 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
5698 a <a href="#t_floating">floating point</a> type or
5699 a <a href="#t_vector">vector</a> of floating point type. They must have
5700 identical types.</p>
5703 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
5704 according to the condition code given as <tt>cond</tt>. If the operands are
5705 vectors, then the vectors are compared element by element. Each comparison
5706 performed always yields an <a href="#t_integer">i1</a> result, as
5710 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
5712 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5713 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5715 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5716 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5718 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5719 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5721 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5722 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5724 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5725 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5727 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5728 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5730 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
5732 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
5733 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5735 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
5736 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5738 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
5739 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5741 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
5742 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5744 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
5745 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5747 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
5748 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5750 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
5752 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5757 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
5758 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5759 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5760 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
5763 <p>Note that the code generator does not yet support vector types with
5764 the <tt>fcmp</tt> instruction.</p>
5768 <!-- _______________________________________________________________________ -->
5770 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
5777 <result> = phi <ty> [ <val0>, <label0>], ...
5781 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
5782 SSA graph representing the function.</p>
5785 <p>The type of the incoming values is specified with the first type field. After
5786 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5787 one pair for each predecessor basic block of the current block. Only values
5788 of <a href="#t_firstclass">first class</a> type may be used as the value
5789 arguments to the PHI node. Only labels may be used as the label
5792 <p>There must be no non-phi instructions between the start of a basic block and
5793 the PHI instructions: i.e. PHI instructions must be first in a basic
5796 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5797 occur on the edge from the corresponding predecessor block to the current
5798 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5799 value on the same edge).</p>
5802 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
5803 specified by the pair corresponding to the predecessor basic block that
5804 executed just prior to the current block.</p>
5808 Loop: ; Infinite loop that counts from 0 on up...
5809 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5810 %nextindvar = add i32 %indvar, 1
5816 <!-- _______________________________________________________________________ -->
5818 <a name="i_select">'<tt>select</tt>' Instruction</a>
5825 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5827 <i>selty</i> is either i1 or {<N x i1>}
5831 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
5832 condition, without branching.</p>
5836 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5837 values indicating the condition, and two values of the
5838 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
5839 vectors and the condition is a scalar, then entire vectors are selected, not
5840 individual elements.</p>
5843 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
5844 first value argument; otherwise, it returns the second value argument.</p>
5846 <p>If the condition is a vector of i1, then the value arguments must be vectors
5847 of the same size, and the selection is done element by element.</p>
5851 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
5854 <p>Note that the code generator does not yet support conditions
5855 with vector type.</p>
5859 <!-- _______________________________________________________________________ -->
5861 <a name="i_call">'<tt>call</tt>' Instruction</a>
5868 <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>]
5872 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
5875 <p>This instruction requires several arguments:</p>
5878 <li>The optional "tail" marker indicates that the callee function does not
5879 access any allocas or varargs in the caller. Note that calls may be
5880 marked "tail" even if they do not occur before
5881 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
5882 present, the function call is eligible for tail call optimization,
5883 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
5884 optimized into a jump</a>. The code generator may optimize calls marked
5885 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
5886 sibling call optimization</a> when the caller and callee have
5887 matching signatures, or 2) forced tail call optimization when the
5888 following extra requirements are met:
5890 <li>Caller and callee both have the calling
5891 convention <tt>fastcc</tt>.</li>
5892 <li>The call is in tail position (ret immediately follows call and ret
5893 uses value of call or is void).</li>
5894 <li>Option <tt>-tailcallopt</tt> is enabled,
5895 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
5896 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
5897 constraints are met.</a></li>
5901 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5902 convention</a> the call should use. If none is specified, the call
5903 defaults to using C calling conventions. The calling convention of the
5904 call must match the calling convention of the target function, or else the
5905 behavior is undefined.</li>
5907 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5908 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5909 '<tt>inreg</tt>' attributes are valid here.</li>
5911 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5912 type of the return value. Functions that return no value are marked
5913 <tt><a href="#t_void">void</a></tt>.</li>
5915 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5916 being invoked. The argument types must match the types implied by this
5917 signature. This type can be omitted if the function is not varargs and if
5918 the function type does not return a pointer to a function.</li>
5920 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5921 be invoked. In most cases, this is a direct function invocation, but
5922 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5923 to function value.</li>
5925 <li>'<tt>function args</tt>': argument list whose types match the function
5926 signature argument types and parameter attributes. All arguments must be
5927 of <a href="#t_firstclass">first class</a> type. If the function
5928 signature indicates the function accepts a variable number of arguments,
5929 the extra arguments can be specified.</li>
5931 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5932 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5933 '<tt>readnone</tt>' attributes are valid here.</li>
5937 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5938 a specified function, with its incoming arguments bound to the specified
5939 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5940 function, control flow continues with the instruction after the function
5941 call, and the return value of the function is bound to the result
5946 %retval = call i32 @test(i32 %argc)
5947 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
5948 %X = tail call i32 @foo() <i>; yields i32</i>
5949 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5950 call void %foo(i8 97 signext)
5952 %struct.A = type { i32, i8 }
5953 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
5954 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5955 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
5956 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
5957 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
5960 <p>llvm treats calls to some functions with names and arguments that match the
5961 standard C99 library as being the C99 library functions, and may perform
5962 optimizations or generate code for them under that assumption. This is
5963 something we'd like to change in the future to provide better support for
5964 freestanding environments and non-C-based languages.</p>
5968 <!-- _______________________________________________________________________ -->
5970 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5977 <resultval> = va_arg <va_list*> <arglist>, <argty>
5981 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
5982 the "variable argument" area of a function call. It is used to implement the
5983 <tt>va_arg</tt> macro in C.</p>
5986 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
5987 argument. It returns a value of the specified argument type and increments
5988 the <tt>va_list</tt> to point to the next argument. The actual type
5989 of <tt>va_list</tt> is target specific.</p>
5992 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
5993 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
5994 to the next argument. For more information, see the variable argument
5995 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
5997 <p>It is legal for this instruction to be called in a function which does not
5998 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6001 <p><tt>va_arg</tt> is an LLVM instruction instead of
6002 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6006 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6008 <p>Note that the code generator does not yet fully support va_arg on many
6009 targets. Also, it does not currently support va_arg with aggregate types on
6014 <!-- _______________________________________________________________________ -->
6016 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6023 <resultval> = landingpad <somety> personality <type> <pers_fn> <clause>+
6024 <resultval> = landingpad <somety> personality <type> <pers_fn> cleanup <clause>*
6026 <clause> := catch <type> <value>
6027 <clause> := filter <array constant type> <array constant>
6031 <p>The '<tt>landingpad</tt>' instruction is used by
6032 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6033 system</a> to specify that a basic block is a landing pad — one where
6034 the exception lands, and corresponds to the code found in the
6035 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6036 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6037 re-entry to the function. The <tt>resultval</tt> has the
6038 type <tt>somety</tt>.</p>
6041 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6042 function associated with the unwinding mechanism. The optional
6043 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6045 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6046 or <tt>filter</tt> — and contains the global variable representing the
6047 "type" that may be caught or filtered respectively. Unlike the
6048 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6049 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6050 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6051 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6054 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6055 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6056 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6057 calling conventions, how the personality function results are represented in
6058 LLVM IR is target specific.</p>
6060 <p>The clauses are applied in order from top to bottom. If two
6061 <tt>landingpad</tt> instructions are merged together through inlining, the
6062 clauses from the calling function are appended to the list of clauses.</p>
6064 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6067 <li>A landing pad block is a basic block which is the unwind destination of an
6068 '<tt>invoke</tt>' instruction.</li>
6069 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6070 first non-PHI instruction.</li>
6071 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6073 <li>A basic block that is not a landing pad block may not include a
6074 '<tt>landingpad</tt>' instruction.</li>
6075 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6076 personality function.</li>
6081 ;; A landing pad which can catch an integer.
6082 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6084 ;; A landing pad that is a cleanup.
6085 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6087 ;; A landing pad which can catch an integer and can only throw a double.
6088 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6090 filter [1 x i8**] [@_ZTId]
6099 <!-- *********************************************************************** -->
6100 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6101 <!-- *********************************************************************** -->
6105 <p>LLVM supports the notion of an "intrinsic function". These functions have
6106 well known names and semantics and are required to follow certain
6107 restrictions. Overall, these intrinsics represent an extension mechanism for
6108 the LLVM language that does not require changing all of the transformations
6109 in LLVM when adding to the language (or the bitcode reader/writer, the
6110 parser, etc...).</p>
6112 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6113 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6114 begin with this prefix. Intrinsic functions must always be external
6115 functions: you cannot define the body of intrinsic functions. Intrinsic
6116 functions may only be used in call or invoke instructions: it is illegal to
6117 take the address of an intrinsic function. Additionally, because intrinsic
6118 functions are part of the LLVM language, it is required if any are added that
6119 they be documented here.</p>
6121 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6122 family of functions that perform the same operation but on different data
6123 types. Because LLVM can represent over 8 million different integer types,
6124 overloading is used commonly to allow an intrinsic function to operate on any
6125 integer type. One or more of the argument types or the result type can be
6126 overloaded to accept any integer type. Argument types may also be defined as
6127 exactly matching a previous argument's type or the result type. This allows
6128 an intrinsic function which accepts multiple arguments, but needs all of them
6129 to be of the same type, to only be overloaded with respect to a single
6130 argument or the result.</p>
6132 <p>Overloaded intrinsics will have the names of its overloaded argument types
6133 encoded into its function name, each preceded by a period. Only those types
6134 which are overloaded result in a name suffix. Arguments whose type is matched
6135 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6136 can take an integer of any width and returns an integer of exactly the same
6137 integer width. This leads to a family of functions such as
6138 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6139 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6140 suffix is required. Because the argument's type is matched against the return
6141 type, it does not require its own name suffix.</p>
6143 <p>To learn how to add an intrinsic function, please see the
6144 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6146 <!-- ======================================================================= -->
6148 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6153 <p>Variable argument support is defined in LLVM with
6154 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6155 intrinsic functions. These functions are related to the similarly named
6156 macros defined in the <tt><stdarg.h></tt> header file.</p>
6158 <p>All of these functions operate on arguments that use a target-specific value
6159 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6160 not define what this type is, so all transformations should be prepared to
6161 handle these functions regardless of the type used.</p>
6163 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6164 instruction and the variable argument handling intrinsic functions are
6167 <pre class="doc_code">
6168 define i32 @test(i32 %X, ...) {
6169 ; Initialize variable argument processing
6171 %ap2 = bitcast i8** %ap to i8*
6172 call void @llvm.va_start(i8* %ap2)
6174 ; Read a single integer argument
6175 %tmp = va_arg i8** %ap, i32
6177 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6179 %aq2 = bitcast i8** %aq to i8*
6180 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6181 call void @llvm.va_end(i8* %aq2)
6183 ; Stop processing of arguments.
6184 call void @llvm.va_end(i8* %ap2)
6188 declare void @llvm.va_start(i8*)
6189 declare void @llvm.va_copy(i8*, i8*)
6190 declare void @llvm.va_end(i8*)
6193 <!-- _______________________________________________________________________ -->
6195 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6203 declare void %llvm.va_start(i8* <arglist>)
6207 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6208 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6211 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6214 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6215 macro available in C. In a target-dependent way, it initializes
6216 the <tt>va_list</tt> element to which the argument points, so that the next
6217 call to <tt>va_arg</tt> will produce the first variable argument passed to
6218 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6219 need to know the last argument of the function as the compiler can figure
6224 <!-- _______________________________________________________________________ -->
6226 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6233 declare void @llvm.va_end(i8* <arglist>)
6237 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6238 which has been initialized previously
6239 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6240 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6243 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6246 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6247 macro available in C. In a target-dependent way, it destroys
6248 the <tt>va_list</tt> element to which the argument points. Calls
6249 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6250 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6251 with calls to <tt>llvm.va_end</tt>.</p>
6255 <!-- _______________________________________________________________________ -->
6257 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6264 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6268 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6269 from the source argument list to the destination argument list.</p>
6272 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6273 The second argument is a pointer to a <tt>va_list</tt> element to copy
6277 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6278 macro available in C. In a target-dependent way, it copies the
6279 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6280 element. This intrinsic is necessary because
6281 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6282 arbitrarily complex and require, for example, memory allocation.</p>
6290 <!-- ======================================================================= -->
6292 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6297 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6298 Collection</a> (GC) requires the implementation and generation of these
6299 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6300 roots on the stack</a>, as well as garbage collector implementations that
6301 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6302 barriers. Front-ends for type-safe garbage collected languages should generate
6303 these intrinsics to make use of the LLVM garbage collectors. For more details,
6304 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6307 <p>The garbage collection intrinsics only operate on objects in the generic
6308 address space (address space zero).</p>
6310 <!-- _______________________________________________________________________ -->
6312 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6319 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6323 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6324 the code generator, and allows some metadata to be associated with it.</p>
6327 <p>The first argument specifies the address of a stack object that contains the
6328 root pointer. The second pointer (which must be either a constant or a
6329 global value address) contains the meta-data to be associated with the
6333 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6334 location. At compile-time, the code generator generates information to allow
6335 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6336 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6341 <!-- _______________________________________________________________________ -->
6343 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6350 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6354 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6355 locations, allowing garbage collector implementations that require read
6359 <p>The second argument is the address to read from, which should be an address
6360 allocated from the garbage collector. The first object is a pointer to the
6361 start of the referenced object, if needed by the language runtime (otherwise
6365 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6366 instruction, but may be replaced with substantially more complex code by the
6367 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6368 may only be used in a function which <a href="#gc">specifies a GC
6373 <!-- _______________________________________________________________________ -->
6375 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6382 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6386 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6387 locations, allowing garbage collector implementations that require write
6388 barriers (such as generational or reference counting collectors).</p>
6391 <p>The first argument is the reference to store, the second is the start of the
6392 object to store it to, and the third is the address of the field of Obj to
6393 store to. If the runtime does not require a pointer to the object, Obj may
6397 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6398 instruction, but may be replaced with substantially more complex code by the
6399 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6400 may only be used in a function which <a href="#gc">specifies a GC
6407 <!-- ======================================================================= -->
6409 <a name="int_codegen">Code Generator Intrinsics</a>
6414 <p>These intrinsics are provided by LLVM to expose special features that may
6415 only be implemented with code generator support.</p>
6417 <!-- _______________________________________________________________________ -->
6419 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6426 declare i8 *@llvm.returnaddress(i32 <level>)
6430 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6431 target-specific value indicating the return address of the current function
6432 or one of its callers.</p>
6435 <p>The argument to this intrinsic indicates which function to return the address
6436 for. Zero indicates the calling function, one indicates its caller, etc.
6437 The argument is <b>required</b> to be a constant integer value.</p>
6440 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6441 indicating the return address of the specified call frame, or zero if it
6442 cannot be identified. The value returned by this intrinsic is likely to be
6443 incorrect or 0 for arguments other than zero, so it should only be used for
6444 debugging purposes.</p>
6446 <p>Note that calling this intrinsic does not prevent function inlining or other
6447 aggressive transformations, so the value returned may not be that of the
6448 obvious source-language caller.</p>
6452 <!-- _______________________________________________________________________ -->
6454 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6461 declare i8* @llvm.frameaddress(i32 <level>)
6465 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6466 target-specific frame pointer value for the specified stack frame.</p>
6469 <p>The argument to this intrinsic indicates which function to return the frame
6470 pointer for. Zero indicates the calling function, one indicates its caller,
6471 etc. The argument is <b>required</b> to be a constant integer value.</p>
6474 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6475 indicating the frame address of the specified call frame, or zero if it
6476 cannot be identified. The value returned by this intrinsic is likely to be
6477 incorrect or 0 for arguments other than zero, so it should only be used for
6478 debugging purposes.</p>
6480 <p>Note that calling this intrinsic does not prevent function inlining or other
6481 aggressive transformations, so the value returned may not be that of the
6482 obvious source-language caller.</p>
6486 <!-- _______________________________________________________________________ -->
6488 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6495 declare i8* @llvm.stacksave()
6499 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6500 of the function stack, for use
6501 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6502 useful for implementing language features like scoped automatic variable
6503 sized arrays in C99.</p>
6506 <p>This intrinsic returns a opaque pointer value that can be passed
6507 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6508 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6509 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6510 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6511 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6512 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6516 <!-- _______________________________________________________________________ -->
6518 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6525 declare void @llvm.stackrestore(i8* %ptr)
6529 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6530 the function stack to the state it was in when the
6531 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6532 executed. This is useful for implementing language features like scoped
6533 automatic variable sized arrays in C99.</p>
6536 <p>See the description
6537 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6541 <!-- _______________________________________________________________________ -->
6543 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6550 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6554 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6555 insert a prefetch instruction if supported; otherwise, it is a noop.
6556 Prefetches have no effect on the behavior of the program but can change its
6557 performance characteristics.</p>
6560 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6561 specifier determining if the fetch should be for a read (0) or write (1),
6562 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6563 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6564 specifies whether the prefetch is performed on the data (1) or instruction (0)
6565 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6566 must be constant integers.</p>
6569 <p>This intrinsic does not modify the behavior of the program. In particular,
6570 prefetches cannot trap and do not produce a value. On targets that support
6571 this intrinsic, the prefetch can provide hints to the processor cache for
6572 better performance.</p>
6576 <!-- _______________________________________________________________________ -->
6578 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
6585 declare void @llvm.pcmarker(i32 <id>)
6589 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
6590 Counter (PC) in a region of code to simulators and other tools. The method
6591 is target specific, but it is expected that the marker will use exported
6592 symbols to transmit the PC of the marker. The marker makes no guarantees
6593 that it will remain with any specific instruction after optimizations. It is
6594 possible that the presence of a marker will inhibit optimizations. The
6595 intended use is to be inserted after optimizations to allow correlations of
6596 simulation runs.</p>
6599 <p><tt>id</tt> is a numerical id identifying the marker.</p>
6602 <p>This intrinsic does not modify the behavior of the program. Backends that do
6603 not support this intrinsic may ignore it.</p>
6607 <!-- _______________________________________________________________________ -->
6609 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
6616 declare i64 @llvm.readcyclecounter()
6620 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
6621 counter register (or similar low latency, high accuracy clocks) on those
6622 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6623 should map to RPCC. As the backing counters overflow quickly (on the order
6624 of 9 seconds on alpha), this should only be used for small timings.</p>
6627 <p>When directly supported, reading the cycle counter should not modify any
6628 memory. Implementations are allowed to either return a application specific
6629 value or a system wide value. On backends without support, this is lowered
6630 to a constant 0.</p>
6636 <!-- ======================================================================= -->
6638 <a name="int_libc">Standard C Library Intrinsics</a>
6643 <p>LLVM provides intrinsics for a few important standard C library functions.
6644 These intrinsics allow source-language front-ends to pass information about
6645 the alignment of the pointer arguments to the code generator, providing
6646 opportunity for more efficient code generation.</p>
6648 <!-- _______________________________________________________________________ -->
6650 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
6656 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
6657 integer bit width and for different address spaces. Not all targets support
6658 all bit widths however.</p>
6661 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6662 i32 <len>, i32 <align>, i1 <isvolatile>)
6663 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6664 i64 <len>, i32 <align>, i1 <isvolatile>)
6668 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6669 source location to the destination location.</p>
6671 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
6672 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6673 and the pointers can be in specified address spaces.</p>
6677 <p>The first argument is a pointer to the destination, the second is a pointer
6678 to the source. The third argument is an integer argument specifying the
6679 number of bytes to copy, the fourth argument is the alignment of the
6680 source and destination locations, and the fifth is a boolean indicating a
6681 volatile access.</p>
6683 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6684 then the caller guarantees that both the source and destination pointers are
6685 aligned to that boundary.</p>
6687 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6688 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
6689 The detailed access behavior is not very cleanly specified and it is unwise
6690 to depend on it.</p>
6694 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6695 source location to the destination location, which are not allowed to
6696 overlap. It copies "len" bytes of memory over. If the argument is known to
6697 be aligned to some boundary, this can be specified as the fourth argument,
6698 otherwise it should be set to 0 or 1.</p>
6702 <!-- _______________________________________________________________________ -->
6704 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
6710 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6711 width and for different address space. Not all targets support all bit
6715 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6716 i32 <len>, i32 <align>, i1 <isvolatile>)
6717 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6718 i64 <len>, i32 <align>, i1 <isvolatile>)
6722 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
6723 source location to the destination location. It is similar to the
6724 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
6727 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
6728 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6729 and the pointers can be in specified address spaces.</p>
6733 <p>The first argument is a pointer to the destination, the second is a pointer
6734 to the source. The third argument is an integer argument specifying the
6735 number of bytes to copy, the fourth argument is the alignment of the
6736 source and destination locations, and the fifth is a boolean indicating a
6737 volatile access.</p>
6739 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6740 then the caller guarantees that the source and destination pointers are
6741 aligned to that boundary.</p>
6743 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6744 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
6745 The detailed access behavior is not very cleanly specified and it is unwise
6746 to depend on it.</p>
6750 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
6751 source location to the destination location, which may overlap. It copies
6752 "len" bytes of memory over. If the argument is known to be aligned to some
6753 boundary, this can be specified as the fourth argument, otherwise it should
6754 be set to 0 or 1.</p>
6758 <!-- _______________________________________________________________________ -->
6760 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
6766 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6767 width and for different address spaces. However, not all targets support all
6771 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6772 i32 <len>, i32 <align>, i1 <isvolatile>)
6773 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6774 i64 <len>, i32 <align>, i1 <isvolatile>)
6778 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
6779 particular byte value.</p>
6781 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
6782 intrinsic does not return a value and takes extra alignment/volatile
6783 arguments. Also, the destination can be in an arbitrary address space.</p>
6786 <p>The first argument is a pointer to the destination to fill, the second is the
6787 byte value with which to fill it, the third argument is an integer argument
6788 specifying the number of bytes to fill, and the fourth argument is the known
6789 alignment of the destination location.</p>
6791 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6792 then the caller guarantees that the destination pointer is aligned to that
6795 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6796 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
6797 The detailed access behavior is not very cleanly specified and it is unwise
6798 to depend on it.</p>
6801 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
6802 at the destination location. If the argument is known to be aligned to some
6803 boundary, this can be specified as the fourth argument, otherwise it should
6804 be set to 0 or 1.</p>
6808 <!-- _______________________________________________________________________ -->
6810 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
6816 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
6817 floating point or vector of floating point type. Not all targets support all
6821 declare float @llvm.sqrt.f32(float %Val)
6822 declare double @llvm.sqrt.f64(double %Val)
6823 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6824 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6825 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6829 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
6830 returning the same value as the libm '<tt>sqrt</tt>' functions would.
6831 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
6832 behavior for negative numbers other than -0.0 (which allows for better
6833 optimization, because there is no need to worry about errno being
6834 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
6837 <p>The argument and return value are floating point numbers of the same
6841 <p>This function returns the sqrt of the specified operand if it is a
6842 nonnegative floating point number.</p>
6846 <!-- _______________________________________________________________________ -->
6848 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
6854 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
6855 floating point or vector of floating point type. Not all targets support all
6859 declare float @llvm.powi.f32(float %Val, i32 %power)
6860 declare double @llvm.powi.f64(double %Val, i32 %power)
6861 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6862 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6863 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6867 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
6868 specified (positive or negative) power. The order of evaluation of
6869 multiplications is not defined. When a vector of floating point type is
6870 used, the second argument remains a scalar integer value.</p>
6873 <p>The second argument is an integer power, and the first is a value to raise to
6877 <p>This function returns the first value raised to the second power with an
6878 unspecified sequence of rounding operations.</p>
6882 <!-- _______________________________________________________________________ -->
6884 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
6890 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
6891 floating point or vector of floating point type. Not all targets support all
6895 declare float @llvm.sin.f32(float %Val)
6896 declare double @llvm.sin.f64(double %Val)
6897 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6898 declare fp128 @llvm.sin.f128(fp128 %Val)
6899 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6903 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
6906 <p>The argument and return value are floating point numbers of the same
6910 <p>This function returns the sine of the specified operand, returning the same
6911 values as the libm <tt>sin</tt> functions would, and handles error conditions
6912 in the same way.</p>
6916 <!-- _______________________________________________________________________ -->
6918 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
6924 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
6925 floating point or vector of floating point type. Not all targets support all
6929 declare float @llvm.cos.f32(float %Val)
6930 declare double @llvm.cos.f64(double %Val)
6931 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6932 declare fp128 @llvm.cos.f128(fp128 %Val)
6933 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6937 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
6940 <p>The argument and return value are floating point numbers of the same
6944 <p>This function returns the cosine of the specified operand, returning the same
6945 values as the libm <tt>cos</tt> functions would, and handles error conditions
6946 in the same way.</p>
6950 <!-- _______________________________________________________________________ -->
6952 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
6958 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
6959 floating point or vector of floating point type. Not all targets support all
6963 declare float @llvm.pow.f32(float %Val, float %Power)
6964 declare double @llvm.pow.f64(double %Val, double %Power)
6965 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6966 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6967 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6971 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
6972 specified (positive or negative) power.</p>
6975 <p>The second argument is a floating point power, and the first is a value to
6976 raise to that power.</p>
6979 <p>This function returns the first value raised to the second power, returning
6980 the same values as the libm <tt>pow</tt> functions would, and handles error
6981 conditions in the same way.</p>
6987 <!-- _______________________________________________________________________ -->
6989 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
6995 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
6996 floating point or vector of floating point type. Not all targets support all
7000 declare float @llvm.exp.f32(float %Val)
7001 declare double @llvm.exp.f64(double %Val)
7002 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7003 declare fp128 @llvm.exp.f128(fp128 %Val)
7004 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7008 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7011 <p>The argument and return value are floating point numbers of the same
7015 <p>This function returns the same values as the libm <tt>exp</tt> functions
7016 would, and handles error conditions in the same way.</p>
7020 <!-- _______________________________________________________________________ -->
7022 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7028 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7029 floating point or vector of floating point type. Not all targets support all
7033 declare float @llvm.log.f32(float %Val)
7034 declare double @llvm.log.f64(double %Val)
7035 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7036 declare fp128 @llvm.log.f128(fp128 %Val)
7037 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7041 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7044 <p>The argument and return value are floating point numbers of the same
7048 <p>This function returns the same values as the libm <tt>log</tt> functions
7049 would, and handles error conditions in the same way.</p>
7052 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7058 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7059 floating point or vector of floating point type. Not all targets support all
7063 declare float @llvm.fma.f32(float %a, float %b, float %c)
7064 declare double @llvm.fma.f64(double %a, double %b, double %c)
7065 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7066 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7067 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7071 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7075 <p>The argument and return value are floating point numbers of the same
7079 <p>This function returns the same values as the libm <tt>fma</tt> functions
7084 <!-- ======================================================================= -->
7086 <a name="int_manip">Bit Manipulation Intrinsics</a>
7091 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7092 These allow efficient code generation for some algorithms.</p>
7094 <!-- _______________________________________________________________________ -->
7096 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7102 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7103 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7106 declare i16 @llvm.bswap.i16(i16 <id>)
7107 declare i32 @llvm.bswap.i32(i32 <id>)
7108 declare i64 @llvm.bswap.i64(i64 <id>)
7112 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7113 values with an even number of bytes (positive multiple of 16 bits). These
7114 are useful for performing operations on data that is not in the target's
7115 native byte order.</p>
7118 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7119 and low byte of the input i16 swapped. Similarly,
7120 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7121 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7122 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7123 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7124 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7125 more, respectively).</p>
7129 <!-- _______________________________________________________________________ -->
7131 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7137 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7138 width, or on any vector with integer elements. Not all targets support all
7139 bit widths or vector types, however.</p>
7142 declare i8 @llvm.ctpop.i8(i8 <src>)
7143 declare i16 @llvm.ctpop.i16(i16 <src>)
7144 declare i32 @llvm.ctpop.i32(i32 <src>)
7145 declare i64 @llvm.ctpop.i64(i64 <src>)
7146 declare i256 @llvm.ctpop.i256(i256 <src>)
7147 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7151 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7155 <p>The only argument is the value to be counted. The argument may be of any
7156 integer type, or a vector with integer elements.
7157 The return type must match the argument type.</p>
7160 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7161 element of a vector.</p>
7165 <!-- _______________________________________________________________________ -->
7167 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7173 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7174 integer bit width, or any vector whose elements are integers. Not all
7175 targets support all bit widths or vector types, however.</p>
7178 declare i8 @llvm.ctlz.i8 (i8 <src>)
7179 declare i16 @llvm.ctlz.i16(i16 <src>)
7180 declare i32 @llvm.ctlz.i32(i32 <src>)
7181 declare i64 @llvm.ctlz.i64(i64 <src>)
7182 declare i256 @llvm.ctlz.i256(i256 <src>)
7183 declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src;gt)
7187 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7188 leading zeros in a variable.</p>
7191 <p>The only argument is the value to be counted. The argument may be of any
7192 integer type, or any vector type with integer element type.
7193 The return type must match the argument type.</p>
7196 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7197 zeros in a variable, or within each element of the vector if the operation
7198 is of vector type. If the src == 0 then the result is the size in bits of
7199 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7203 <!-- _______________________________________________________________________ -->
7205 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7211 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7212 integer bit width, or any vector of integer elements. Not all targets
7213 support all bit widths or vector types, however.</p>
7216 declare i8 @llvm.cttz.i8 (i8 <src>)
7217 declare i16 @llvm.cttz.i16(i16 <src>)
7218 declare i32 @llvm.cttz.i32(i32 <src>)
7219 declare i64 @llvm.cttz.i64(i64 <src>)
7220 declare i256 @llvm.cttz.i256(i256 <src>)
7221 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>)
7225 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7229 <p>The only argument is the value to be counted. The argument may be of any
7230 integer type, or a vectory with integer element type.. The return type
7231 must match the argument type.</p>
7234 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7235 zeros in a variable, or within each element of a vector.
7236 If the src == 0 then the result is the size in bits of
7237 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7243 <!-- ======================================================================= -->
7245 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7250 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7252 <!-- _______________________________________________________________________ -->
7254 <a name="int_sadd_overflow">
7255 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7262 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7263 on any integer bit width.</p>
7266 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7267 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7268 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7272 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7273 a signed addition of the two arguments, and indicate whether an overflow
7274 occurred during the signed summation.</p>
7277 <p>The arguments (%a and %b) and the first element of the result structure may
7278 be of integer types of any bit width, but they must have the same bit
7279 width. The second element of the result structure must be of
7280 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7281 undergo signed addition.</p>
7284 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7285 a signed addition of the two variables. They return a structure — the
7286 first element of which is the signed summation, and the second element of
7287 which is a bit specifying if the signed summation resulted in an
7292 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7293 %sum = extractvalue {i32, i1} %res, 0
7294 %obit = extractvalue {i32, i1} %res, 1
7295 br i1 %obit, label %overflow, label %normal
7300 <!-- _______________________________________________________________________ -->
7302 <a name="int_uadd_overflow">
7303 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7310 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7311 on any integer bit width.</p>
7314 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7315 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7316 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7320 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7321 an unsigned addition of the two arguments, and indicate whether a carry
7322 occurred during the unsigned summation.</p>
7325 <p>The arguments (%a and %b) and the first element of the result structure may
7326 be of integer types of any bit width, but they must have the same bit
7327 width. The second element of the result structure must be of
7328 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7329 undergo unsigned addition.</p>
7332 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7333 an unsigned addition of the two arguments. They return a structure —
7334 the first element of which is the sum, and the second element of which is a
7335 bit specifying if the unsigned summation resulted in a carry.</p>
7339 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7340 %sum = extractvalue {i32, i1} %res, 0
7341 %obit = extractvalue {i32, i1} %res, 1
7342 br i1 %obit, label %carry, label %normal
7347 <!-- _______________________________________________________________________ -->
7349 <a name="int_ssub_overflow">
7350 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7357 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7358 on any integer bit width.</p>
7361 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7362 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7363 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7367 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7368 a signed subtraction of the two arguments, and indicate whether an overflow
7369 occurred during the signed subtraction.</p>
7372 <p>The arguments (%a and %b) and the first element of the result structure may
7373 be of integer types of any bit width, but they must have the same bit
7374 width. The second element of the result structure must be of
7375 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7376 undergo signed subtraction.</p>
7379 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7380 a signed subtraction of the two arguments. They return a structure —
7381 the first element of which is the subtraction, and the second element of
7382 which is a bit specifying if the signed subtraction resulted in an
7387 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7388 %sum = extractvalue {i32, i1} %res, 0
7389 %obit = extractvalue {i32, i1} %res, 1
7390 br i1 %obit, label %overflow, label %normal
7395 <!-- _______________________________________________________________________ -->
7397 <a name="int_usub_overflow">
7398 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7405 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7406 on any integer bit width.</p>
7409 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7410 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7411 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7415 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7416 an unsigned subtraction of the two arguments, and indicate whether an
7417 overflow occurred during the unsigned subtraction.</p>
7420 <p>The arguments (%a and %b) and the first element of the result structure may
7421 be of integer types of any bit width, but they must have the same bit
7422 width. The second element of the result structure must be of
7423 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7424 undergo unsigned subtraction.</p>
7427 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7428 an unsigned subtraction of the two arguments. They return a structure —
7429 the first element of which is the subtraction, and the second element of
7430 which is a bit specifying if the unsigned subtraction resulted in an
7435 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7436 %sum = extractvalue {i32, i1} %res, 0
7437 %obit = extractvalue {i32, i1} %res, 1
7438 br i1 %obit, label %overflow, label %normal
7443 <!-- _______________________________________________________________________ -->
7445 <a name="int_smul_overflow">
7446 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7453 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7454 on any integer bit width.</p>
7457 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7458 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7459 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7464 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7465 a signed multiplication of the two arguments, and indicate whether an
7466 overflow occurred during the signed multiplication.</p>
7469 <p>The arguments (%a and %b) and the first element of the result structure may
7470 be of integer types of any bit width, but they must have the same bit
7471 width. The second element of the result structure must be of
7472 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7473 undergo signed multiplication.</p>
7476 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7477 a signed multiplication of the two arguments. They return a structure —
7478 the first element of which is the multiplication, and the second element of
7479 which is a bit specifying if the signed multiplication resulted in an
7484 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7485 %sum = extractvalue {i32, i1} %res, 0
7486 %obit = extractvalue {i32, i1} %res, 1
7487 br i1 %obit, label %overflow, label %normal
7492 <!-- _______________________________________________________________________ -->
7494 <a name="int_umul_overflow">
7495 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7502 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
7503 on any integer bit width.</p>
7506 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7507 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7508 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7512 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7513 a unsigned multiplication of the two arguments, and indicate whether an
7514 overflow occurred during the unsigned multiplication.</p>
7517 <p>The arguments (%a and %b) and the first element of the result structure may
7518 be of integer types of any bit width, but they must have the same bit
7519 width. The second element of the result structure must be of
7520 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7521 undergo unsigned multiplication.</p>
7524 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7525 an unsigned multiplication of the two arguments. They return a structure
7526 — the first element of which is the multiplication, and the second
7527 element of which is a bit specifying if the unsigned multiplication resulted
7532 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7533 %sum = extractvalue {i32, i1} %res, 0
7534 %obit = extractvalue {i32, i1} %res, 1
7535 br i1 %obit, label %overflow, label %normal
7542 <!-- ======================================================================= -->
7544 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
7549 <p>Half precision floating point is a storage-only format. This means that it is
7550 a dense encoding (in memory) but does not support computation in the
7553 <p>This means that code must first load the half-precision floating point
7554 value as an i16, then convert it to float with <a
7555 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
7556 Computation can then be performed on the float value (including extending to
7557 double etc). To store the value back to memory, it is first converted to
7558 float if needed, then converted to i16 with
7559 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
7560 storing as an i16 value.</p>
7562 <!-- _______________________________________________________________________ -->
7564 <a name="int_convert_to_fp16">
7565 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
7573 declare i16 @llvm.convert.to.fp16(f32 %a)
7577 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7578 a conversion from single precision floating point format to half precision
7579 floating point format.</p>
7582 <p>The intrinsic function contains single argument - the value to be
7586 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7587 a conversion from single precision floating point format to half precision
7588 floating point format. The return value is an <tt>i16</tt> which
7589 contains the converted number.</p>
7593 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7594 store i16 %res, i16* @x, align 2
7599 <!-- _______________________________________________________________________ -->
7601 <a name="int_convert_from_fp16">
7602 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
7610 declare f32 @llvm.convert.from.fp16(i16 %a)
7614 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
7615 a conversion from half precision floating point format to single precision
7616 floating point format.</p>
7619 <p>The intrinsic function contains single argument - the value to be
7623 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
7624 conversion from half single precision floating point format to single
7625 precision floating point format. The input half-float value is represented by
7626 an <tt>i16</tt> value.</p>
7630 %a = load i16* @x, align 2
7631 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7638 <!-- ======================================================================= -->
7640 <a name="int_debugger">Debugger Intrinsics</a>
7645 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
7646 prefix), are described in
7647 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
7648 Level Debugging</a> document.</p>
7652 <!-- ======================================================================= -->
7654 <a name="int_eh">Exception Handling Intrinsics</a>
7659 <p>The LLVM exception handling intrinsics (which all start with
7660 <tt>llvm.eh.</tt> prefix), are described in
7661 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
7662 Handling</a> document.</p>
7666 <!-- ======================================================================= -->
7668 <a name="int_trampoline">Trampoline Intrinsic</a>
7673 <p>This intrinsic makes it possible to excise one parameter, marked with
7674 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
7675 The result is a callable
7676 function pointer lacking the nest parameter - the caller does not need to
7677 provide a value for it. Instead, the value to use is stored in advance in a
7678 "trampoline", a block of memory usually allocated on the stack, which also
7679 contains code to splice the nest value into the argument list. This is used
7680 to implement the GCC nested function address extension.</p>
7682 <p>For example, if the function is
7683 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
7684 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
7687 <pre class="doc_code">
7688 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7689 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
7690 %p = call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval)
7691 %fp = bitcast i8* %p to i32 (i32, i32)*
7694 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
7695 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
7697 <!-- _______________________________________________________________________ -->
7700 '<tt>llvm.init.trampoline</tt>' Intrinsic
7708 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
7712 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
7713 function pointer suitable for executing it.</p>
7716 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
7717 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
7718 sufficiently aligned block of memory; this memory is written to by the
7719 intrinsic. Note that the size and the alignment are target-specific - LLVM
7720 currently provides no portable way of determining them, so a front-end that
7721 generates this intrinsic needs to have some target-specific knowledge.
7722 The <tt>func</tt> argument must hold a function bitcast to
7723 an <tt>i8*</tt>.</p>
7726 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
7727 dependent code, turning it into a function. A pointer to this function is
7728 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
7729 function pointer type</a> before being called. The new function's signature
7730 is the same as that of <tt>func</tt> with any arguments marked with
7731 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
7732 is allowed, and it must be of pointer type. Calling the new function is
7733 equivalent to calling <tt>func</tt> with the same argument list, but
7734 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
7735 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
7736 by <tt>tramp</tt> is modified, then the effect of any later call to the
7737 returned function pointer is undefined.</p>
7743 <!-- ======================================================================= -->
7745 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
7750 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
7751 hardware constructs for atomic operations and memory synchronization. This
7752 provides an interface to the hardware, not an interface to the programmer. It
7753 is aimed at a low enough level to allow any programming models or APIs
7754 (Application Programming Interfaces) which need atomic behaviors to map
7755 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
7756 hardware provides a "universal IR" for source languages, it also provides a
7757 starting point for developing a "universal" atomic operation and
7758 synchronization IR.</p>
7760 <p>These do <em>not</em> form an API such as high-level threading libraries,
7761 software transaction memory systems, atomic primitives, and intrinsic
7762 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
7763 application libraries. The hardware interface provided by LLVM should allow
7764 a clean implementation of all of these APIs and parallel programming models.
7765 No one model or paradigm should be selected above others unless the hardware
7766 itself ubiquitously does so.</p>
7768 <!-- _______________________________________________________________________ -->
7770 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
7776 declare void @llvm.memory.barrier(i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device>)
7780 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
7781 specific pairs of memory access types.</p>
7784 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
7785 The first four arguments enables a specific barrier as listed below. The
7786 fifth argument specifies that the barrier applies to io or device or uncached
7790 <li><tt>ll</tt>: load-load barrier</li>
7791 <li><tt>ls</tt>: load-store barrier</li>
7792 <li><tt>sl</tt>: store-load barrier</li>
7793 <li><tt>ss</tt>: store-store barrier</li>
7794 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
7798 <p>This intrinsic causes the system to enforce some ordering constraints upon
7799 the loads and stores of the program. This barrier does not
7800 indicate <em>when</em> any events will occur, it only enforces
7801 an <em>order</em> in which they occur. For any of the specified pairs of load
7802 and store operations (f.ex. load-load, or store-load), all of the first
7803 operations preceding the barrier will complete before any of the second
7804 operations succeeding the barrier begin. Specifically the semantics for each
7805 pairing is as follows:</p>
7808 <li><tt>ll</tt>: All loads before the barrier must complete before any load
7809 after the barrier begins.</li>
7810 <li><tt>ls</tt>: All loads before the barrier must complete before any
7811 store after the barrier begins.</li>
7812 <li><tt>ss</tt>: All stores before the barrier must complete before any
7813 store after the barrier begins.</li>
7814 <li><tt>sl</tt>: All stores before the barrier must complete before any
7815 load after the barrier begins.</li>
7818 <p>These semantics are applied with a logical "and" behavior when more than one
7819 is enabled in a single memory barrier intrinsic.</p>
7821 <p>Backends may implement stronger barriers than those requested when they do
7822 not support as fine grained a barrier as requested. Some architectures do
7823 not need all types of barriers and on such architectures, these become
7828 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7829 %ptr = bitcast i8* %mallocP to i32*
7832 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
7833 call void @llvm.memory.barrier(i1 false, i1 true, i1 false, i1 false, i1 true)
7834 <i>; guarantee the above finishes</i>
7835 store i32 8, %ptr <i>; before this begins</i>
7840 <!-- _______________________________________________________________________ -->
7842 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
7848 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
7849 any integer bit width and for different address spaces. Not all targets
7850 support all bit widths however.</p>
7853 declare i8 @llvm.atomic.cmp.swap.i8.p0i8(i8* <ptr>, i8 <cmp>, i8 <val>)
7854 declare i16 @llvm.atomic.cmp.swap.i16.p0i16(i16* <ptr>, i16 <cmp>, i16 <val>)
7855 declare i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* <ptr>, i32 <cmp>, i32 <val>)
7856 declare i64 @llvm.atomic.cmp.swap.i64.p0i64(i64* <ptr>, i64 <cmp>, i64 <val>)
7860 <p>This loads a value in memory and compares it to a given value. If they are
7861 equal, it stores a new value into the memory.</p>
7864 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
7865 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
7866 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
7867 this integer type. While any bit width integer may be used, targets may only
7868 lower representations they support in hardware.</p>
7871 <p>This entire intrinsic must be executed atomically. It first loads the value
7872 in memory pointed to by <tt>ptr</tt> and compares it with the
7873 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
7874 memory. The loaded value is yielded in all cases. This provides the
7875 equivalent of an atomic compare-and-swap operation within the SSA
7880 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7881 %ptr = bitcast i8* %mallocP to i32*
7884 %val1 = add i32 4, 4
7885 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 4, %val1)
7886 <i>; yields {i32}:result1 = 4</i>
7887 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7888 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7890 %val2 = add i32 1, 1
7891 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 5, %val2)
7892 <i>; yields {i32}:result2 = 8</i>
7893 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
7895 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
7900 <!-- _______________________________________________________________________ -->
7902 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
7908 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
7909 integer bit width. Not all targets support all bit widths however.</p>
7912 declare i8 @llvm.atomic.swap.i8.p0i8(i8* <ptr>, i8 <val>)
7913 declare i16 @llvm.atomic.swap.i16.p0i16(i16* <ptr>, i16 <val>)
7914 declare i32 @llvm.atomic.swap.i32.p0i32(i32* <ptr>, i32 <val>)
7915 declare i64 @llvm.atomic.swap.i64.p0i64(i64* <ptr>, i64 <val>)
7919 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
7920 the value from memory. It then stores the value in <tt>val</tt> in the memory
7921 at <tt>ptr</tt>.</p>
7924 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
7925 the <tt>val</tt> argument and the result must be integers of the same bit
7926 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
7927 integer type. The targets may only lower integer representations they
7931 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
7932 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
7933 equivalent of an atomic swap operation within the SSA framework.</p>
7937 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7938 %ptr = bitcast i8* %mallocP to i32*
7941 %val1 = add i32 4, 4
7942 %result1 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val1)
7943 <i>; yields {i32}:result1 = 4</i>
7944 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7945 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7947 %val2 = add i32 1, 1
7948 %result2 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val2)
7949 <i>; yields {i32}:result2 = 8</i>
7951 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
7952 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
7957 <!-- _______________________________________________________________________ -->
7959 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
7965 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
7966 any integer bit width. Not all targets support all bit widths however.</p>
7969 declare i8 @llvm.atomic.load.add.i8.p0i8(i8* <ptr>, i8 <delta>)
7970 declare i16 @llvm.atomic.load.add.i16.p0i16(i16* <ptr>, i16 <delta>)
7971 declare i32 @llvm.atomic.load.add.i32.p0i32(i32* <ptr>, i32 <delta>)
7972 declare i64 @llvm.atomic.load.add.i64.p0i64(i64* <ptr>, i64 <delta>)
7976 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
7977 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7980 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7981 and the second an integer value. The result is also an integer value. These
7982 integer types can have any bit width, but they must all have the same bit
7983 width. The targets may only lower integer representations they support.</p>
7986 <p>This intrinsic does a series of operations atomically. It first loads the
7987 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
7988 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
7992 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7993 %ptr = bitcast i8* %mallocP to i32*
7995 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 4)
7996 <i>; yields {i32}:result1 = 4</i>
7997 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 2)
7998 <i>; yields {i32}:result2 = 8</i>
7999 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 5)
8000 <i>; yields {i32}:result3 = 10</i>
8001 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
8006 <!-- _______________________________________________________________________ -->
8008 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
8014 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
8015 any integer bit width and for different address spaces. Not all targets
8016 support all bit widths however.</p>
8019 declare i8 @llvm.atomic.load.sub.i8.p0i32(i8* <ptr>, i8 <delta>)
8020 declare i16 @llvm.atomic.load.sub.i16.p0i32(i16* <ptr>, i16 <delta>)
8021 declare i32 @llvm.atomic.load.sub.i32.p0i32(i32* <ptr>, i32 <delta>)
8022 declare i64 @llvm.atomic.load.sub.i64.p0i32(i64* <ptr>, i64 <delta>)
8026 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
8027 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
8030 <p>The intrinsic takes two arguments, the first a pointer to an integer value
8031 and the second an integer value. The result is also an integer value. These
8032 integer types can have any bit width, but they must all have the same bit
8033 width. The targets may only lower integer representations they support.</p>
8036 <p>This intrinsic does a series of operations atomically. It first loads the
8037 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
8038 result to <tt>ptr</tt>. It yields the original value stored
8039 at <tt>ptr</tt>.</p>
8043 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8044 %ptr = bitcast i8* %mallocP to i32*
8046 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 4)
8047 <i>; yields {i32}:result1 = 8</i>
8048 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 2)
8049 <i>; yields {i32}:result2 = 4</i>
8050 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 5)
8051 <i>; yields {i32}:result3 = 2</i>
8052 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
8057 <!-- _______________________________________________________________________ -->
8059 <a name="int_atomic_load_and">
8060 '<tt>llvm.atomic.load.and.*</tt>' Intrinsic
8063 <a name="int_atomic_load_nand">
8064 '<tt>llvm.atomic.load.nand.*</tt>' Intrinsic
8067 <a name="int_atomic_load_or">
8068 '<tt>llvm.atomic.load.or.*</tt>' Intrinsic
8071 <a name="int_atomic_load_xor">
8072 '<tt>llvm.atomic.load.xor.*</tt>' Intrinsic
8079 <p>These are overloaded intrinsics. You can
8080 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
8081 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
8082 bit width and for different address spaces. Not all targets support all bit
8086 declare i8 @llvm.atomic.load.and.i8.p0i8(i8* <ptr>, i8 <delta>)
8087 declare i16 @llvm.atomic.load.and.i16.p0i16(i16* <ptr>, i16 <delta>)
8088 declare i32 @llvm.atomic.load.and.i32.p0i32(i32* <ptr>, i32 <delta>)
8089 declare i64 @llvm.atomic.load.and.i64.p0i64(i64* <ptr>, i64 <delta>)
8093 declare i8 @llvm.atomic.load.or.i8.p0i8(i8* <ptr>, i8 <delta>)
8094 declare i16 @llvm.atomic.load.or.i16.p0i16(i16* <ptr>, i16 <delta>)
8095 declare i32 @llvm.atomic.load.or.i32.p0i32(i32* <ptr>, i32 <delta>)
8096 declare i64 @llvm.atomic.load.or.i64.p0i64(i64* <ptr>, i64 <delta>)
8100 declare i8 @llvm.atomic.load.nand.i8.p0i32(i8* <ptr>, i8 <delta>)
8101 declare i16 @llvm.atomic.load.nand.i16.p0i32(i16* <ptr>, i16 <delta>)
8102 declare i32 @llvm.atomic.load.nand.i32.p0i32(i32* <ptr>, i32 <delta>)
8103 declare i64 @llvm.atomic.load.nand.i64.p0i32(i64* <ptr>, i64 <delta>)
8107 declare i8 @llvm.atomic.load.xor.i8.p0i32(i8* <ptr>, i8 <delta>)
8108 declare i16 @llvm.atomic.load.xor.i16.p0i32(i16* <ptr>, i16 <delta>)
8109 declare i32 @llvm.atomic.load.xor.i32.p0i32(i32* <ptr>, i32 <delta>)
8110 declare i64 @llvm.atomic.load.xor.i64.p0i32(i64* <ptr>, i64 <delta>)
8114 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
8115 the value stored in memory at <tt>ptr</tt>. It yields the original value
8116 at <tt>ptr</tt>.</p>
8119 <p>These intrinsics take two arguments, the first a pointer to an integer value
8120 and the second an integer value. The result is also an integer value. These
8121 integer types can have any bit width, but they must all have the same bit
8122 width. The targets may only lower integer representations they support.</p>
8125 <p>These intrinsics does a series of operations atomically. They first load the
8126 value stored at <tt>ptr</tt>. They then do the bitwise
8127 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
8128 original value stored at <tt>ptr</tt>.</p>
8132 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8133 %ptr = bitcast i8* %mallocP to i32*
8134 store i32 0x0F0F, %ptr
8135 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32(i32* %ptr, i32 0xFF)
8136 <i>; yields {i32}:result0 = 0x0F0F</i>
8137 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32(i32* %ptr, i32 0xFF)
8138 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
8139 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32(i32* %ptr, i32 0F)
8140 <i>; yields {i32}:result2 = 0xF0</i>
8141 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32(i32* %ptr, i32 0F)
8142 <i>; yields {i32}:result3 = FF</i>
8143 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
8148 <!-- _______________________________________________________________________ -->
8150 <a name="int_atomic_load_max">
8151 '<tt>llvm.atomic.load.max.*</tt>' Intrinsic
8154 <a name="int_atomic_load_min">
8155 '<tt>llvm.atomic.load.min.*</tt>' Intrinsic
8158 <a name="int_atomic_load_umax">
8159 '<tt>llvm.atomic.load.umax.*</tt>' Intrinsic
8162 <a name="int_atomic_load_umin">
8163 '<tt>llvm.atomic.load.umin.*</tt>' Intrinsic
8170 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
8171 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
8172 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
8173 address spaces. Not all targets support all bit widths however.</p>
8176 declare i8 @llvm.atomic.load.max.i8.p0i8(i8* <ptr>, i8 <delta>)
8177 declare i16 @llvm.atomic.load.max.i16.p0i16(i16* <ptr>, i16 <delta>)
8178 declare i32 @llvm.atomic.load.max.i32.p0i32(i32* <ptr>, i32 <delta>)
8179 declare i64 @llvm.atomic.load.max.i64.p0i64(i64* <ptr>, i64 <delta>)
8183 declare i8 @llvm.atomic.load.min.i8.p0i8(i8* <ptr>, i8 <delta>)
8184 declare i16 @llvm.atomic.load.min.i16.p0i16(i16* <ptr>, i16 <delta>)
8185 declare i32 @llvm.atomic.load.min.i32.p0i32(i32* <ptr>, i32 <delta>)
8186 declare i64 @llvm.atomic.load.min.i64.p0i64(i64* <ptr>, i64 <delta>)
8190 declare i8 @llvm.atomic.load.umax.i8.p0i8(i8* <ptr>, i8 <delta>)
8191 declare i16 @llvm.atomic.load.umax.i16.p0i16(i16* <ptr>, i16 <delta>)
8192 declare i32 @llvm.atomic.load.umax.i32.p0i32(i32* <ptr>, i32 <delta>)
8193 declare i64 @llvm.atomic.load.umax.i64.p0i64(i64* <ptr>, i64 <delta>)
8197 declare i8 @llvm.atomic.load.umin.i8.p0i8(i8* <ptr>, i8 <delta>)
8198 declare i16 @llvm.atomic.load.umin.i16.p0i16(i16* <ptr>, i16 <delta>)
8199 declare i32 @llvm.atomic.load.umin.i32.p0i32(i32* <ptr>, i32 <delta>)
8200 declare i64 @llvm.atomic.load.umin.i64.p0i64(i64* <ptr>, i64 <delta>)
8204 <p>These intrinsics takes the signed or unsigned minimum or maximum of
8205 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
8206 original value at <tt>ptr</tt>.</p>
8209 <p>These intrinsics take two arguments, the first a pointer to an integer value
8210 and the second an integer value. The result is also an integer value. These
8211 integer types can have any bit width, but they must all have the same bit
8212 width. The targets may only lower integer representations they support.</p>
8215 <p>These intrinsics does a series of operations atomically. They first load the
8216 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
8217 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
8218 yield the original value stored at <tt>ptr</tt>.</p>
8222 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8223 %ptr = bitcast i8* %mallocP to i32*
8225 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32(i32* %ptr, i32 -2)
8226 <i>; yields {i32}:result0 = 7</i>
8227 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32(i32* %ptr, i32 8)
8228 <i>; yields {i32}:result1 = -2</i>
8229 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32(i32* %ptr, i32 10)
8230 <i>; yields {i32}:result2 = 8</i>
8231 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32(i32* %ptr, i32 30)
8232 <i>; yields {i32}:result3 = 8</i>
8233 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
8240 <!-- ======================================================================= -->
8242 <a name="int_memorymarkers">Memory Use Markers</a>
8247 <p>This class of intrinsics exists to information about the lifetime of memory
8248 objects and ranges where variables are immutable.</p>
8250 <!-- _______________________________________________________________________ -->
8252 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8259 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8263 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8264 object's lifetime.</p>
8267 <p>The first argument is a constant integer representing the size of the
8268 object, or -1 if it is variable sized. The second argument is a pointer to
8272 <p>This intrinsic indicates that before this point in the code, the value of the
8273 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8274 never be used and has an undefined value. A load from the pointer that
8275 precedes this intrinsic can be replaced with
8276 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8280 <!-- _______________________________________________________________________ -->
8282 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8289 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8293 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8294 object's lifetime.</p>
8297 <p>The first argument is a constant integer representing the size of the
8298 object, or -1 if it is variable sized. The second argument is a pointer to
8302 <p>This intrinsic indicates that after this point in the code, the value of the
8303 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8304 never be used and has an undefined value. Any stores into the memory object
8305 following this intrinsic may be removed as dead.
8309 <!-- _______________________________________________________________________ -->
8311 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8318 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8322 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8323 a memory object will not change.</p>
8326 <p>The first argument is a constant integer representing the size of the
8327 object, or -1 if it is variable sized. The second argument is a pointer to
8331 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8332 the return value, the referenced memory location is constant and
8337 <!-- _______________________________________________________________________ -->
8339 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8346 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8350 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8351 a memory object are mutable.</p>
8354 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8355 The second argument is a constant integer representing the size of the
8356 object, or -1 if it is variable sized and the third argument is a pointer
8360 <p>This intrinsic indicates that the memory is mutable again.</p>
8366 <!-- ======================================================================= -->
8368 <a name="int_general">General Intrinsics</a>
8373 <p>This class of intrinsics is designed to be generic and has no specific
8376 <!-- _______________________________________________________________________ -->
8378 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8385 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8389 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8392 <p>The first argument is a pointer to a value, the second is a pointer to a
8393 global string, the third is a pointer to a global string which is the source
8394 file name, and the last argument is the line number.</p>
8397 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8398 This can be useful for special purpose optimizations that want to look for
8399 these annotations. These have no other defined use; they are ignored by code
8400 generation and optimization.</p>
8404 <!-- _______________________________________________________________________ -->
8406 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8412 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8413 any integer bit width.</p>
8416 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8417 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8418 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8419 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8420 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8424 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8427 <p>The first argument is an integer value (result of some expression), the
8428 second is a pointer to a global string, the third is a pointer to a global
8429 string which is the source file name, and the last argument is the line
8430 number. It returns the value of the first argument.</p>
8433 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8434 arbitrary strings. This can be useful for special purpose optimizations that
8435 want to look for these annotations. These have no other defined use; they
8436 are ignored by code generation and optimization.</p>
8440 <!-- _______________________________________________________________________ -->
8442 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8449 declare void @llvm.trap()
8453 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8459 <p>This intrinsics is lowered to the target dependent trap instruction. If the
8460 target does not have a trap instruction, this intrinsic will be lowered to
8461 the call of the <tt>abort()</tt> function.</p>
8465 <!-- _______________________________________________________________________ -->
8467 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8474 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8478 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8479 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8480 ensure that it is placed on the stack before local variables.</p>
8483 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8484 arguments. The first argument is the value loaded from the stack
8485 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8486 that has enough space to hold the value of the guard.</p>
8489 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8490 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8491 stack. This is to ensure that if a local variable on the stack is
8492 overwritten, it will destroy the value of the guard. When the function exits,
8493 the guard on the stack is checked against the original guard. If they are
8494 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8499 <!-- _______________________________________________________________________ -->
8501 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8508 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <type>)
8509 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <type>)
8513 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8514 the optimizers to determine at compile time whether a) an operation (like
8515 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8516 runtime check for overflow isn't necessary. An object in this context means
8517 an allocation of a specific class, structure, array, or other object.</p>
8520 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8521 argument is a pointer to or into the <tt>object</tt>. The second argument
8522 is a boolean 0 or 1. This argument determines whether you want the
8523 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
8524 1, variables are not allowed.</p>
8527 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
8528 representing the size of the object concerned, or <tt>i32/i64 -1 or 0</tt>,
8529 depending on the <tt>type</tt> argument, if the size cannot be determined at
8538 <!-- *********************************************************************** -->
8541 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8542 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8543 <a href="http://validator.w3.org/check/referer"><img
8544 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8546 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8547 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8548 Last modified: $Date$