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>
209 <li><a href="#intrinsics">Intrinsic Functions</a>
211 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
213 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
214 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
215 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
218 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
220 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
221 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
222 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
225 <li><a href="#int_codegen">Code Generator Intrinsics</a>
227 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
228 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
229 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
230 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
231 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
232 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
233 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
236 <li><a href="#int_libc">Standard C Library Intrinsics</a>
238 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
239 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
240 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
241 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
242 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
243 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
244 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
245 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
246 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
247 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
248 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
251 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
253 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
254 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
255 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
256 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
259 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
261 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
262 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
263 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
264 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
265 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
266 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
269 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
271 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
272 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
275 <li><a href="#int_debugger">Debugger intrinsics</a></li>
276 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
277 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
279 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
282 <li><a href="#int_atomics">Atomic intrinsics</a>
284 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
285 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
286 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
287 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
288 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
289 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
290 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
291 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
292 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
293 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
294 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
295 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
296 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
299 <li><a href="#int_memorymarkers">Memory Use Markers</a>
301 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
302 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
303 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
304 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
307 <li><a href="#int_general">General intrinsics</a>
309 <li><a href="#int_var_annotation">
310 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
311 <li><a href="#int_annotation">
312 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
313 <li><a href="#int_trap">
314 '<tt>llvm.trap</tt>' Intrinsic</a></li>
315 <li><a href="#int_stackprotector">
316 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
317 <li><a href="#int_objectsize">
318 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
325 <div class="doc_author">
326 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
327 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
330 <!-- *********************************************************************** -->
331 <h2><a name="abstract">Abstract</a></h2>
332 <!-- *********************************************************************** -->
336 <p>This document is a reference manual for the LLVM assembly language. LLVM is
337 a Static Single Assignment (SSA) based representation that provides type
338 safety, low-level operations, flexibility, and the capability of representing
339 'all' high-level languages cleanly. It is the common code representation
340 used throughout all phases of the LLVM compilation strategy.</p>
344 <!-- *********************************************************************** -->
345 <h2><a name="introduction">Introduction</a></h2>
346 <!-- *********************************************************************** -->
350 <p>The LLVM code representation is designed to be used in three different forms:
351 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
352 for fast loading by a Just-In-Time compiler), and as a human readable
353 assembly language representation. This allows LLVM to provide a powerful
354 intermediate representation for efficient compiler transformations and
355 analysis, while providing a natural means to debug and visualize the
356 transformations. The three different forms of LLVM are all equivalent. This
357 document describes the human readable representation and notation.</p>
359 <p>The LLVM representation aims to be light-weight and low-level while being
360 expressive, typed, and extensible at the same time. It aims to be a
361 "universal IR" of sorts, by being at a low enough level that high-level ideas
362 may be cleanly mapped to it (similar to how microprocessors are "universal
363 IR's", allowing many source languages to be mapped to them). By providing
364 type information, LLVM can be used as the target of optimizations: for
365 example, through pointer analysis, it can be proven that a C automatic
366 variable is never accessed outside of the current function, allowing it to
367 be promoted to a simple SSA value instead of a memory location.</p>
369 <!-- _______________________________________________________________________ -->
371 <a name="wellformed">Well-Formedness</a>
376 <p>It is important to note that this document describes 'well formed' LLVM
377 assembly language. There is a difference between what the parser accepts and
378 what is considered 'well formed'. For example, the following instruction is
379 syntactically okay, but not well formed:</p>
381 <pre class="doc_code">
382 %x = <a href="#i_add">add</a> i32 1, %x
385 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
386 LLVM infrastructure provides a verification pass that may be used to verify
387 that an LLVM module is well formed. This pass is automatically run by the
388 parser after parsing input assembly and by the optimizer before it outputs
389 bitcode. The violations pointed out by the verifier pass indicate bugs in
390 transformation passes or input to the parser.</p>
396 <!-- Describe the typesetting conventions here. -->
398 <!-- *********************************************************************** -->
399 <h2><a name="identifiers">Identifiers</a></h2>
400 <!-- *********************************************************************** -->
404 <p>LLVM identifiers come in two basic types: global and local. Global
405 identifiers (functions, global variables) begin with the <tt>'@'</tt>
406 character. Local identifiers (register names, types) begin with
407 the <tt>'%'</tt> character. Additionally, there are three different formats
408 for identifiers, for different purposes:</p>
411 <li>Named values are represented as a string of characters with their prefix.
412 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
413 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
414 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
415 other characters in their names can be surrounded with quotes. Special
416 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
417 ASCII code for the character in hexadecimal. In this way, any character
418 can be used in a name value, even quotes themselves.</li>
420 <li>Unnamed values are represented as an unsigned numeric value with their
421 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
423 <li>Constants, which are described in a <a href="#constants">section about
424 constants</a>, below.</li>
427 <p>LLVM requires that values start with a prefix for two reasons: Compilers
428 don't need to worry about name clashes with reserved words, and the set of
429 reserved words may be expanded in the future without penalty. Additionally,
430 unnamed identifiers allow a compiler to quickly come up with a temporary
431 variable without having to avoid symbol table conflicts.</p>
433 <p>Reserved words in LLVM are very similar to reserved words in other
434 languages. There are keywords for different opcodes
435 ('<tt><a href="#i_add">add</a></tt>',
436 '<tt><a href="#i_bitcast">bitcast</a></tt>',
437 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
438 ('<tt><a href="#t_void">void</a></tt>',
439 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
440 reserved words cannot conflict with variable names, because none of them
441 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
443 <p>Here is an example of LLVM code to multiply the integer variable
444 '<tt>%X</tt>' by 8:</p>
448 <pre class="doc_code">
449 %result = <a href="#i_mul">mul</a> i32 %X, 8
452 <p>After strength reduction:</p>
454 <pre class="doc_code">
455 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
458 <p>And the hard way:</p>
460 <pre class="doc_code">
461 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
462 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
463 %result = <a href="#i_add">add</a> i32 %1, %1
466 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
467 lexical features of LLVM:</p>
470 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
473 <li>Unnamed temporaries are created when the result of a computation is not
474 assigned to a named value.</li>
476 <li>Unnamed temporaries are numbered sequentially</li>
479 <p>It also shows a convention that we follow in this document. When
480 demonstrating instructions, we will follow an instruction with a comment that
481 defines the type and name of value produced. Comments are shown in italic
486 <!-- *********************************************************************** -->
487 <h2><a name="highlevel">High Level Structure</a></h2>
488 <!-- *********************************************************************** -->
490 <!-- ======================================================================= -->
492 <a name="modulestructure">Module Structure</a>
497 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
498 of the input programs. Each module consists of functions, global variables,
499 and symbol table entries. Modules may be combined together with the LLVM
500 linker, which merges function (and global variable) definitions, resolves
501 forward declarations, and merges symbol table entries. Here is an example of
502 the "hello world" module:</p>
504 <pre class="doc_code">
505 <i>; Declare the string constant as a global constant.</i>
506 <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>
508 <i>; External declaration of the puts function</i>
509 <a href="#functionstructure">declare</a> i32 @puts(i8*) <i>; i32 (i8*)* </i>
511 <i>; Definition of main function</i>
512 define i32 @main() { <i>; i32()* </i>
513 <i>; Convert [13 x i8]* to i8 *...</i>
514 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8*</i>
516 <i>; Call puts function to write out the string to stdout.</i>
517 <a href="#i_call">call</a> i32 @puts(i8* %cast210) <i>; i32</i>
518 <a href="#i_ret">ret</a> i32 0
521 <i>; Named metadata</i>
522 !1 = metadata !{i32 41}
526 <p>This example is made up of a <a href="#globalvars">global variable</a> named
527 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
528 a <a href="#functionstructure">function definition</a> for
529 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
532 <p>In general, a module is made up of a list of global values, where both
533 functions and global variables are global values. Global values are
534 represented by a pointer to a memory location (in this case, a pointer to an
535 array of char, and a pointer to a function), and have one of the
536 following <a href="#linkage">linkage types</a>.</p>
540 <!-- ======================================================================= -->
542 <a name="linkage">Linkage Types</a>
547 <p>All Global Variables and Functions have one of the following types of
551 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
552 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
553 by objects in the current module. In particular, linking code into a
554 module with an private global value may cause the private to be renamed as
555 necessary to avoid collisions. Because the symbol is private to the
556 module, all references can be updated. This doesn't show up in any symbol
557 table in the object file.</dd>
559 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
560 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
561 assembler and evaluated by the linker. Unlike normal strong symbols, they
562 are removed by the linker from the final linked image (executable or
563 dynamic library).</dd>
565 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
566 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
567 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
568 linker. The symbols are removed by the linker from the final linked image
569 (executable or dynamic library).</dd>
571 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
572 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
573 of the object is not taken. For instance, functions that had an inline
574 definition, but the compiler decided not to inline it. Note,
575 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
576 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
577 visibility. The symbols are removed by the linker from the final linked
578 image (executable or dynamic library).</dd>
580 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
581 <dd>Similar to private, but the value shows as a local symbol
582 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
583 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
585 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
586 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
587 into the object file corresponding to the LLVM module. They exist to
588 allow inlining and other optimizations to take place given knowledge of
589 the definition of the global, which is known to be somewhere outside the
590 module. Globals with <tt>available_externally</tt> linkage are allowed to
591 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
592 This linkage type is only allowed on definitions, not declarations.</dd>
594 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
595 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
596 the same name when linkage occurs. This can be used to implement
597 some forms of inline functions, templates, or other code which must be
598 generated in each translation unit that uses it, but where the body may
599 be overridden with a more definitive definition later. Unreferenced
600 <tt>linkonce</tt> globals are allowed to be discarded. Note that
601 <tt>linkonce</tt> linkage does not actually allow the optimizer to
602 inline the body of this function into callers because it doesn't know if
603 this definition of the function is the definitive definition within the
604 program or whether it will be overridden by a stronger definition.
605 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
608 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
609 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
610 <tt>linkonce</tt> linkage, except that unreferenced globals with
611 <tt>weak</tt> linkage may not be discarded. This is used for globals that
612 are declared "weak" in C source code.</dd>
614 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
615 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
616 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
618 Symbols with "<tt>common</tt>" linkage are merged in the same way as
619 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
620 <tt>common</tt> symbols may not have an explicit section,
621 must have a zero initializer, and may not be marked '<a
622 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
623 have common linkage.</dd>
626 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
627 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
628 pointer to array type. When two global variables with appending linkage
629 are linked together, the two global arrays are appended together. This is
630 the LLVM, typesafe, equivalent of having the system linker append together
631 "sections" with identical names when .o files are linked.</dd>
633 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
634 <dd>The semantics of this linkage follow the ELF object file model: the symbol
635 is weak until linked, if not linked, the symbol becomes null instead of
636 being an undefined reference.</dd>
638 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
639 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
640 <dd>Some languages allow differing globals to be merged, such as two functions
641 with different semantics. Other languages, such as <tt>C++</tt>, ensure
642 that only equivalent globals are ever merged (the "one definition rule"
643 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
644 and <tt>weak_odr</tt> linkage types to indicate that the global will only
645 be merged with equivalent globals. These linkage types are otherwise the
646 same as their non-<tt>odr</tt> versions.</dd>
648 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
649 <dd>If none of the above identifiers are used, the global is externally
650 visible, meaning that it participates in linkage and can be used to
651 resolve external symbol references.</dd>
654 <p>The next two types of linkage are targeted for Microsoft Windows platform
655 only. They are designed to support importing (exporting) symbols from (to)
656 DLLs (Dynamic Link Libraries).</p>
659 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
660 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
661 or variable via a global pointer to a pointer that is set up by the DLL
662 exporting the symbol. On Microsoft Windows targets, the pointer name is
663 formed by combining <code>__imp_</code> and the function or variable
666 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
667 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
668 pointer to a pointer in a DLL, so that it can be referenced with the
669 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
670 name is formed by combining <code>__imp_</code> and the function or
674 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
675 another module defined a "<tt>.LC0</tt>" variable and was linked with this
676 one, one of the two would be renamed, preventing a collision. Since
677 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
678 declarations), they are accessible outside of the current module.</p>
680 <p>It is illegal for a function <i>declaration</i> to have any linkage type
681 other than "externally visible", <tt>dllimport</tt>
682 or <tt>extern_weak</tt>.</p>
684 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
685 or <tt>weak_odr</tt> linkages.</p>
689 <!-- ======================================================================= -->
691 <a name="callingconv">Calling Conventions</a>
696 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
697 and <a href="#i_invoke">invokes</a> can all have an optional calling
698 convention specified for the call. The calling convention of any pair of
699 dynamic caller/callee must match, or the behavior of the program is
700 undefined. The following calling conventions are supported by LLVM, and more
701 may be added in the future:</p>
704 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
705 <dd>This calling convention (the default if no other calling convention is
706 specified) matches the target C calling conventions. This calling
707 convention supports varargs function calls and tolerates some mismatch in
708 the declared prototype and implemented declaration of the function (as
711 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
712 <dd>This calling convention attempts to make calls as fast as possible
713 (e.g. by passing things in registers). This calling convention allows the
714 target to use whatever tricks it wants to produce fast code for the
715 target, without having to conform to an externally specified ABI
716 (Application Binary Interface).
717 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
718 when this or the GHC convention is used.</a> This calling convention
719 does not support varargs and requires the prototype of all callees to
720 exactly match the prototype of the function definition.</dd>
722 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
723 <dd>This calling convention attempts to make code in the caller as efficient
724 as possible under the assumption that the call is not commonly executed.
725 As such, these calls often preserve all registers so that the call does
726 not break any live ranges in the caller side. This calling convention
727 does not support varargs and requires the prototype of all callees to
728 exactly match the prototype of the function definition.</dd>
730 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
731 <dd>This calling convention has been implemented specifically for use by the
732 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
733 It passes everything in registers, going to extremes to achieve this by
734 disabling callee save registers. This calling convention should not be
735 used lightly but only for specific situations such as an alternative to
736 the <em>register pinning</em> performance technique often used when
737 implementing functional programming languages.At the moment only X86
738 supports this convention and it has the following limitations:
740 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
741 floating point types are supported.</li>
742 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
743 6 floating point parameters.</li>
745 This calling convention supports
746 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
747 requires both the caller and callee are using it.
750 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
751 <dd>Any calling convention may be specified by number, allowing
752 target-specific calling conventions to be used. Target specific calling
753 conventions start at 64.</dd>
756 <p>More calling conventions can be added/defined on an as-needed basis, to
757 support Pascal conventions or any other well-known target-independent
762 <!-- ======================================================================= -->
764 <a name="visibility">Visibility Styles</a>
769 <p>All Global Variables and Functions have one of the following visibility
773 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
774 <dd>On targets that use the ELF object file format, default visibility means
775 that the declaration is visible to other modules and, in shared libraries,
776 means that the declared entity may be overridden. On Darwin, default
777 visibility means that the declaration is visible to other modules. Default
778 visibility corresponds to "external linkage" in the language.</dd>
780 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
781 <dd>Two declarations of an object with hidden visibility refer to the same
782 object if they are in the same shared object. Usually, hidden visibility
783 indicates that the symbol will not be placed into the dynamic symbol
784 table, so no other module (executable or shared library) can reference it
787 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
788 <dd>On ELF, protected visibility indicates that the symbol will be placed in
789 the dynamic symbol table, but that references within the defining module
790 will bind to the local symbol. That is, the symbol cannot be overridden by
796 <!-- ======================================================================= -->
798 <a name="namedtypes">Named Types</a>
803 <p>LLVM IR allows you to specify name aliases for certain types. This can make
804 it easier to read the IR and make the IR more condensed (particularly when
805 recursive types are involved). An example of a name specification is:</p>
807 <pre class="doc_code">
808 %mytype = type { %mytype*, i32 }
811 <p>You may give a name to any <a href="#typesystem">type</a> except
812 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
813 is expected with the syntax "%mytype".</p>
815 <p>Note that type names are aliases for the structural type that they indicate,
816 and that you can therefore specify multiple names for the same type. This
817 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
818 uses structural typing, the name is not part of the type. When printing out
819 LLVM IR, the printer will pick <em>one name</em> to render all types of a
820 particular shape. This means that if you have code where two different
821 source types end up having the same LLVM type, that the dumper will sometimes
822 print the "wrong" or unexpected type. This is an important design point and
823 isn't going to change.</p>
827 <!-- ======================================================================= -->
829 <a name="globalvars">Global Variables</a>
834 <p>Global variables define regions of memory allocated at compilation time
835 instead of run-time. Global variables may optionally be initialized, may
836 have an explicit section to be placed in, and may have an optional explicit
837 alignment specified. A variable may be defined as "thread_local", which
838 means that it will not be shared by threads (each thread will have a
839 separated copy of the variable). A variable may be defined as a global
840 "constant," which indicates that the contents of the variable
841 will <b>never</b> be modified (enabling better optimization, allowing the
842 global data to be placed in the read-only section of an executable, etc).
843 Note that variables that need runtime initialization cannot be marked
844 "constant" as there is a store to the variable.</p>
846 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
847 constant, even if the final definition of the global is not. This capability
848 can be used to enable slightly better optimization of the program, but
849 requires the language definition to guarantee that optimizations based on the
850 'constantness' are valid for the translation units that do not include the
853 <p>As SSA values, global variables define pointer values that are in scope
854 (i.e. they dominate) all basic blocks in the program. Global variables
855 always define a pointer to their "content" type because they describe a
856 region of memory, and all memory objects in LLVM are accessed through
859 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
860 that the address is not significant, only the content. Constants marked
861 like this can be merged with other constants if they have the same
862 initializer. Note that a constant with significant address <em>can</em>
863 be merged with a <tt>unnamed_addr</tt> constant, the result being a
864 constant whose address is significant.</p>
866 <p>A global variable may be declared to reside in a target-specific numbered
867 address space. For targets that support them, address spaces may affect how
868 optimizations are performed and/or what target instructions are used to
869 access the variable. The default address space is zero. The address space
870 qualifier must precede any other attributes.</p>
872 <p>LLVM allows an explicit section to be specified for globals. If the target
873 supports it, it will emit globals to the section specified.</p>
875 <p>An explicit alignment may be specified for a global, which must be a power
876 of 2. If not present, or if the alignment is set to zero, the alignment of
877 the global is set by the target to whatever it feels convenient. If an
878 explicit alignment is specified, the global is forced to have exactly that
879 alignment. Targets and optimizers are not allowed to over-align the global
880 if the global has an assigned section. In this case, the extra alignment
881 could be observable: for example, code could assume that the globals are
882 densely packed in their section and try to iterate over them as an array,
883 alignment padding would break this iteration.</p>
885 <p>For example, the following defines a global in a numbered address space with
886 an initializer, section, and alignment:</p>
888 <pre class="doc_code">
889 @G = addrspace(5) constant float 1.0, section "foo", align 4
895 <!-- ======================================================================= -->
897 <a name="functionstructure">Functions</a>
902 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
903 optional <a href="#linkage">linkage type</a>, an optional
904 <a href="#visibility">visibility style</a>, an optional
905 <a href="#callingconv">calling convention</a>,
906 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
907 <a href="#paramattrs">parameter attribute</a> for the return type, a function
908 name, a (possibly empty) argument list (each with optional
909 <a href="#paramattrs">parameter attributes</a>), optional
910 <a href="#fnattrs">function attributes</a>, an optional section, an optional
911 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
912 curly brace, a list of basic blocks, and a closing curly brace.</p>
914 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
915 optional <a href="#linkage">linkage type</a>, an optional
916 <a href="#visibility">visibility style</a>, an optional
917 <a href="#callingconv">calling convention</a>,
918 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
919 <a href="#paramattrs">parameter attribute</a> for the return type, a function
920 name, a possibly empty list of arguments, an optional alignment, and an
921 optional <a href="#gc">garbage collector name</a>.</p>
923 <p>A function definition contains a list of basic blocks, forming the CFG
924 (Control Flow Graph) for the function. Each basic block may optionally start
925 with a label (giving the basic block a symbol table entry), contains a list
926 of instructions, and ends with a <a href="#terminators">terminator</a>
927 instruction (such as a branch or function return).</p>
929 <p>The first basic block in a function is special in two ways: it is immediately
930 executed on entrance to the function, and it is not allowed to have
931 predecessor basic blocks (i.e. there can not be any branches to the entry
932 block of a function). Because the block can have no predecessors, it also
933 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
935 <p>LLVM allows an explicit section to be specified for functions. If the target
936 supports it, it will emit functions to the section specified.</p>
938 <p>An explicit alignment may be specified for a function. If not present, or if
939 the alignment is set to zero, the alignment of the function is set by the
940 target to whatever it feels convenient. If an explicit alignment is
941 specified, the function is forced to have at least that much alignment. All
942 alignments must be a power of 2.</p>
944 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
945 be significant and two identical functions can be merged</p>.
948 <pre class="doc_code">
949 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
950 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
951 <ResultType> @<FunctionName> ([argument list])
952 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
953 [<a href="#gc">gc</a>] { ... }
958 <!-- ======================================================================= -->
960 <a name="aliasstructure">Aliases</a>
965 <p>Aliases act as "second name" for the aliasee value (which can be either
966 function, global variable, another alias or bitcast of global value). Aliases
967 may have an optional <a href="#linkage">linkage type</a>, and an
968 optional <a href="#visibility">visibility style</a>.</p>
971 <pre class="doc_code">
972 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
977 <!-- ======================================================================= -->
979 <a name="namedmetadatastructure">Named Metadata</a>
984 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
985 nodes</a> (but not metadata strings) are the only valid operands for
986 a named metadata.</p>
989 <pre class="doc_code">
990 ; Some unnamed metadata nodes, which are referenced by the named metadata.
991 !0 = metadata !{metadata !"zero"}
992 !1 = metadata !{metadata !"one"}
993 !2 = metadata !{metadata !"two"}
995 !name = !{!0, !1, !2}
1000 <!-- ======================================================================= -->
1002 <a name="paramattrs">Parameter Attributes</a>
1007 <p>The return type and each parameter of a function type may have a set of
1008 <i>parameter attributes</i> associated with them. Parameter attributes are
1009 used to communicate additional information about the result or parameters of
1010 a function. Parameter attributes are considered to be part of the function,
1011 not of the function type, so functions with different parameter attributes
1012 can have the same function type.</p>
1014 <p>Parameter attributes are simple keywords that follow the type specified. If
1015 multiple parameter attributes are needed, they are space separated. For
1018 <pre class="doc_code">
1019 declare i32 @printf(i8* noalias nocapture, ...)
1020 declare i32 @atoi(i8 zeroext)
1021 declare signext i8 @returns_signed_char()
1024 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1025 <tt>readonly</tt>) come immediately after the argument list.</p>
1027 <p>Currently, only the following parameter attributes are defined:</p>
1030 <dt><tt><b>zeroext</b></tt></dt>
1031 <dd>This indicates to the code generator that the parameter or return value
1032 should be zero-extended to the extent required by the target's ABI (which
1033 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1034 parameter) or the callee (for a return value).</dd>
1036 <dt><tt><b>signext</b></tt></dt>
1037 <dd>This indicates to the code generator that the parameter or return value
1038 should be sign-extended to the extent required by the target's ABI (which
1039 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1042 <dt><tt><b>inreg</b></tt></dt>
1043 <dd>This indicates that this parameter or return value should be treated in a
1044 special target-dependent fashion during while emitting code for a function
1045 call or return (usually, by putting it in a register as opposed to memory,
1046 though some targets use it to distinguish between two different kinds of
1047 registers). Use of this attribute is target-specific.</dd>
1049 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1050 <dd><p>This indicates that the pointer parameter should really be passed by
1051 value to the function. The attribute implies that a hidden copy of the
1053 is made between the caller and the callee, so the callee is unable to
1054 modify the value in the callee. This attribute is only valid on LLVM
1055 pointer arguments. It is generally used to pass structs and arrays by
1056 value, but is also valid on pointers to scalars. The copy is considered
1057 to belong to the caller not the callee (for example,
1058 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1059 <tt>byval</tt> parameters). This is not a valid attribute for return
1062 <p>The byval attribute also supports specifying an alignment with
1063 the align attribute. It indicates the alignment of the stack slot to
1064 form and the known alignment of the pointer specified to the call site. If
1065 the alignment is not specified, then the code generator makes a
1066 target-specific assumption.</p></dd>
1068 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1069 <dd>This indicates that the pointer parameter specifies the address of a
1070 structure that is the return value of the function in the source program.
1071 This pointer must be guaranteed by the caller to be valid: loads and
1072 stores to the structure may be assumed by the callee to not to trap. This
1073 may only be applied to the first parameter. This is not a valid attribute
1074 for return values. </dd>
1076 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1077 <dd>This indicates that pointer values
1078 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1079 value do not alias pointer values which are not <i>based</i> on it,
1080 ignoring certain "irrelevant" dependencies.
1081 For a call to the parent function, dependencies between memory
1082 references from before or after the call and from those during the call
1083 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1084 return value used in that call.
1085 The caller shares the responsibility with the callee for ensuring that
1086 these requirements are met.
1087 For further details, please see the discussion of the NoAlias response in
1088 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1090 Note that this definition of <tt>noalias</tt> is intentionally
1091 similar to the definition of <tt>restrict</tt> in C99 for function
1092 arguments, though it is slightly weaker.
1094 For function return values, C99's <tt>restrict</tt> is not meaningful,
1095 while LLVM's <tt>noalias</tt> is.
1098 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1099 <dd>This indicates that the callee does not make any copies of the pointer
1100 that outlive the callee itself. This is not a valid attribute for return
1103 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1104 <dd>This indicates that the pointer parameter can be excised using the
1105 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1106 attribute for return values.</dd>
1111 <!-- ======================================================================= -->
1113 <a name="gc">Garbage Collector Names</a>
1118 <p>Each function may specify a garbage collector name, which is simply a
1121 <pre class="doc_code">
1122 define void @f() gc "name" { ... }
1125 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1126 collector which will cause the compiler to alter its output in order to
1127 support the named garbage collection algorithm.</p>
1131 <!-- ======================================================================= -->
1133 <a name="fnattrs">Function Attributes</a>
1138 <p>Function attributes are set to communicate additional information about a
1139 function. Function attributes are considered to be part of the function, not
1140 of the function type, so functions with different parameter attributes can
1141 have the same function type.</p>
1143 <p>Function attributes are simple keywords that follow the type specified. If
1144 multiple attributes are needed, they are space separated. For example:</p>
1146 <pre class="doc_code">
1147 define void @f() noinline { ... }
1148 define void @f() alwaysinline { ... }
1149 define void @f() alwaysinline optsize { ... }
1150 define void @f() optsize { ... }
1154 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1155 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1156 the backend should forcibly align the stack pointer. Specify the
1157 desired alignment, which must be a power of two, in parentheses.
1159 <dt><tt><b>alwaysinline</b></tt></dt>
1160 <dd>This attribute indicates that the inliner should attempt to inline this
1161 function into callers whenever possible, ignoring any active inlining size
1162 threshold for this caller.</dd>
1164 <dt><tt><b>hotpatch</b></tt></dt>
1165 <dd>This attribute indicates that the function should be 'hotpatchable',
1166 meaning the function can be patched and/or hooked even while it is
1167 loaded into memory. On x86, the function prologue will be preceded
1168 by six bytes of padding and will begin with a two-byte instruction.
1169 Most of the functions in the Windows system DLLs in Windows XP SP2 or
1170 higher were compiled in this fashion.</dd>
1172 <dt><tt><b>nonlazybind</b></tt></dt>
1173 <dd>This attribute suppresses lazy symbol binding for the function. This
1174 may make calls to the function faster, at the cost of extra program
1175 startup time if the function is not called during program startup.</dd>
1177 <dt><tt><b>inlinehint</b></tt></dt>
1178 <dd>This attribute indicates that the source code contained a hint that inlining
1179 this function is desirable (such as the "inline" keyword in C/C++). It
1180 is just a hint; it imposes no requirements on the inliner.</dd>
1182 <dt><tt><b>naked</b></tt></dt>
1183 <dd>This attribute disables prologue / epilogue emission for the function.
1184 This can have very system-specific consequences.</dd>
1186 <dt><tt><b>noimplicitfloat</b></tt></dt>
1187 <dd>This attributes disables implicit floating point instructions.</dd>
1189 <dt><tt><b>noinline</b></tt></dt>
1190 <dd>This attribute indicates that the inliner should never inline this
1191 function in any situation. This attribute may not be used together with
1192 the <tt>alwaysinline</tt> attribute.</dd>
1194 <dt><tt><b>noredzone</b></tt></dt>
1195 <dd>This attribute indicates that the code generator should not use a red
1196 zone, even if the target-specific ABI normally permits it.</dd>
1198 <dt><tt><b>noreturn</b></tt></dt>
1199 <dd>This function attribute indicates that the function never returns
1200 normally. This produces undefined behavior at runtime if the function
1201 ever does dynamically return.</dd>
1203 <dt><tt><b>nounwind</b></tt></dt>
1204 <dd>This function attribute indicates that the function never returns with an
1205 unwind or exceptional control flow. If the function does unwind, its
1206 runtime behavior is undefined.</dd>
1208 <dt><tt><b>optsize</b></tt></dt>
1209 <dd>This attribute suggests that optimization passes and code generator passes
1210 make choices that keep the code size of this function low, and otherwise
1211 do optimizations specifically to reduce code size.</dd>
1213 <dt><tt><b>readnone</b></tt></dt>
1214 <dd>This attribute indicates that the function computes its result (or decides
1215 to unwind an exception) based strictly on its arguments, without
1216 dereferencing any pointer arguments or otherwise accessing any mutable
1217 state (e.g. memory, control registers, etc) visible to caller functions.
1218 It does not write through any pointer arguments
1219 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1220 changes any state visible to callers. This means that it cannot unwind
1221 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1222 could use the <tt>unwind</tt> instruction.</dd>
1224 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1225 <dd>This attribute indicates that the function does not write through any
1226 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1227 arguments) or otherwise modify any state (e.g. memory, control registers,
1228 etc) visible to caller functions. It may dereference pointer arguments
1229 and read state that may be set in the caller. A readonly function always
1230 returns the same value (or unwinds an exception identically) when called
1231 with the same set of arguments and global state. It cannot unwind an
1232 exception by calling the <tt>C++</tt> exception throwing methods, but may
1233 use the <tt>unwind</tt> instruction.</dd>
1235 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1236 <dd>This attribute indicates that the function should emit a stack smashing
1237 protector. It is in the form of a "canary"—a random value placed on
1238 the stack before the local variables that's checked upon return from the
1239 function to see if it has been overwritten. A heuristic is used to
1240 determine if a function needs stack protectors or not.<br>
1242 If a function that has an <tt>ssp</tt> attribute is inlined into a
1243 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1244 function will have an <tt>ssp</tt> attribute.</dd>
1246 <dt><tt><b>sspreq</b></tt></dt>
1247 <dd>This attribute indicates that the function should <em>always</em> emit a
1248 stack smashing protector. This overrides
1249 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1251 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1252 function that doesn't have an <tt>sspreq</tt> attribute or which has
1253 an <tt>ssp</tt> attribute, then the resulting function will have
1254 an <tt>sspreq</tt> attribute.</dd>
1256 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1257 <dd>This attribute indicates that the ABI being targeted requires that
1258 an unwind table entry be produce for this function even if we can
1259 show that no exceptions passes by it. This is normally the case for
1260 the ELF x86-64 abi, but it can be disabled for some compilation
1267 <!-- ======================================================================= -->
1269 <a name="moduleasm">Module-Level Inline Assembly</a>
1274 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1275 the GCC "file scope inline asm" blocks. These blocks are internally
1276 concatenated by LLVM and treated as a single unit, but may be separated in
1277 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1279 <pre class="doc_code">
1280 module asm "inline asm code goes here"
1281 module asm "more can go here"
1284 <p>The strings can contain any character by escaping non-printable characters.
1285 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1288 <p>The inline asm code is simply printed to the machine code .s file when
1289 assembly code is generated.</p>
1293 <!-- ======================================================================= -->
1295 <a name="datalayout">Data Layout</a>
1300 <p>A module may specify a target specific data layout string that specifies how
1301 data is to be laid out in memory. The syntax for the data layout is
1304 <pre class="doc_code">
1305 target datalayout = "<i>layout specification</i>"
1308 <p>The <i>layout specification</i> consists of a list of specifications
1309 separated by the minus sign character ('-'). Each specification starts with
1310 a letter and may include other information after the letter to define some
1311 aspect of the data layout. The specifications accepted are as follows:</p>
1315 <dd>Specifies that the target lays out data in big-endian form. That is, the
1316 bits with the most significance have the lowest address location.</dd>
1319 <dd>Specifies that the target lays out data in little-endian form. That is,
1320 the bits with the least significance have the lowest address
1323 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1324 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1325 <i>preferred</i> alignments. All sizes are in bits. Specifying
1326 the <i>pref</i> alignment is optional. If omitted, the
1327 preceding <tt>:</tt> should be omitted too.</dd>
1329 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1330 <dd>This specifies the alignment for an integer type of a given bit
1331 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1333 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1334 <dd>This specifies the alignment for a vector type of a given bit
1337 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1338 <dd>This specifies the alignment for a floating point type of a given bit
1339 <i>size</i>. Only values of <i>size</i> that are supported by the target
1340 will work. 32 (float) and 64 (double) are supported on all targets;
1341 80 or 128 (different flavors of long double) are also supported on some
1344 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1345 <dd>This specifies the alignment for an aggregate type of a given bit
1348 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1349 <dd>This specifies the alignment for a stack object of a given bit
1352 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1353 <dd>This specifies a set of native integer widths for the target CPU
1354 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1355 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1356 this set are considered to support most general arithmetic
1357 operations efficiently.</dd>
1360 <p>When constructing the data layout for a given target, LLVM starts with a
1361 default set of specifications which are then (possibly) overridden by the
1362 specifications in the <tt>datalayout</tt> keyword. The default specifications
1363 are given in this list:</p>
1366 <li><tt>E</tt> - big endian</li>
1367 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1368 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1369 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1370 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1371 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1372 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1373 alignment of 64-bits</li>
1374 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1375 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1376 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1377 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1378 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1379 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1382 <p>When LLVM is determining the alignment for a given type, it uses the
1383 following rules:</p>
1386 <li>If the type sought is an exact match for one of the specifications, that
1387 specification is used.</li>
1389 <li>If no match is found, and the type sought is an integer type, then the
1390 smallest integer type that is larger than the bitwidth of the sought type
1391 is used. If none of the specifications are larger than the bitwidth then
1392 the the largest integer type is used. For example, given the default
1393 specifications above, the i7 type will use the alignment of i8 (next
1394 largest) while both i65 and i256 will use the alignment of i64 (largest
1397 <li>If no match is found, and the type sought is a vector type, then the
1398 largest vector type that is smaller than the sought vector type will be
1399 used as a fall back. This happens because <128 x double> can be
1400 implemented in terms of 64 <2 x double>, for example.</li>
1405 <!-- ======================================================================= -->
1407 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1412 <p>Any memory access must be done through a pointer value associated
1413 with an address range of the memory access, otherwise the behavior
1414 is undefined. Pointer values are associated with address ranges
1415 according to the following rules:</p>
1418 <li>A pointer value is associated with the addresses associated with
1419 any value it is <i>based</i> on.
1420 <li>An address of a global variable is associated with the address
1421 range of the variable's storage.</li>
1422 <li>The result value of an allocation instruction is associated with
1423 the address range of the allocated storage.</li>
1424 <li>A null pointer in the default address-space is associated with
1426 <li>An integer constant other than zero or a pointer value returned
1427 from a function not defined within LLVM may be associated with address
1428 ranges allocated through mechanisms other than those provided by
1429 LLVM. Such ranges shall not overlap with any ranges of addresses
1430 allocated by mechanisms provided by LLVM.</li>
1433 <p>A pointer value is <i>based</i> on another pointer value according
1434 to the following rules:</p>
1437 <li>A pointer value formed from a
1438 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1439 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1440 <li>The result value of a
1441 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1442 of the <tt>bitcast</tt>.</li>
1443 <li>A pointer value formed by an
1444 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1445 pointer values that contribute (directly or indirectly) to the
1446 computation of the pointer's value.</li>
1447 <li>The "<i>based</i> on" relationship is transitive.</li>
1450 <p>Note that this definition of <i>"based"</i> is intentionally
1451 similar to the definition of <i>"based"</i> in C99, though it is
1452 slightly weaker.</p>
1454 <p>LLVM IR does not associate types with memory. The result type of a
1455 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1456 alignment of the memory from which to load, as well as the
1457 interpretation of the value. The first operand type of a
1458 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1459 and alignment of the store.</p>
1461 <p>Consequently, type-based alias analysis, aka TBAA, aka
1462 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1463 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1464 additional information which specialized optimization passes may use
1465 to implement type-based alias analysis.</p>
1469 <!-- ======================================================================= -->
1471 <a name="volatile">Volatile Memory Accesses</a>
1476 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1477 href="#i_store"><tt>store</tt></a>s, and <a
1478 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1479 The optimizers must not change the number of volatile operations or change their
1480 order of execution relative to other volatile operations. The optimizers
1481 <i>may</i> change the order of volatile operations relative to non-volatile
1482 operations. This is not Java's "volatile" and has no cross-thread
1483 synchronization behavior.</p>
1487 <!-- ======================================================================= -->
1489 <a name="memmodel">Memory Model for Concurrent Operations</a>
1494 <p>The LLVM IR does not define any way to start parallel threads of execution
1495 or to register signal handlers. Nonetheless, there are platform-specific
1496 ways to create them, and we define LLVM IR's behavior in their presence. This
1497 model is inspired by the C++0x memory model.</p>
1499 <p>We define a <i>happens-before</i> partial order as the least partial order
1502 <li>Is a superset of single-thread program order, and</li>
1503 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1504 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1505 by platform-specific techniques, like pthread locks, thread
1506 creation, thread joining, etc., and by atomic instructions.
1507 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1511 <p>Note that program order does not introduce <i>happens-before</i> edges
1512 between a thread and signals executing inside that thread.</p>
1514 <p>Every (defined) read operation (load instructions, memcpy, atomic
1515 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1516 (defined) write operations (store instructions, atomic
1517 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1518 initialized globals are considered to have a write of the initializer which is
1519 atomic and happens before any other read or write of the memory in question.
1520 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1521 any write to the same byte, except:</p>
1524 <li>If <var>write<sub>1</sub></var> happens before
1525 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1526 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1527 does not see <var>write<sub>1</sub></var>.
1528 <li>If <var>R<sub>byte</sub></var> happens before <var>write<sub>3</var>,
1529 then <var>R<sub>byte</sub></var> does not see
1530 <var>write<sub>3</sub></var>.
1533 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1535 <li>If there is no write to the same byte that happens before
1536 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1537 <tt>undef</tt> for that byte.
1538 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1539 <var>R<sub>byte</sub></var> returns the value written by that
1541 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1542 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1543 values written. See the <a href="#ordering">Atomic Memory Ordering
1544 Constraints</a> section for additional constraints on how the choice
1546 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1549 <p><var>R</var> returns the value composed of the series of bytes it read.
1550 This implies that some bytes within the value may be <tt>undef</tt>
1551 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1552 defines the semantics of the operation; it doesn't mean that targets will
1553 emit more than one instruction to read the series of bytes.</p>
1555 <p>Note that in cases where none of the atomic intrinsics are used, this model
1556 places only one restriction on IR transformations on top of what is required
1557 for single-threaded execution: introducing a store to a byte which might not
1558 otherwise be stored to can introduce undefined behavior. (Specifically, in
1559 the case where another thread might write to and read from an address,
1560 introducing a store can change a load that may see exactly one write into
1561 a load that may see multiple writes.)</p>
1563 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1564 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1565 none of the backends currently in the tree fall into this category; however,
1566 there might be targets which care. If there are, we want a paragraph
1569 Targets may specify that stores narrower than a certain width are not
1570 available; on such a target, for the purposes of this model, treat any
1571 non-atomic write with an alignment or width less than the minimum width
1572 as if it writes to the relevant surrounding bytes.
1577 <!-- ======================================================================= -->
1578 <div class="doc_subsection">
1579 <a name="ordering">Atomic Memory Ordering Constraints</a>
1582 <div class="doc_text">
1584 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1585 <a href="#i_atomicrmw"><code>atomicrmw</code></a>, and
1586 <a href="#i_fence"><code>fence</code></a>) take an ordering parameter
1587 that determines which other atomic instructions on the same address they
1588 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1589 but are somewhat more colloquial. If these descriptions aren't precise enough,
1590 check those specs. <a href="#i_fence"><code>fence</code></a> instructions
1591 treat these orderings somewhat differently since they don't take an address.
1592 See that instruction's documentation for details.</p>
1594 <!-- FIXME Note atomic load+store here once those get added. -->
1597 <!-- FIXME: unordered is intended to be used for atomic load and store;
1598 it isn't allowed for any instruction yet. -->
1599 <dt><code>unordered</code></dt>
1600 <dd>The set of values that can be read is governed by the happens-before
1601 partial order. A value cannot be read unless some operation wrote it.
1602 This is intended to provide a guarantee strong enough to model Java's
1603 non-volatile shared variables. This ordering cannot be specified for
1604 read-modify-write operations; it is not strong enough to make them atomic
1605 in any interesting way.</dd>
1606 <dt><code>monotonic</code></dt>
1607 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1608 total order for modifications by <code>monotonic</code> operations on each
1609 address. All modification orders must be compatible with the happens-before
1610 order. There is no guarantee that the modification orders can be combined to
1611 a global total order for the whole program (and this often will not be
1612 possible). The read in an atomic read-modify-write operation
1613 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1614 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1615 reads the value in the modification order immediately before the value it
1616 writes. If one atomic read happens before another atomic read of the same
1617 address, the later read must see the same value or a later value in the
1618 address's modification order. This disallows reordering of
1619 <code>monotonic</code> (or stronger) operations on the same address. If an
1620 address is written <code>monotonic</code>ally by one thread, and other threads
1621 <code>monotonic</code>ally read that address repeatedly, the other threads must
1622 eventually see the write. This is intended to model C++'s relaxed atomic
1624 <dt><code>acquire</code></dt>
1625 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1626 reads a value written by a <code>release</code> atomic operation, it
1627 <i>synchronizes-with</i> that operation.</dd>
1628 <dt><code>release</code></dt>
1629 <dd>In addition to the guarantees of <code>monotonic</code>,
1630 a <i>synchronizes-with</i> edge may be formed by an <code>acquire</code>
1632 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1633 <code>acquire</code> and <code>release</code> operation on its address.</dd>
1634 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1635 <dd>In addition to the guarantees of <code>acq_rel</code>
1636 (<code>acquire</code> for an operation which only reads, <code>release</code>
1637 for an operation which only writes), there is a global total order on all
1638 sequentially-consistent operations on all addresses, which is consistent with
1639 the <i>happens-before</i> partial order and with the modification orders of
1640 all the affected addresses. Each sequentially-consistent read sees the last
1641 preceding write to the same address in this global order. This is intended
1642 to model C++'s sequentially-consistent atomic variables and Java's volatile
1643 shared variables.</dd>
1646 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1647 it only <i>synchronizes with</i> or participates in modification and seq_cst
1648 total orderings with other operations running in the same thread (for example,
1649 in signal handlers).</p>
1655 <!-- *********************************************************************** -->
1656 <h2><a name="typesystem">Type System</a></h2>
1657 <!-- *********************************************************************** -->
1661 <p>The LLVM type system is one of the most important features of the
1662 intermediate representation. Being typed enables a number of optimizations
1663 to be performed on the intermediate representation directly, without having
1664 to do extra analyses on the side before the transformation. A strong type
1665 system makes it easier to read the generated code and enables novel analyses
1666 and transformations that are not feasible to perform on normal three address
1667 code representations.</p>
1669 <!-- ======================================================================= -->
1671 <a name="t_classifications">Type Classifications</a>
1676 <p>The types fall into a few useful classifications:</p>
1678 <table border="1" cellspacing="0" cellpadding="4">
1680 <tr><th>Classification</th><th>Types</th></tr>
1682 <td><a href="#t_integer">integer</a></td>
1683 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1686 <td><a href="#t_floating">floating point</a></td>
1687 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1690 <td><a name="t_firstclass">first class</a></td>
1691 <td><a href="#t_integer">integer</a>,
1692 <a href="#t_floating">floating point</a>,
1693 <a href="#t_pointer">pointer</a>,
1694 <a href="#t_vector">vector</a>,
1695 <a href="#t_struct">structure</a>,
1696 <a href="#t_array">array</a>,
1697 <a href="#t_label">label</a>,
1698 <a href="#t_metadata">metadata</a>.
1702 <td><a href="#t_primitive">primitive</a></td>
1703 <td><a href="#t_label">label</a>,
1704 <a href="#t_void">void</a>,
1705 <a href="#t_integer">integer</a>,
1706 <a href="#t_floating">floating point</a>,
1707 <a href="#t_x86mmx">x86mmx</a>,
1708 <a href="#t_metadata">metadata</a>.</td>
1711 <td><a href="#t_derived">derived</a></td>
1712 <td><a href="#t_array">array</a>,
1713 <a href="#t_function">function</a>,
1714 <a href="#t_pointer">pointer</a>,
1715 <a href="#t_struct">structure</a>,
1716 <a href="#t_vector">vector</a>,
1717 <a href="#t_opaque">opaque</a>.
1723 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1724 important. Values of these types are the only ones which can be produced by
1729 <!-- ======================================================================= -->
1731 <a name="t_primitive">Primitive Types</a>
1736 <p>The primitive types are the fundamental building blocks of the LLVM
1739 <!-- _______________________________________________________________________ -->
1741 <a name="t_integer">Integer Type</a>
1747 <p>The integer type is a very simple type that simply specifies an arbitrary
1748 bit width for the integer type desired. Any bit width from 1 bit to
1749 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1756 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1760 <table class="layout">
1762 <td class="left"><tt>i1</tt></td>
1763 <td class="left">a single-bit integer.</td>
1766 <td class="left"><tt>i32</tt></td>
1767 <td class="left">a 32-bit integer.</td>
1770 <td class="left"><tt>i1942652</tt></td>
1771 <td class="left">a really big integer of over 1 million bits.</td>
1777 <!-- _______________________________________________________________________ -->
1779 <a name="t_floating">Floating Point Types</a>
1786 <tr><th>Type</th><th>Description</th></tr>
1787 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1788 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1789 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1790 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1791 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1797 <!-- _______________________________________________________________________ -->
1799 <a name="t_x86mmx">X86mmx Type</a>
1805 <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>
1814 <!-- _______________________________________________________________________ -->
1816 <a name="t_void">Void Type</a>
1822 <p>The void type does not represent any value and has no size.</p>
1831 <!-- _______________________________________________________________________ -->
1833 <a name="t_label">Label Type</a>
1839 <p>The label type represents code labels.</p>
1848 <!-- _______________________________________________________________________ -->
1850 <a name="t_metadata">Metadata Type</a>
1856 <p>The metadata type represents embedded metadata. No derived types may be
1857 created from metadata except for <a href="#t_function">function</a>
1869 <!-- ======================================================================= -->
1871 <a name="t_derived">Derived Types</a>
1876 <p>The real power in LLVM comes from the derived types in the system. This is
1877 what allows a programmer to represent arrays, functions, pointers, and other
1878 useful types. Each of these types contain one or more element types which
1879 may be a primitive type, or another derived type. For example, it is
1880 possible to have a two dimensional array, using an array as the element type
1881 of another array.</p>
1886 <!-- _______________________________________________________________________ -->
1888 <a name="t_aggregate">Aggregate Types</a>
1893 <p>Aggregate Types are a subset of derived types that can contain multiple
1894 member types. <a href="#t_array">Arrays</a>,
1895 <a href="#t_struct">structs</a>, and <a href="#t_vector">vectors</a> are
1896 aggregate types.</p>
1900 <!-- _______________________________________________________________________ -->
1902 <a name="t_array">Array Type</a>
1908 <p>The array type is a very simple derived type that arranges elements
1909 sequentially in memory. The array type requires a size (number of elements)
1910 and an underlying data type.</p>
1914 [<# elements> x <elementtype>]
1917 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1918 be any type with a size.</p>
1921 <table class="layout">
1923 <td class="left"><tt>[40 x i32]</tt></td>
1924 <td class="left">Array of 40 32-bit integer values.</td>
1927 <td class="left"><tt>[41 x i32]</tt></td>
1928 <td class="left">Array of 41 32-bit integer values.</td>
1931 <td class="left"><tt>[4 x i8]</tt></td>
1932 <td class="left">Array of 4 8-bit integer values.</td>
1935 <p>Here are some examples of multidimensional arrays:</p>
1936 <table class="layout">
1938 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1939 <td class="left">3x4 array of 32-bit integer values.</td>
1942 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1943 <td class="left">12x10 array of single precision floating point values.</td>
1946 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1947 <td class="left">2x3x4 array of 16-bit integer values.</td>
1951 <p>There is no restriction on indexing beyond the end of the array implied by
1952 a static type (though there are restrictions on indexing beyond the bounds
1953 of an allocated object in some cases). This means that single-dimension
1954 'variable sized array' addressing can be implemented in LLVM with a zero
1955 length array type. An implementation of 'pascal style arrays' in LLVM could
1956 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1960 <!-- _______________________________________________________________________ -->
1962 <a name="t_function">Function Type</a>
1968 <p>The function type can be thought of as a function signature. It consists of
1969 a return type and a list of formal parameter types. The return type of a
1970 function type is a first class type or a void type.</p>
1974 <returntype> (<parameter list>)
1977 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1978 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1979 which indicates that the function takes a variable number of arguments.
1980 Variable argument functions can access their arguments with
1981 the <a href="#int_varargs">variable argument handling intrinsic</a>
1982 functions. '<tt><returntype></tt>' is any type except
1983 <a href="#t_label">label</a>.</p>
1986 <table class="layout">
1988 <td class="left"><tt>i32 (i32)</tt></td>
1989 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1991 </tr><tr class="layout">
1992 <td class="left"><tt>float (i16, i32 *) *
1994 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1995 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
1996 returning <tt>float</tt>.
1998 </tr><tr class="layout">
1999 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2000 <td class="left">A vararg function that takes at least one
2001 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2002 which returns an integer. This is the signature for <tt>printf</tt> in
2005 </tr><tr class="layout">
2006 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2007 <td class="left">A function taking an <tt>i32</tt>, returning a
2008 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2015 <!-- _______________________________________________________________________ -->
2017 <a name="t_struct">Structure Type</a>
2023 <p>The structure type is used to represent a collection of data members together
2024 in memory. The elements of a structure may be any type that has a size.</p>
2026 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2027 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2028 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2029 Structures in registers are accessed using the
2030 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2031 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2033 <p>Structures may optionally be "packed" structures, which indicate that the
2034 alignment of the struct is one byte, and that there is no padding between
2035 the elements. In non-packed structs, padding between field types is defined
2036 by the target data string to match the underlying processor.</p>
2038 <p>Structures can either be "anonymous" or "named". An anonymous structure is
2039 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) and a named types
2040 are always defined at the top level with a name. Anonmyous types are uniqued
2041 by their contents and can never be recursive since there is no way to write
2042 one. Named types can be recursive.
2047 %T1 = type { <type list> } <i>; Named normal struct type</i>
2048 %T2 = type <{ <type list> }> <i>; Named packed struct type</i>
2052 <table class="layout">
2054 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2055 <td class="left">A triple of three <tt>i32</tt> values</td>
2058 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2059 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2060 second element is a <a href="#t_pointer">pointer</a> to a
2061 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2062 an <tt>i32</tt>.</td>
2065 <td class="left"><tt><{ i8, i32 }></tt></td>
2066 <td class="left">A packed struct known to be 5 bytes in size.</td>
2072 <!-- _______________________________________________________________________ -->
2074 <a name="t_opaque">Opaque Structure Types</a>
2080 <p>Opaque structure types are used to represent named structure types that do
2081 not have a body specified. This corresponds (for example) to the C notion of
2082 a forward declared structure.</p>
2091 <table class="layout">
2093 <td class="left"><tt>opaque</tt></td>
2094 <td class="left">An opaque type.</td>
2102 <!-- _______________________________________________________________________ -->
2104 <a name="t_pointer">Pointer Type</a>
2110 <p>The pointer type is used to specify memory locations.
2111 Pointers are commonly used to reference objects in memory.</p>
2113 <p>Pointer types may have an optional address space attribute defining the
2114 numbered address space where the pointed-to object resides. The default
2115 address space is number zero. The semantics of non-zero address
2116 spaces are target-specific.</p>
2118 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2119 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2127 <table class="layout">
2129 <td class="left"><tt>[4 x i32]*</tt></td>
2130 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2131 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2134 <td class="left"><tt>i32 (i32*) *</tt></td>
2135 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2136 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2140 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2141 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2142 that resides in address space #5.</td>
2148 <!-- _______________________________________________________________________ -->
2150 <a name="t_vector">Vector Type</a>
2156 <p>A vector type is a simple derived type that represents a vector of elements.
2157 Vector types are used when multiple primitive data are operated in parallel
2158 using a single instruction (SIMD). A vector type requires a size (number of
2159 elements) and an underlying primitive data type. Vector types are considered
2160 <a href="#t_firstclass">first class</a>.</p>
2164 < <# elements> x <elementtype> >
2167 <p>The number of elements is a constant integer value larger than 0; elementtype
2168 may be any integer or floating point type. Vectors of size zero are not
2169 allowed, and pointers are not allowed as the element type.</p>
2172 <table class="layout">
2174 <td class="left"><tt><4 x i32></tt></td>
2175 <td class="left">Vector of 4 32-bit integer values.</td>
2178 <td class="left"><tt><8 x float></tt></td>
2179 <td class="left">Vector of 8 32-bit floating-point values.</td>
2182 <td class="left"><tt><2 x i64></tt></td>
2183 <td class="left">Vector of 2 64-bit integer values.</td>
2189 <!-- *********************************************************************** -->
2190 <h2><a name="constants">Constants</a></h2>
2191 <!-- *********************************************************************** -->
2195 <p>LLVM has several different basic types of constants. This section describes
2196 them all and their syntax.</p>
2198 <!-- ======================================================================= -->
2200 <a name="simpleconstants">Simple Constants</a>
2206 <dt><b>Boolean constants</b></dt>
2207 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2208 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2210 <dt><b>Integer constants</b></dt>
2211 <dd>Standard integers (such as '4') are constants of
2212 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2213 with integer types.</dd>
2215 <dt><b>Floating point constants</b></dt>
2216 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2217 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2218 notation (see below). The assembler requires the exact decimal value of a
2219 floating-point constant. For example, the assembler accepts 1.25 but
2220 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2221 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2223 <dt><b>Null pointer constants</b></dt>
2224 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2225 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2228 <p>The one non-intuitive notation for constants is the hexadecimal form of
2229 floating point constants. For example, the form '<tt>double
2230 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2231 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2232 constants are required (and the only time that they are generated by the
2233 disassembler) is when a floating point constant must be emitted but it cannot
2234 be represented as a decimal floating point number in a reasonable number of
2235 digits. For example, NaN's, infinities, and other special values are
2236 represented in their IEEE hexadecimal format so that assembly and disassembly
2237 do not cause any bits to change in the constants.</p>
2239 <p>When using the hexadecimal form, constants of types float and double are
2240 represented using the 16-digit form shown above (which matches the IEEE754
2241 representation for double); float values must, however, be exactly
2242 representable as IEE754 single precision. Hexadecimal format is always used
2243 for long double, and there are three forms of long double. The 80-bit format
2244 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2245 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2246 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2247 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2248 currently supported target uses this format. Long doubles will only work if
2249 they match the long double format on your target. All hexadecimal formats
2250 are big-endian (sign bit at the left).</p>
2252 <p>There are no constants of type x86mmx.</p>
2255 <!-- ======================================================================= -->
2257 <a name="aggregateconstants"></a> <!-- old anchor -->
2258 <a name="complexconstants">Complex Constants</a>
2263 <p>Complex constants are a (potentially recursive) combination of simple
2264 constants and smaller complex constants.</p>
2267 <dt><b>Structure constants</b></dt>
2268 <dd>Structure constants are represented with notation similar to structure
2269 type definitions (a comma separated list of elements, surrounded by braces
2270 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2271 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2272 Structure constants must have <a href="#t_struct">structure type</a>, and
2273 the number and types of elements must match those specified by the
2276 <dt><b>Array constants</b></dt>
2277 <dd>Array constants are represented with notation similar to array type
2278 definitions (a comma separated list of elements, surrounded by square
2279 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2280 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2281 the number and types of elements must match those specified by the
2284 <dt><b>Vector constants</b></dt>
2285 <dd>Vector constants are represented with notation similar to vector type
2286 definitions (a comma separated list of elements, surrounded by
2287 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2288 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2289 have <a href="#t_vector">vector type</a>, and the number and types of
2290 elements must match those specified by the type.</dd>
2292 <dt><b>Zero initialization</b></dt>
2293 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2294 value to zero of <em>any</em> type, including scalar and
2295 <a href="#t_aggregate">aggregate</a> types.
2296 This is often used to avoid having to print large zero initializers
2297 (e.g. for large arrays) and is always exactly equivalent to using explicit
2298 zero initializers.</dd>
2300 <dt><b>Metadata node</b></dt>
2301 <dd>A metadata node is a structure-like constant with
2302 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2303 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2304 be interpreted as part of the instruction stream, metadata is a place to
2305 attach additional information such as debug info.</dd>
2310 <!-- ======================================================================= -->
2312 <a name="globalconstants">Global Variable and Function Addresses</a>
2317 <p>The addresses of <a href="#globalvars">global variables</a>
2318 and <a href="#functionstructure">functions</a> are always implicitly valid
2319 (link-time) constants. These constants are explicitly referenced when
2320 the <a href="#identifiers">identifier for the global</a> is used and always
2321 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2322 legal LLVM file:</p>
2324 <pre class="doc_code">
2327 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2332 <!-- ======================================================================= -->
2334 <a name="undefvalues">Undefined Values</a>
2339 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2340 indicates that the user of the value may receive an unspecified bit-pattern.
2341 Undefined values may be of any type (other than '<tt>label</tt>'
2342 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2344 <p>Undefined values are useful because they indicate to the compiler that the
2345 program is well defined no matter what value is used. This gives the
2346 compiler more freedom to optimize. Here are some examples of (potentially
2347 surprising) transformations that are valid (in pseudo IR):</p>
2350 <pre class="doc_code">
2360 <p>This is safe because all of the output bits are affected by the undef bits.
2361 Any output bit can have a zero or one depending on the input bits.</p>
2363 <pre class="doc_code">
2374 <p>These logical operations have bits that are not always affected by the input.
2375 For example, if <tt>%X</tt> has a zero bit, then the output of the
2376 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2377 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2378 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2379 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2380 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2381 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2382 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2384 <pre class="doc_code">
2385 %A = select undef, %X, %Y
2386 %B = select undef, 42, %Y
2387 %C = select %X, %Y, undef
2398 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2399 branch) conditions can go <em>either way</em>, but they have to come from one
2400 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2401 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2402 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2403 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2404 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2407 <pre class="doc_code">
2408 %A = xor undef, undef
2426 <p>This example points out that two '<tt>undef</tt>' operands are not
2427 necessarily the same. This can be surprising to people (and also matches C
2428 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2429 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2430 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2431 its value over its "live range". This is true because the variable doesn't
2432 actually <em>have a live range</em>. Instead, the value is logically read
2433 from arbitrary registers that happen to be around when needed, so the value
2434 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2435 need to have the same semantics or the core LLVM "replace all uses with"
2436 concept would not hold.</p>
2438 <pre class="doc_code">
2446 <p>These examples show the crucial difference between an <em>undefined
2447 value</em> and <em>undefined behavior</em>. An undefined value (like
2448 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2449 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2450 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2451 defined on SNaN's. However, in the second example, we can make a more
2452 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2453 arbitrary value, we are allowed to assume that it could be zero. Since a
2454 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2455 the operation does not execute at all. This allows us to delete the divide and
2456 all code after it. Because the undefined operation "can't happen", the
2457 optimizer can assume that it occurs in dead code.</p>
2459 <pre class="doc_code">
2460 a: store undef -> %X
2461 b: store %X -> undef
2467 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2468 undefined value can be assumed to not have any effect; we can assume that the
2469 value is overwritten with bits that happen to match what was already there.
2470 However, a store <em>to</em> an undefined location could clobber arbitrary
2471 memory, therefore, it has undefined behavior.</p>
2475 <!-- ======================================================================= -->
2477 <a name="trapvalues">Trap Values</a>
2482 <p>Trap values are similar to <a href="#undefvalues">undef values</a>, however
2483 instead of representing an unspecified bit pattern, they represent the
2484 fact that an instruction or constant expression which cannot evoke side
2485 effects has nevertheless detected a condition which results in undefined
2488 <p>There is currently no way of representing a trap value in the IR; they
2489 only exist when produced by operations such as
2490 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2492 <p>Trap value behavior is defined in terms of value <i>dependence</i>:</p>
2495 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2496 their operands.</li>
2498 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2499 to their dynamic predecessor basic block.</li>
2501 <li>Function arguments depend on the corresponding actual argument values in
2502 the dynamic callers of their functions.</li>
2504 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2505 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2506 control back to them.</li>
2508 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2509 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_unwind"><tt>unwind</tt></a>,
2510 or exception-throwing call instructions that dynamically transfer control
2513 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2514 referenced memory addresses, following the order in the IR
2515 (including loads and stores implied by intrinsics such as
2516 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2518 <!-- TODO: In the case of multiple threads, this only applies if the store
2519 "happens-before" the load or store. -->
2521 <!-- TODO: floating-point exception state -->
2523 <li>An instruction with externally visible side effects depends on the most
2524 recent preceding instruction with externally visible side effects, following
2525 the order in the IR. (This includes
2526 <a href="#volatile">volatile operations</a>.)</li>
2528 <li>An instruction <i>control-depends</i> on a
2529 <a href="#terminators">terminator instruction</a>
2530 if the terminator instruction has multiple successors and the instruction
2531 is always executed when control transfers to one of the successors, and
2532 may not be executed when control is transferred to another.</li>
2534 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2535 instruction if the set of instructions it otherwise depends on would be
2536 different if the terminator had transferred control to a different
2539 <li>Dependence is transitive.</li>
2543 <p>Whenever a trap value is generated, all values which depend on it evaluate
2544 to trap. If they have side effects, the evoke their side effects as if each
2545 operand with a trap value were undef. If they have externally-visible side
2546 effects, the behavior is undefined.</p>
2548 <p>Here are some examples:</p>
2550 <pre class="doc_code">
2552 %trap = sub nuw i32 0, 1 ; Results in a trap value.
2553 %still_trap = and i32 %trap, 0 ; Whereas (and i32 undef, 0) would return 0.
2554 %trap_yet_again = getelementptr i32* @h, i32 %still_trap
2555 store i32 0, i32* %trap_yet_again ; undefined behavior
2557 store i32 %trap, i32* @g ; Trap value conceptually stored to memory.
2558 %trap2 = load i32* @g ; Returns a trap value, not just undef.
2560 volatile store i32 %trap, i32* @g ; External observation; undefined behavior.
2562 %narrowaddr = bitcast i32* @g to i16*
2563 %wideaddr = bitcast i32* @g to i64*
2564 %trap3 = load i16* %narrowaddr ; Returns a trap value.
2565 %trap4 = load i64* %wideaddr ; Returns a trap value.
2567 %cmp = icmp slt i32 %trap, 0 ; Returns a trap value.
2568 br i1 %cmp, label %true, label %end ; Branch to either destination.
2571 volatile store i32 0, i32* @g ; This is control-dependent on %cmp, so
2572 ; it has undefined behavior.
2576 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2577 ; Both edges into this PHI are
2578 ; control-dependent on %cmp, so this
2579 ; always results in a trap value.
2581 volatile store i32 0, i32* @g ; This would depend on the store in %true
2582 ; if %cmp is true, or the store in %entry
2583 ; otherwise, so this is undefined behavior.
2585 br i1 %cmp, label %second_true, label %second_end
2586 ; The same branch again, but this time the
2587 ; true block doesn't have side effects.
2594 volatile store i32 0, i32* @g ; This time, the instruction always depends
2595 ; on the store in %end. Also, it is
2596 ; control-equivalent to %end, so this is
2597 ; well-defined (again, ignoring earlier
2598 ; undefined behavior in this example).
2603 <!-- ======================================================================= -->
2605 <a name="blockaddress">Addresses of Basic Blocks</a>
2610 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2612 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2613 basic block in the specified function, and always has an i8* type. Taking
2614 the address of the entry block is illegal.</p>
2616 <p>This value only has defined behavior when used as an operand to the
2617 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2618 comparisons against null. Pointer equality tests between labels addresses
2619 results in undefined behavior — though, again, comparison against null
2620 is ok, and no label is equal to the null pointer. This may be passed around
2621 as an opaque pointer sized value as long as the bits are not inspected. This
2622 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2623 long as the original value is reconstituted before the <tt>indirectbr</tt>
2626 <p>Finally, some targets may provide defined semantics when using the value as
2627 the operand to an inline assembly, but that is target specific.</p>
2632 <!-- ======================================================================= -->
2634 <a name="constantexprs">Constant Expressions</a>
2639 <p>Constant expressions are used to allow expressions involving other constants
2640 to be used as constants. Constant expressions may be of
2641 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2642 operation that does not have side effects (e.g. load and call are not
2643 supported). The following is the syntax for constant expressions:</p>
2646 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2647 <dd>Truncate a constant to another type. The bit size of CST must be larger
2648 than the bit size of TYPE. Both types must be integers.</dd>
2650 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2651 <dd>Zero extend a constant to another type. The bit size of CST must be
2652 smaller than the bit size of TYPE. Both types must be integers.</dd>
2654 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2655 <dd>Sign extend a constant to another type. The bit size of CST must be
2656 smaller than the bit size of TYPE. Both types must be integers.</dd>
2658 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2659 <dd>Truncate a floating point constant to another floating point type. The
2660 size of CST must be larger than the size of TYPE. Both types must be
2661 floating point.</dd>
2663 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2664 <dd>Floating point extend a constant to another type. The size of CST must be
2665 smaller or equal to the size of TYPE. Both types must be floating
2668 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2669 <dd>Convert a floating point constant to the corresponding unsigned integer
2670 constant. TYPE must be a scalar or vector integer type. CST must be of
2671 scalar or vector floating point type. Both CST and TYPE must be scalars,
2672 or vectors of the same number of elements. If the value won't fit in the
2673 integer type, the results are undefined.</dd>
2675 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2676 <dd>Convert a floating point constant to the corresponding signed integer
2677 constant. TYPE must be a scalar or vector integer type. CST must be of
2678 scalar or vector floating point type. Both CST and TYPE must be scalars,
2679 or vectors of the same number of elements. If the value won't fit in the
2680 integer type, the results are undefined.</dd>
2682 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2683 <dd>Convert an unsigned integer constant to the corresponding floating point
2684 constant. TYPE must be a scalar or vector floating point type. CST must be
2685 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2686 vectors of the same number of elements. If the value won't fit in the
2687 floating point type, the results are undefined.</dd>
2689 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2690 <dd>Convert a signed integer constant to the corresponding floating point
2691 constant. TYPE must be a scalar or vector floating point type. CST must be
2692 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2693 vectors of the same number of elements. If the value won't fit in the
2694 floating point type, the results are undefined.</dd>
2696 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2697 <dd>Convert a pointer typed constant to the corresponding integer constant
2698 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2699 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2700 make it fit in <tt>TYPE</tt>.</dd>
2702 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2703 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2704 type. CST must be of integer type. The CST value is zero extended,
2705 truncated, or unchanged to make it fit in a pointer size. This one is
2706 <i>really</i> dangerous!</dd>
2708 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2709 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2710 are the same as those for the <a href="#i_bitcast">bitcast
2711 instruction</a>.</dd>
2713 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2714 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2715 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2716 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2717 instruction, the index list may have zero or more indexes, which are
2718 required to make sense for the type of "CSTPTR".</dd>
2720 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2721 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2723 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2724 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2726 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2727 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2729 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2730 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2733 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2734 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2737 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2738 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2741 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2742 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2743 constants. The index list is interpreted in a similar manner as indices in
2744 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2745 index value must be specified.</dd>
2747 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2748 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2749 constants. The index list is interpreted in a similar manner as indices in
2750 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2751 index value must be specified.</dd>
2753 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2754 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2755 be any of the <a href="#binaryops">binary</a>
2756 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2757 on operands are the same as those for the corresponding instruction
2758 (e.g. no bitwise operations on floating point values are allowed).</dd>
2765 <!-- *********************************************************************** -->
2766 <h2><a name="othervalues">Other Values</a></h2>
2767 <!-- *********************************************************************** -->
2769 <!-- ======================================================================= -->
2771 <a name="inlineasm">Inline Assembler Expressions</a>
2776 <p>LLVM supports inline assembler expressions (as opposed
2777 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2778 a special value. This value represents the inline assembler as a string
2779 (containing the instructions to emit), a list of operand constraints (stored
2780 as a string), a flag that indicates whether or not the inline asm
2781 expression has side effects, and a flag indicating whether the function
2782 containing the asm needs to align its stack conservatively. An example
2783 inline assembler expression is:</p>
2785 <pre class="doc_code">
2786 i32 (i32) asm "bswap $0", "=r,r"
2789 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2790 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2793 <pre class="doc_code">
2794 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2797 <p>Inline asms with side effects not visible in the constraint list must be
2798 marked as having side effects. This is done through the use of the
2799 '<tt>sideeffect</tt>' keyword, like so:</p>
2801 <pre class="doc_code">
2802 call void asm sideeffect "eieio", ""()
2805 <p>In some cases inline asms will contain code that will not work unless the
2806 stack is aligned in some way, such as calls or SSE instructions on x86,
2807 yet will not contain code that does that alignment within the asm.
2808 The compiler should make conservative assumptions about what the asm might
2809 contain and should generate its usual stack alignment code in the prologue
2810 if the '<tt>alignstack</tt>' keyword is present:</p>
2812 <pre class="doc_code">
2813 call void asm alignstack "eieio", ""()
2816 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2819 <p>TODO: The format of the asm and constraints string still need to be
2820 documented here. Constraints on what can be done (e.g. duplication, moving,
2821 etc need to be documented). This is probably best done by reference to
2822 another document that covers inline asm from a holistic perspective.</p>
2825 <a name="inlineasm_md">Inline Asm Metadata</a>
2830 <p>The call instructions that wrap inline asm nodes may have a "!srcloc" MDNode
2831 attached to it that contains a list of constant integers. If present, the
2832 code generator will use the integer as the location cookie value when report
2833 errors through the LLVMContext error reporting mechanisms. This allows a
2834 front-end to correlate backend errors that occur with inline asm back to the
2835 source code that produced it. For example:</p>
2837 <pre class="doc_code">
2838 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2840 !42 = !{ i32 1234567 }
2843 <p>It is up to the front-end to make sense of the magic numbers it places in the
2844 IR. If the MDNode contains multiple constants, the code generator will use
2845 the one that corresponds to the line of the asm that the error occurs on.</p>
2851 <!-- ======================================================================= -->
2853 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2858 <p>LLVM IR allows metadata to be attached to instructions in the program that
2859 can convey extra information about the code to the optimizers and code
2860 generator. One example application of metadata is source-level debug
2861 information. There are two metadata primitives: strings and nodes. All
2862 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2863 preceding exclamation point ('<tt>!</tt>').</p>
2865 <p>A metadata string is a string surrounded by double quotes. It can contain
2866 any character by escaping non-printable characters with "\xx" where "xx" is
2867 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2869 <p>Metadata nodes are represented with notation similar to structure constants
2870 (a comma separated list of elements, surrounded by braces and preceded by an
2871 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2872 10}</tt>". Metadata nodes can have any values as their operand.</p>
2874 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2875 metadata nodes, which can be looked up in the module symbol table. For
2876 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2878 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2879 function is using two metadata arguments.</p>
2881 <div class="doc_code">
2883 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2887 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2888 attached with <tt>add</tt> instruction using <tt>!dbg</tt> identifier.</p>
2890 <div class="doc_code">
2892 %indvar.next = add i64 %indvar, 1, !dbg !21
2900 <!-- *********************************************************************** -->
2902 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2904 <!-- *********************************************************************** -->
2906 <p>LLVM has a number of "magic" global variables that contain data that affect
2907 code generation or other IR semantics. These are documented here. All globals
2908 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2909 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2912 <!-- ======================================================================= -->
2914 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2919 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2920 href="#linkage_appending">appending linkage</a>. This array contains a list of
2921 pointers to global variables and functions which may optionally have a pointer
2922 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2928 @llvm.used = appending global [2 x i8*] [
2930 i8* bitcast (i32* @Y to i8*)
2931 ], section "llvm.metadata"
2934 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2935 compiler, assembler, and linker are required to treat the symbol as if there is
2936 a reference to the global that it cannot see. For example, if a variable has
2937 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2938 list, it cannot be deleted. This is commonly used to represent references from
2939 inline asms and other things the compiler cannot "see", and corresponds to
2940 "attribute((used))" in GNU C.</p>
2942 <p>On some targets, the code generator must emit a directive to the assembler or
2943 object file to prevent the assembler and linker from molesting the symbol.</p>
2947 <!-- ======================================================================= -->
2949 <a name="intg_compiler_used">
2950 The '<tt>llvm.compiler.used</tt>' Global Variable
2956 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2957 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2958 touching the symbol. On targets that support it, this allows an intelligent
2959 linker to optimize references to the symbol without being impeded as it would be
2960 by <tt>@llvm.used</tt>.</p>
2962 <p>This is a rare construct that should only be used in rare circumstances, and
2963 should not be exposed to source languages.</p>
2967 <!-- ======================================================================= -->
2969 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2974 %0 = type { i32, void ()* }
2975 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2977 <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.
2982 <!-- ======================================================================= -->
2984 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2989 %0 = type { i32, void ()* }
2990 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2993 <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.
3000 <!-- *********************************************************************** -->
3001 <h2><a name="instref">Instruction Reference</a></h2>
3002 <!-- *********************************************************************** -->
3006 <p>The LLVM instruction set consists of several different classifications of
3007 instructions: <a href="#terminators">terminator
3008 instructions</a>, <a href="#binaryops">binary instructions</a>,
3009 <a href="#bitwiseops">bitwise binary instructions</a>,
3010 <a href="#memoryops">memory instructions</a>, and
3011 <a href="#otherops">other instructions</a>.</p>
3013 <!-- ======================================================================= -->
3015 <a name="terminators">Terminator Instructions</a>
3020 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3021 in a program ends with a "Terminator" instruction, which indicates which
3022 block should be executed after the current block is finished. These
3023 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3024 control flow, not values (the one exception being the
3025 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3027 <p>There are eight different terminator instructions: the
3028 '<a href="#i_ret"><tt>ret</tt></a>' instruction, the
3029 '<a href="#i_br"><tt>br</tt></a>' instruction, the
3030 '<a href="#i_switch"><tt>switch</tt></a>' instruction, the
3031 '<a href="#i_indirectbr">'<tt>indirectbr</tt></a>' Instruction, the
3032 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the
3033 '<a href="#i_unwind"><tt>unwind</tt></a>' instruction, the
3034 '<a href="#i_resume"><tt>resume</tt></a>' instruction, and the
3035 '<a href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
3037 <!-- _______________________________________________________________________ -->
3039 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3046 ret <type> <value> <i>; Return a value from a non-void function</i>
3047 ret void <i>; Return from void function</i>
3051 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3052 a value) from a function back to the caller.</p>
3054 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3055 value and then causes control flow, and one that just causes control flow to
3059 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3060 return value. The type of the return value must be a
3061 '<a href="#t_firstclass">first class</a>' type.</p>
3063 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3064 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3065 value or a return value with a type that does not match its type, or if it
3066 has a void return type and contains a '<tt>ret</tt>' instruction with a
3070 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3071 the calling function's context. If the caller is a
3072 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3073 instruction after the call. If the caller was an
3074 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3075 the beginning of the "normal" destination block. If the instruction returns
3076 a value, that value shall set the call or invoke instruction's return
3081 ret i32 5 <i>; Return an integer value of 5</i>
3082 ret void <i>; Return from a void function</i>
3083 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3087 <!-- _______________________________________________________________________ -->
3089 <a name="i_br">'<tt>br</tt>' Instruction</a>
3096 br i1 <cond>, label <iftrue>, label <iffalse>
3097 br label <dest> <i>; Unconditional branch</i>
3101 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3102 different basic block in the current function. There are two forms of this
3103 instruction, corresponding to a conditional branch and an unconditional
3107 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3108 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3109 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3113 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3114 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3115 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3116 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3121 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3122 br i1 %cond, label %IfEqual, label %IfUnequal
3124 <a href="#i_ret">ret</a> i32 1
3126 <a href="#i_ret">ret</a> i32 0
3131 <!-- _______________________________________________________________________ -->
3133 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3140 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3144 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3145 several different places. It is a generalization of the '<tt>br</tt>'
3146 instruction, allowing a branch to occur to one of many possible
3150 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3151 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3152 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3153 The table is not allowed to contain duplicate constant entries.</p>
3156 <p>The <tt>switch</tt> instruction specifies a table of values and
3157 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3158 is searched for the given value. If the value is found, control flow is
3159 transferred to the corresponding destination; otherwise, control flow is
3160 transferred to the default destination.</p>
3162 <h5>Implementation:</h5>
3163 <p>Depending on properties of the target machine and the particular
3164 <tt>switch</tt> instruction, this instruction may be code generated in
3165 different ways. For example, it could be generated as a series of chained
3166 conditional branches or with a lookup table.</p>
3170 <i>; Emulate a conditional br instruction</i>
3171 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3172 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3174 <i>; Emulate an unconditional br instruction</i>
3175 switch i32 0, label %dest [ ]
3177 <i>; Implement a jump table:</i>
3178 switch i32 %val, label %otherwise [ i32 0, label %onzero
3180 i32 2, label %ontwo ]
3186 <!-- _______________________________________________________________________ -->
3188 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3195 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3200 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3201 within the current function, whose address is specified by
3202 "<tt>address</tt>". Address must be derived from a <a
3203 href="#blockaddress">blockaddress</a> constant.</p>
3207 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3208 rest of the arguments indicate the full set of possible destinations that the
3209 address may point to. Blocks are allowed to occur multiple times in the
3210 destination list, though this isn't particularly useful.</p>
3212 <p>This destination list is required so that dataflow analysis has an accurate
3213 understanding of the CFG.</p>
3217 <p>Control transfers to the block specified in the address argument. All
3218 possible destination blocks must be listed in the label list, otherwise this
3219 instruction has undefined behavior. This implies that jumps to labels
3220 defined in other functions have undefined behavior as well.</p>
3222 <h5>Implementation:</h5>
3224 <p>This is typically implemented with a jump through a register.</p>
3228 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3234 <!-- _______________________________________________________________________ -->
3236 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3243 <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>]
3244 to label <normal label> unwind label <exception label>
3248 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3249 function, with the possibility of control flow transfer to either the
3250 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3251 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3252 control flow will return to the "normal" label. If the callee (or any
3253 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
3254 instruction, control is interrupted and continued at the dynamically nearest
3255 "exception" label.</p>
3258 <p>This instruction requires several arguments:</p>
3261 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3262 convention</a> the call should use. If none is specified, the call
3263 defaults to using C calling conventions.</li>
3265 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3266 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3267 '<tt>inreg</tt>' attributes are valid here.</li>
3269 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3270 function value being invoked. In most cases, this is a direct function
3271 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3272 off an arbitrary pointer to function value.</li>
3274 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3275 function to be invoked. </li>
3277 <li>'<tt>function args</tt>': argument list whose types match the function
3278 signature argument types and parameter attributes. All arguments must be
3279 of <a href="#t_firstclass">first class</a> type. If the function
3280 signature indicates the function accepts a variable number of arguments,
3281 the extra arguments can be specified.</li>
3283 <li>'<tt>normal label</tt>': the label reached when the called function
3284 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3286 <li>'<tt>exception label</tt>': the label reached when a callee returns with
3287 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
3289 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3290 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3291 '<tt>readnone</tt>' attributes are valid here.</li>
3295 <p>This instruction is designed to operate as a standard
3296 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3297 primary difference is that it establishes an association with a label, which
3298 is used by the runtime library to unwind the stack.</p>
3300 <p>This instruction is used in languages with destructors to ensure that proper
3301 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3302 exception. Additionally, this is important for implementation of
3303 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3305 <p>For the purposes of the SSA form, the definition of the value returned by the
3306 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3307 block to the "normal" label. If the callee unwinds then no return value is
3310 <p>Note that the code generator does not yet completely support unwind, and
3311 that the invoke/unwind semantics are likely to change in future versions.</p>
3315 %retval = invoke i32 @Test(i32 15) to label %Continue
3316 unwind label %TestCleanup <i>; {i32}:retval set</i>
3317 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3318 unwind label %TestCleanup <i>; {i32}:retval set</i>
3323 <!-- _______________________________________________________________________ -->
3326 <a name="i_unwind">'<tt>unwind</tt>' Instruction</a>
3337 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
3338 at the first callee in the dynamic call stack which used
3339 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
3340 This is primarily used to implement exception handling.</p>
3343 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
3344 immediately halt. The dynamic call stack is then searched for the
3345 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
3346 Once found, execution continues at the "exceptional" destination block
3347 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
3348 instruction in the dynamic call chain, undefined behavior results.</p>
3350 <p>Note that the code generator does not yet completely support unwind, and
3351 that the invoke/unwind semantics are likely to change in future versions.</p>
3355 <!-- _______________________________________________________________________ -->
3358 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3365 resume <type> <value>
3369 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3370 successors. Its operand must have the same type as the result of any
3371 '<tt>landingpad</tt>' instruction in the same function.</p>
3374 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3375 (in-flight) exception.</p>
3379 resume { i8*, i32 } %exn
3384 <!-- _______________________________________________________________________ -->
3387 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3398 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3399 instruction is used to inform the optimizer that a particular portion of the
3400 code is not reachable. This can be used to indicate that the code after a
3401 no-return function cannot be reached, and other facts.</p>
3404 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3410 <!-- ======================================================================= -->
3412 <a name="binaryops">Binary Operations</a>
3417 <p>Binary operators are used to do most of the computation in a program. They
3418 require two operands of the same type, execute an operation on them, and
3419 produce a single value. The operands might represent multiple data, as is
3420 the case with the <a href="#t_vector">vector</a> data type. The result value
3421 has the same type as its operands.</p>
3423 <p>There are several different binary operators:</p>
3425 <!-- _______________________________________________________________________ -->
3427 <a name="i_add">'<tt>add</tt>' Instruction</a>
3434 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3435 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3436 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3437 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3441 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3444 <p>The two arguments to the '<tt>add</tt>' instruction must
3445 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3446 integer values. Both arguments must have identical types.</p>
3449 <p>The value produced is the integer sum of the two operands.</p>
3451 <p>If the sum has unsigned overflow, the result returned is the mathematical
3452 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3454 <p>Because LLVM integers use a two's complement representation, this instruction
3455 is appropriate for both signed and unsigned integers.</p>
3457 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3458 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3459 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3460 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3461 respectively, occurs.</p>
3465 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3470 <!-- _______________________________________________________________________ -->
3472 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3479 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3483 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3486 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3487 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3488 floating point values. Both arguments must have identical types.</p>
3491 <p>The value produced is the floating point sum of the two operands.</p>
3495 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3500 <!-- _______________________________________________________________________ -->
3502 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3509 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3510 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3511 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3512 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3516 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3519 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3520 '<tt>neg</tt>' instruction present in most other intermediate
3521 representations.</p>
3524 <p>The two arguments to the '<tt>sub</tt>' instruction must
3525 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3526 integer values. Both arguments must have identical types.</p>
3529 <p>The value produced is the integer difference of the two operands.</p>
3531 <p>If the difference has unsigned overflow, the result returned is the
3532 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3535 <p>Because LLVM integers use a two's complement representation, this instruction
3536 is appropriate for both signed and unsigned integers.</p>
3538 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3539 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3540 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3541 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3542 respectively, occurs.</p>
3546 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3547 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3552 <!-- _______________________________________________________________________ -->
3554 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3561 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3565 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3568 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3569 '<tt>fneg</tt>' instruction present in most other intermediate
3570 representations.</p>
3573 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3574 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3575 floating point values. Both arguments must have identical types.</p>
3578 <p>The value produced is the floating point difference of the two operands.</p>
3582 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3583 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3588 <!-- _______________________________________________________________________ -->
3590 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3597 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3598 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3599 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3600 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3604 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3607 <p>The two arguments to the '<tt>mul</tt>' instruction must
3608 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3609 integer values. Both arguments must have identical types.</p>
3612 <p>The value produced is the integer product of the two operands.</p>
3614 <p>If the result of the multiplication has unsigned overflow, the result
3615 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3616 width of the result.</p>
3618 <p>Because LLVM integers use a two's complement representation, and the result
3619 is the same width as the operands, this instruction returns the correct
3620 result for both signed and unsigned integers. If a full product
3621 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3622 be sign-extended or zero-extended as appropriate to the width of the full
3625 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3626 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3627 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3628 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3629 respectively, occurs.</p>
3633 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3638 <!-- _______________________________________________________________________ -->
3640 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3647 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3651 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3654 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3655 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3656 floating point values. Both arguments must have identical types.</p>
3659 <p>The value produced is the floating point product of the two operands.</p>
3663 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3668 <!-- _______________________________________________________________________ -->
3670 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
3677 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3678 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3682 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3685 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3686 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3687 values. Both arguments must have identical types.</p>
3690 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3692 <p>Note that unsigned integer division and signed integer division are distinct
3693 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3695 <p>Division by zero leads to undefined behavior.</p>
3697 <p>If the <tt>exact</tt> keyword is present, the result value of the
3698 <tt>udiv</tt> is a <a href="#trapvalues">trap value</a> if %op1 is not a
3699 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
3704 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3709 <!-- _______________________________________________________________________ -->
3711 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
3718 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3719 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3723 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3726 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3727 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3728 values. Both arguments must have identical types.</p>
3731 <p>The value produced is the signed integer quotient of the two operands rounded
3734 <p>Note that signed integer division and unsigned integer division are distinct
3735 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3737 <p>Division by zero leads to undefined behavior. Overflow also leads to
3738 undefined behavior; this is a rare case, but can occur, for example, by doing
3739 a 32-bit division of -2147483648 by -1.</p>
3741 <p>If the <tt>exact</tt> keyword is present, the result value of the
3742 <tt>sdiv</tt> is a <a href="#trapvalues">trap value</a> if the result would
3747 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3752 <!-- _______________________________________________________________________ -->
3754 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
3761 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3765 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3768 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3769 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3770 floating point values. Both arguments must have identical types.</p>
3773 <p>The value produced is the floating point quotient of the two operands.</p>
3777 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3782 <!-- _______________________________________________________________________ -->
3784 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3791 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3795 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3796 division of its two arguments.</p>
3799 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3800 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3801 values. Both arguments must have identical types.</p>
3804 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3805 This instruction always performs an unsigned division to get the
3808 <p>Note that unsigned integer remainder and signed integer remainder are
3809 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3811 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3815 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3820 <!-- _______________________________________________________________________ -->
3822 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3829 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3833 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3834 division of its two operands. This instruction can also take
3835 <a href="#t_vector">vector</a> versions of the values in which case the
3836 elements must be integers.</p>
3839 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3840 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3841 values. Both arguments must have identical types.</p>
3844 <p>This instruction returns the <i>remainder</i> of a division (where the result
3845 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
3846 <i>modulo</i> operator (where the result is either zero or has the same sign
3847 as the divisor, <tt>op2</tt>) of a value.
3848 For more information about the difference,
3849 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3850 Math Forum</a>. For a table of how this is implemented in various languages,
3851 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3852 Wikipedia: modulo operation</a>.</p>
3854 <p>Note that signed integer remainder and unsigned integer remainder are
3855 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3857 <p>Taking the remainder of a division by zero leads to undefined behavior.
3858 Overflow also leads to undefined behavior; this is a rare case, but can
3859 occur, for example, by taking the remainder of a 32-bit division of
3860 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3861 lets srem be implemented using instructions that return both the result of
3862 the division and the remainder.)</p>
3866 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3871 <!-- _______________________________________________________________________ -->
3873 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
3880 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3884 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3885 its two operands.</p>
3888 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3889 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3890 floating point values. Both arguments must have identical types.</p>
3893 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3894 has the same sign as the dividend.</p>
3898 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3905 <!-- ======================================================================= -->
3907 <a name="bitwiseops">Bitwise Binary Operations</a>
3912 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3913 program. They are generally very efficient instructions and can commonly be
3914 strength reduced from other instructions. They require two operands of the
3915 same type, execute an operation on them, and produce a single value. The
3916 resulting value is the same type as its operands.</p>
3918 <!-- _______________________________________________________________________ -->
3920 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
3927 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3928 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3929 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3930 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3934 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3935 a specified number of bits.</p>
3938 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3939 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3940 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3943 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3944 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3945 is (statically or dynamically) negative or equal to or larger than the number
3946 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3947 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3948 shift amount in <tt>op2</tt>.</p>
3950 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
3951 <a href="#trapvalues">trap value</a> if it shifts out any non-zero bits. If
3952 the <tt>nsw</tt> keyword is present, then the shift produces a
3953 <a href="#trapvalues">trap value</a> if it shifts out any bits that disagree
3954 with the resultant sign bit. As such, NUW/NSW have the same semantics as
3955 they would if the shift were expressed as a mul instruction with the same
3956 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
3960 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3961 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3962 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3963 <result> = shl i32 1, 32 <i>; undefined</i>
3964 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3969 <!-- _______________________________________________________________________ -->
3971 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
3978 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3979 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3983 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
3984 operand shifted to the right a specified number of bits with zero fill.</p>
3987 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
3988 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3989 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3992 <p>This instruction always performs a logical shift right operation. The most
3993 significant bits of the result will be filled with zero bits after the shift.
3994 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
3995 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3996 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3997 shift amount in <tt>op2</tt>.</p>
3999 <p>If the <tt>exact</tt> keyword is present, the result value of the
4000 <tt>lshr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
4001 shifted out are non-zero.</p>
4006 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4007 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4008 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4009 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4010 <result> = lshr i32 1, 32 <i>; undefined</i>
4011 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4016 <!-- _______________________________________________________________________ -->
4018 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4025 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4026 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4030 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4031 operand shifted to the right a specified number of bits with sign
4035 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4036 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4037 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4040 <p>This instruction always performs an arithmetic shift right operation, The
4041 most significant bits of the result will be filled with the sign bit
4042 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4043 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4044 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4045 the corresponding shift amount in <tt>op2</tt>.</p>
4047 <p>If the <tt>exact</tt> keyword is present, the result value of the
4048 <tt>ashr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
4049 shifted out are non-zero.</p>
4053 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4054 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4055 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4056 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4057 <result> = ashr i32 1, 32 <i>; undefined</i>
4058 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4063 <!-- _______________________________________________________________________ -->
4065 <a name="i_and">'<tt>and</tt>' Instruction</a>
4072 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4076 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4080 <p>The two arguments to the '<tt>and</tt>' instruction must be
4081 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4082 values. Both arguments must have identical types.</p>
4085 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4087 <table border="1" cellspacing="0" cellpadding="4">
4119 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4120 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4121 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4124 <!-- _______________________________________________________________________ -->
4126 <a name="i_or">'<tt>or</tt>' Instruction</a>
4133 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4137 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4141 <p>The two arguments to the '<tt>or</tt>' instruction must be
4142 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4143 values. Both arguments must have identical types.</p>
4146 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4148 <table border="1" cellspacing="0" cellpadding="4">
4180 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4181 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4182 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4187 <!-- _______________________________________________________________________ -->
4189 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4196 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4200 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4201 its two operands. The <tt>xor</tt> is used to implement the "one's
4202 complement" operation, which is the "~" operator in C.</p>
4205 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4206 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4207 values. Both arguments must have identical types.</p>
4210 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4212 <table border="1" cellspacing="0" cellpadding="4">
4244 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4245 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4246 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4247 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4254 <!-- ======================================================================= -->
4256 <a name="vectorops">Vector Operations</a>
4261 <p>LLVM supports several instructions to represent vector operations in a
4262 target-independent manner. These instructions cover the element-access and
4263 vector-specific operations needed to process vectors effectively. While LLVM
4264 does directly support these vector operations, many sophisticated algorithms
4265 will want to use target-specific intrinsics to take full advantage of a
4266 specific target.</p>
4268 <!-- _______________________________________________________________________ -->
4270 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4277 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4281 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4282 from a vector at a specified index.</p>
4286 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4287 of <a href="#t_vector">vector</a> type. The second operand is an index
4288 indicating the position from which to extract the element. The index may be
4292 <p>The result is a scalar of the same type as the element type of
4293 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4294 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4295 results are undefined.</p>
4299 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4304 <!-- _______________________________________________________________________ -->
4306 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4313 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4317 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4318 vector at a specified index.</p>
4321 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4322 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4323 whose type must equal the element type of the first operand. The third
4324 operand is an index indicating the position at which to insert the value.
4325 The index may be a variable.</p>
4328 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4329 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4330 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4331 results are undefined.</p>
4335 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4340 <!-- _______________________________________________________________________ -->
4342 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4349 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4353 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4354 from two input vectors, returning a vector with the same element type as the
4355 input and length that is the same as the shuffle mask.</p>
4358 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4359 with types that match each other. The third argument is a shuffle mask whose
4360 element type is always 'i32'. The result of the instruction is a vector
4361 whose length is the same as the shuffle mask and whose element type is the
4362 same as the element type of the first two operands.</p>
4364 <p>The shuffle mask operand is required to be a constant vector with either
4365 constant integer or undef values.</p>
4368 <p>The elements of the two input vectors are numbered from left to right across
4369 both of the vectors. The shuffle mask operand specifies, for each element of
4370 the result vector, which element of the two input vectors the result element
4371 gets. The element selector may be undef (meaning "don't care") and the
4372 second operand may be undef if performing a shuffle from only one vector.</p>
4376 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4377 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4378 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4379 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4380 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4381 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4382 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4383 <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>
4390 <!-- ======================================================================= -->
4392 <a name="aggregateops">Aggregate Operations</a>
4397 <p>LLVM supports several instructions for working with
4398 <a href="#t_aggregate">aggregate</a> values.</p>
4400 <!-- _______________________________________________________________________ -->
4402 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4409 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4413 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4414 from an <a href="#t_aggregate">aggregate</a> value.</p>
4417 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4418 of <a href="#t_struct">struct</a> or
4419 <a href="#t_array">array</a> type. The operands are constant indices to
4420 specify which value to extract in a similar manner as indices in a
4421 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4422 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4424 <li>Since the value being indexed is not a pointer, the first index is
4425 omitted and assumed to be zero.</li>
4426 <li>At least one index must be specified.</li>
4427 <li>Not only struct indices but also array indices must be in
4432 <p>The result is the value at the position in the aggregate specified by the
4437 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4442 <!-- _______________________________________________________________________ -->
4444 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4451 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4455 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4456 in an <a href="#t_aggregate">aggregate</a> value.</p>
4459 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4460 of <a href="#t_struct">struct</a> or
4461 <a href="#t_array">array</a> type. The second operand is a first-class
4462 value to insert. The following operands are constant indices indicating
4463 the position at which to insert the value in a similar manner as indices in a
4464 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4465 value to insert must have the same type as the value identified by the
4469 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4470 that of <tt>val</tt> except that the value at the position specified by the
4471 indices is that of <tt>elt</tt>.</p>
4475 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4476 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4477 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4484 <!-- ======================================================================= -->
4486 <a name="memoryops">Memory Access and Addressing Operations</a>
4491 <p>A key design point of an SSA-based representation is how it represents
4492 memory. In LLVM, no memory locations are in SSA form, which makes things
4493 very simple. This section describes how to read, write, and allocate
4496 <!-- _______________________________________________________________________ -->
4498 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4505 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4509 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4510 currently executing function, to be automatically released when this function
4511 returns to its caller. The object is always allocated in the generic address
4512 space (address space zero).</p>
4515 <p>The '<tt>alloca</tt>' instruction
4516 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4517 runtime stack, returning a pointer of the appropriate type to the program.
4518 If "NumElements" is specified, it is the number of elements allocated,
4519 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4520 specified, the value result of the allocation is guaranteed to be aligned to
4521 at least that boundary. If not specified, or if zero, the target can choose
4522 to align the allocation on any convenient boundary compatible with the
4525 <p>'<tt>type</tt>' may be any sized type.</p>
4528 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4529 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4530 memory is automatically released when the function returns. The
4531 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4532 variables that must have an address available. When the function returns
4533 (either with the <tt><a href="#i_ret">ret</a></tt>
4534 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
4535 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4539 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4540 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4541 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4542 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4547 <!-- _______________________________________________________________________ -->
4549 <a name="i_load">'<tt>load</tt>' Instruction</a>
4556 <result> = load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4557 <result> = volatile load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4558 !<index> = !{ i32 1 }
4562 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4565 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4566 from which to load. The pointer must point to
4567 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4568 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4569 number or order of execution of this <tt>load</tt> with other <a
4570 href="#volatile">volatile operations</a>.</p>
4572 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4573 operation (that is, the alignment of the memory address). A value of 0 or an
4574 omitted <tt>align</tt> argument means that the operation has the preferential
4575 alignment for the target. It is the responsibility of the code emitter to
4576 ensure that the alignment information is correct. Overestimating the
4577 alignment results in undefined behavior. Underestimating the alignment may
4578 produce less efficient code. An alignment of 1 is always safe.</p>
4580 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4581 metatadata name <index> corresponding to a metadata node with
4582 one <tt>i32</tt> entry of value 1. The existence of
4583 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4584 and code generator that this load is not expected to be reused in the cache.
4585 The code generator may select special instructions to save cache bandwidth,
4586 such as the <tt>MOVNT</tt> instruction on x86.</p>
4589 <p>The location of memory pointed to is loaded. If the value being loaded is of
4590 scalar type then the number of bytes read does not exceed the minimum number
4591 of bytes needed to hold all bits of the type. For example, loading an
4592 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4593 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4594 is undefined if the value was not originally written using a store of the
4599 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4600 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4601 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4606 <!-- _______________________________________________________________________ -->
4608 <a name="i_store">'<tt>store</tt>' Instruction</a>
4615 store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4616 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4620 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4623 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4624 and an address at which to store it. The type of the
4625 '<tt><pointer></tt>' operand must be a pointer to
4626 the <a href="#t_firstclass">first class</a> type of the
4627 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
4628 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
4629 order of execution of this <tt>store</tt> with other <a
4630 href="#volatile">volatile operations</a>.</p>
4632 <p>The optional constant "align" argument specifies the alignment of the
4633 operation (that is, the alignment of the memory address). A value of 0 or an
4634 omitted "align" argument means that the operation has the preferential
4635 alignment for the target. It is the responsibility of the code emitter to
4636 ensure that the alignment information is correct. Overestimating the
4637 alignment results in an undefined behavior. Underestimating the alignment may
4638 produce less efficient code. An alignment of 1 is always safe.</p>
4640 <p>The optional !nontemporal metadata must reference a single metatadata
4641 name <index> corresponding to a metadata node with one i32 entry of
4642 value 1. The existence of the !nontemporal metatadata on the
4643 instruction tells the optimizer and code generator that this load is
4644 not expected to be reused in the cache. The code generator may
4645 select special instructions to save cache bandwidth, such as the
4646 MOVNT instruction on x86.</p>
4650 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4651 location specified by the '<tt><pointer></tt>' operand. If
4652 '<tt><value></tt>' is of scalar type then the number of bytes written
4653 does not exceed the minimum number of bytes needed to hold all bits of the
4654 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4655 writing a value of a type like <tt>i20</tt> with a size that is not an
4656 integral number of bytes, it is unspecified what happens to the extra bits
4657 that do not belong to the type, but they will typically be overwritten.</p>
4661 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4662 store i32 3, i32* %ptr <i>; yields {void}</i>
4663 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4668 <!-- _______________________________________________________________________ -->
4669 <div class="doc_subsubsection"> <a name="i_fence">'<tt>fence</tt>'
4670 Instruction</a> </div>
4672 <div class="doc_text">
4676 fence [singlethread] <ordering> <i>; yields {void}</i>
4680 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
4681 between operations.</p>
4683 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
4684 href="#ordering">ordering</a> argument which defines what
4685 <i>synchronizes-with</i> edges they add. They can only be given
4686 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
4687 <code>seq_cst</code> orderings.</p>
4690 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
4691 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
4692 <code>acquire</code> ordering semantics if and only if there exist atomic
4693 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
4694 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
4695 <var>X</var> modifies <var>M</var> (either directly or through some side effect
4696 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
4697 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
4698 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
4699 than an explicit <code>fence</code>, one (but not both) of the atomic operations
4700 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
4701 <code>acquire</code> (resp.) ordering constraint and still
4702 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
4703 <i>happens-before</i> edge.</p>
4705 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
4706 having both <code>acquire</code> and <code>release</code> semantics specified
4707 above, participates in the global program order of other <code>seq_cst</code>
4708 operations and/or fences.</p>
4710 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
4711 specifies that the fence only synchronizes with other fences in the same
4712 thread. (This is useful for interacting with signal handlers.)</p>
4714 <p>FIXME: This instruction is a work in progress; until it is finished, use
4715 llvm.memory.barrier.
4719 fence acquire <i>; yields {void}</i>
4720 fence singlethread seq_cst <i>; yields {void}</i>
4725 <!-- _______________________________________________________________________ -->
4726 <div class="doc_subsubsection"> <a name="i_cmpxchg">'<tt>cmpxchg</tt>'
4727 Instruction</a> </div>
4729 <div class="doc_text">
4733 [volatile] cmpxchg <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
4737 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
4738 It loads a value in memory and compares it to a given value. If they are
4739 equal, it stores a new value into the memory.</p>
4742 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
4743 address to operate on, a value to compare to the value currently be at that
4744 address, and a new value to place at that address if the compared values are
4745 equal. The type of '<var><cmp></var>' must be an integer type whose
4746 bit width is a power of two greater than or equal to eight and less than
4747 or equal to a target-specific size limit. '<var><cmp></var>' and
4748 '<var><new></var>' must have the same type, and the type of
4749 '<var><pointer></var>' must be a pointer to that type. If the
4750 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
4751 optimizer is not allowed to modify the number or order of execution
4752 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
4755 <!-- FIXME: Extend allowed types. -->
4757 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
4758 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
4760 <p>The optional "<code>singlethread</code>" argument declares that the
4761 <code>cmpxchg</code> is only atomic with respect to code (usually signal
4762 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
4763 cmpxchg is atomic with respect to all other code in the system.</p>
4765 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
4766 the size in memory of the operand.
4769 <p>The contents of memory at the location specified by the
4770 '<tt><pointer></tt>' operand is read and compared to
4771 '<tt><cmp></tt>'; if the read value is the equal,
4772 '<tt><new></tt>' is written. The original value at the location
4775 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
4776 purpose of identifying <a href="#release_sequence">release sequences</a>. A
4777 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
4778 parameter determined by dropping any <code>release</code> part of the
4779 <code>cmpxchg</code>'s ordering.</p>
4782 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
4783 optimization work on ARM.)
4785 FIXME: Is a weaker ordering constraint on failure helpful in practice?
4791 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
4792 <a href="#i_br">br</a> label %loop
4795 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
4796 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
4797 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
4798 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
4799 <a href="#i_br">br</a> i1 %success, label %done, label %loop
4807 <!-- _______________________________________________________________________ -->
4808 <div class="doc_subsubsection"> <a name="i_atomicrmw">'<tt>atomicrmw</tt>'
4809 Instruction</a> </div>
4811 <div class="doc_text">
4815 [volatile] atomicrmw <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
4819 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
4822 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
4823 operation to apply, an address whose value to modify, an argument to the
4824 operation. The operation must be one of the following keywords:</p>
4839 <p>The type of '<var><value></var>' must be an integer type whose
4840 bit width is a power of two greater than or equal to eight and less than
4841 or equal to a target-specific size limit. The type of the
4842 '<code><pointer></code>' operand must be a pointer to that type.
4843 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
4844 optimizer is not allowed to modify the number or order of execution of this
4845 <code>atomicrmw</code> with other <a href="#volatile">volatile
4848 <!-- FIXME: Extend allowed types. -->
4851 <p>The contents of memory at the location specified by the
4852 '<tt><pointer></tt>' operand are atomically read, modified, and written
4853 back. The original value at the location is returned. The modification is
4854 specified by the <var>operation</var> argument:</p>
4857 <li>xchg: <code>*ptr = val</code></li>
4858 <li>add: <code>*ptr = *ptr + val</code></li>
4859 <li>sub: <code>*ptr = *ptr - val</code></li>
4860 <li>and: <code>*ptr = *ptr & val</code></li>
4861 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
4862 <li>or: <code>*ptr = *ptr | val</code></li>
4863 <li>xor: <code>*ptr = *ptr ^ val</code></li>
4864 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
4865 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
4866 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
4867 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
4872 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
4877 <!-- _______________________________________________________________________ -->
4879 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
4886 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4887 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4891 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4892 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
4893 It performs address calculation only and does not access memory.</p>
4896 <p>The first argument is always a pointer, and forms the basis of the
4897 calculation. The remaining arguments are indices that indicate which of the
4898 elements of the aggregate object are indexed. The interpretation of each
4899 index is dependent on the type being indexed into. The first index always
4900 indexes the pointer value given as the first argument, the second index
4901 indexes a value of the type pointed to (not necessarily the value directly
4902 pointed to, since the first index can be non-zero), etc. The first type
4903 indexed into must be a pointer value, subsequent types can be arrays,
4904 vectors, and structs. Note that subsequent types being indexed into
4905 can never be pointers, since that would require loading the pointer before
4906 continuing calculation.</p>
4908 <p>The type of each index argument depends on the type it is indexing into.
4909 When indexing into a (optionally packed) structure, only <tt>i32</tt>
4910 integer <b>constants</b> are allowed. When indexing into an array, pointer
4911 or vector, integers of any width are allowed, and they are not required to be
4914 <p>For example, let's consider a C code fragment and how it gets compiled to
4917 <pre class="doc_code">
4929 int *foo(struct ST *s) {
4930 return &s[1].Z.B[5][13];
4934 <p>The LLVM code generated by the GCC frontend is:</p>
4936 <pre class="doc_code">
4937 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4938 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
4940 define i32* @foo(%ST* %s) {
4942 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
4948 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
4949 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
4950 }</tt>' type, a structure. The second index indexes into the third element
4951 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
4952 i8 }</tt>' type, another structure. The third index indexes into the second
4953 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
4954 array. The two dimensions of the array are subscripted into, yielding an
4955 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
4956 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
4958 <p>Note that it is perfectly legal to index partially through a structure,
4959 returning a pointer to an inner element. Because of this, the LLVM code for
4960 the given testcase is equivalent to:</p>
4963 define i32* @foo(%ST* %s) {
4964 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
4965 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
4966 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
4967 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
4968 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
4973 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
4974 <tt>getelementptr</tt> is a <a href="#trapvalues">trap value</a> if the
4975 base pointer is not an <i>in bounds</i> address of an allocated object,
4976 or if any of the addresses that would be formed by successive addition of
4977 the offsets implied by the indices to the base address with infinitely
4978 precise arithmetic are not an <i>in bounds</i> address of that allocated
4979 object. The <i>in bounds</i> addresses for an allocated object are all
4980 the addresses that point into the object, plus the address one byte past
4983 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
4984 the base address with silently-wrapping two's complement arithmetic, and
4985 the result value of the <tt>getelementptr</tt> may be outside the object
4986 pointed to by the base pointer. The result value may not necessarily be
4987 used to access memory though, even if it happens to point into allocated
4988 storage. See the <a href="#pointeraliasing">Pointer Aliasing Rules</a>
4989 section for more information.</p>
4991 <p>The getelementptr instruction is often confusing. For some more insight into
4992 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
4996 <i>; yields [12 x i8]*:aptr</i>
4997 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4998 <i>; yields i8*:vptr</i>
4999 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5000 <i>; yields i8*:eptr</i>
5001 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5002 <i>; yields i32*:iptr</i>
5003 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5010 <!-- ======================================================================= -->
5012 <a name="convertops">Conversion Operations</a>
5017 <p>The instructions in this category are the conversion instructions (casting)
5018 which all take a single operand and a type. They perform various bit
5019 conversions on the operand.</p>
5021 <!-- _______________________________________________________________________ -->
5023 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5030 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5034 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5035 type <tt>ty2</tt>.</p>
5038 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5039 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5040 of the same number of integers.
5041 The bit size of the <tt>value</tt> must be larger than
5042 the bit size of the destination type, <tt>ty2</tt>.
5043 Equal sized types are not allowed.</p>
5046 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5047 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5048 source size must be larger than the destination size, <tt>trunc</tt> cannot
5049 be a <i>no-op cast</i>. It will always truncate bits.</p>
5053 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5054 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5055 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5056 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5061 <!-- _______________________________________________________________________ -->
5063 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5070 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5074 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5079 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5080 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5081 of the same number of integers.
5082 The bit size of the <tt>value</tt> must be smaller than
5083 the bit size of the destination type,
5087 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5088 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5090 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5094 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5095 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5096 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5101 <!-- _______________________________________________________________________ -->
5103 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5110 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5114 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5117 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5118 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5119 of the same number of integers.
5120 The bit size of the <tt>value</tt> must be smaller than
5121 the bit size of the destination type,
5125 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5126 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5127 of the type <tt>ty2</tt>.</p>
5129 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5133 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5134 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5135 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5140 <!-- _______________________________________________________________________ -->
5142 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5149 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5153 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5157 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5158 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5159 to cast it to. The size of <tt>value</tt> must be larger than the size of
5160 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5161 <i>no-op cast</i>.</p>
5164 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5165 <a href="#t_floating">floating point</a> type to a smaller
5166 <a href="#t_floating">floating point</a> type. If the value cannot fit
5167 within the destination type, <tt>ty2</tt>, then the results are
5172 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5173 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5178 <!-- _______________________________________________________________________ -->
5180 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5187 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5191 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5192 floating point value.</p>
5195 <p>The '<tt>fpext</tt>' instruction takes a
5196 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5197 a <a href="#t_floating">floating point</a> type to cast it to. The source
5198 type must be smaller than the destination type.</p>
5201 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5202 <a href="#t_floating">floating point</a> type to a larger
5203 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5204 used to make a <i>no-op cast</i> because it always changes bits. Use
5205 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5209 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5210 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5215 <!-- _______________________________________________________________________ -->
5217 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5224 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5228 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5229 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5232 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5233 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5234 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5235 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5236 vector integer type with the same number of elements as <tt>ty</tt></p>
5239 <p>The '<tt>fptoui</tt>' instruction converts its
5240 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5241 towards zero) unsigned integer value. If the value cannot fit
5242 in <tt>ty2</tt>, the results are undefined.</p>
5246 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5247 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5248 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5253 <!-- _______________________________________________________________________ -->
5255 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5262 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5266 <p>The '<tt>fptosi</tt>' instruction converts
5267 <a href="#t_floating">floating point</a> <tt>value</tt> to
5268 type <tt>ty2</tt>.</p>
5271 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5272 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5273 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5274 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5275 vector integer type with the same number of elements as <tt>ty</tt></p>
5278 <p>The '<tt>fptosi</tt>' instruction converts its
5279 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5280 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5281 the results are undefined.</p>
5285 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5286 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5287 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5292 <!-- _______________________________________________________________________ -->
5294 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5301 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5305 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5306 integer and converts that value to the <tt>ty2</tt> type.</p>
5309 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5310 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5311 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5312 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5313 floating point type with the same number of elements as <tt>ty</tt></p>
5316 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5317 integer quantity and converts it to the corresponding floating point
5318 value. If the value cannot fit in the floating point value, the results are
5323 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5324 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5329 <!-- _______________________________________________________________________ -->
5331 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5338 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5342 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5343 and converts that value to the <tt>ty2</tt> type.</p>
5346 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5347 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5348 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5349 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5350 floating point type with the same number of elements as <tt>ty</tt></p>
5353 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5354 quantity and converts it to the corresponding floating point value. If the
5355 value cannot fit in the floating point value, the results are undefined.</p>
5359 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5360 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5365 <!-- _______________________________________________________________________ -->
5367 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5374 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5378 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
5379 the integer type <tt>ty2</tt>.</p>
5382 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5383 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
5384 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
5387 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5388 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5389 truncating or zero extending that value to the size of the integer type. If
5390 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5391 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5392 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5397 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
5398 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
5403 <!-- _______________________________________________________________________ -->
5405 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5412 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5416 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5417 pointer type, <tt>ty2</tt>.</p>
5420 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5421 value to cast, and a type to cast it to, which must be a
5422 <a href="#t_pointer">pointer</a> type.</p>
5425 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5426 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5427 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5428 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5429 than the size of a pointer then a zero extension is done. If they are the
5430 same size, nothing is done (<i>no-op cast</i>).</p>
5434 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5435 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5436 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5441 <!-- _______________________________________________________________________ -->
5443 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5450 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5454 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5455 <tt>ty2</tt> without changing any bits.</p>
5458 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5459 non-aggregate first class value, and a type to cast it to, which must also be
5460 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5461 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5462 identical. If the source type is a pointer, the destination type must also be
5463 a pointer. This instruction supports bitwise conversion of vectors to
5464 integers and to vectors of other types (as long as they have the same
5468 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5469 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5470 this conversion. The conversion is done as if the <tt>value</tt> had been
5471 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
5472 be converted to other pointer types with this instruction. To convert
5473 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5474 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5478 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5479 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5480 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5487 <!-- ======================================================================= -->
5489 <a name="otherops">Other Operations</a>
5494 <p>The instructions in this category are the "miscellaneous" instructions, which
5495 defy better classification.</p>
5497 <!-- _______________________________________________________________________ -->
5499 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5506 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5510 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5511 boolean values based on comparison of its two integer, integer vector, or
5512 pointer operands.</p>
5515 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5516 the condition code indicating the kind of comparison to perform. It is not a
5517 value, just a keyword. The possible condition code are:</p>
5520 <li><tt>eq</tt>: equal</li>
5521 <li><tt>ne</tt>: not equal </li>
5522 <li><tt>ugt</tt>: unsigned greater than</li>
5523 <li><tt>uge</tt>: unsigned greater or equal</li>
5524 <li><tt>ult</tt>: unsigned less than</li>
5525 <li><tt>ule</tt>: unsigned less or equal</li>
5526 <li><tt>sgt</tt>: signed greater than</li>
5527 <li><tt>sge</tt>: signed greater or equal</li>
5528 <li><tt>slt</tt>: signed less than</li>
5529 <li><tt>sle</tt>: signed less or equal</li>
5532 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
5533 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5534 typed. They must also be identical types.</p>
5537 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5538 condition code given as <tt>cond</tt>. The comparison performed always yields
5539 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
5540 result, as follows:</p>
5543 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
5544 <tt>false</tt> otherwise. No sign interpretation is necessary or
5547 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
5548 <tt>false</tt> otherwise. No sign interpretation is necessary or
5551 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
5552 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5554 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
5555 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5556 to <tt>op2</tt>.</li>
5558 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
5559 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5561 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
5562 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5564 <li><tt>sgt</tt>: interprets the operands as signed values and yields
5565 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5567 <li><tt>sge</tt>: interprets the operands as signed values and yields
5568 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5569 to <tt>op2</tt>.</li>
5571 <li><tt>slt</tt>: interprets the operands as signed values and yields
5572 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5574 <li><tt>sle</tt>: interprets the operands as signed values and yields
5575 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5578 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
5579 values are compared as if they were integers.</p>
5581 <p>If the operands are integer vectors, then they are compared element by
5582 element. The result is an <tt>i1</tt> vector with the same number of elements
5583 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
5587 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
5588 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
5589 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
5590 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
5591 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
5592 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
5595 <p>Note that the code generator does not yet support vector types with
5596 the <tt>icmp</tt> instruction.</p>
5600 <!-- _______________________________________________________________________ -->
5602 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
5609 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5613 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
5614 values based on comparison of its operands.</p>
5616 <p>If the operands are floating point scalars, then the result type is a boolean
5617 (<a href="#t_integer"><tt>i1</tt></a>).</p>
5619 <p>If the operands are floating point vectors, then the result type is a vector
5620 of boolean with the same number of elements as the operands being
5624 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
5625 the condition code indicating the kind of comparison to perform. It is not a
5626 value, just a keyword. The possible condition code are:</p>
5629 <li><tt>false</tt>: no comparison, always returns false</li>
5630 <li><tt>oeq</tt>: ordered and equal</li>
5631 <li><tt>ogt</tt>: ordered and greater than </li>
5632 <li><tt>oge</tt>: ordered and greater than or equal</li>
5633 <li><tt>olt</tt>: ordered and less than </li>
5634 <li><tt>ole</tt>: ordered and less than or equal</li>
5635 <li><tt>one</tt>: ordered and not equal</li>
5636 <li><tt>ord</tt>: ordered (no nans)</li>
5637 <li><tt>ueq</tt>: unordered or equal</li>
5638 <li><tt>ugt</tt>: unordered or greater than </li>
5639 <li><tt>uge</tt>: unordered or greater than or equal</li>
5640 <li><tt>ult</tt>: unordered or less than </li>
5641 <li><tt>ule</tt>: unordered or less than or equal</li>
5642 <li><tt>une</tt>: unordered or not equal</li>
5643 <li><tt>uno</tt>: unordered (either nans)</li>
5644 <li><tt>true</tt>: no comparison, always returns true</li>
5647 <p><i>Ordered</i> means that neither operand is a QNAN while
5648 <i>unordered</i> means that either operand may be a QNAN.</p>
5650 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
5651 a <a href="#t_floating">floating point</a> type or
5652 a <a href="#t_vector">vector</a> of floating point type. They must have
5653 identical types.</p>
5656 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
5657 according to the condition code given as <tt>cond</tt>. If the operands are
5658 vectors, then the vectors are compared element by element. Each comparison
5659 performed always yields an <a href="#t_integer">i1</a> result, as
5663 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
5665 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5666 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5668 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5669 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5671 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5672 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5674 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5675 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5677 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5678 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5680 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5681 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5683 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
5685 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
5686 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5688 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
5689 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5691 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
5692 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5694 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
5695 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5697 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
5698 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5700 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
5701 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5703 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
5705 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5710 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
5711 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5712 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5713 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
5716 <p>Note that the code generator does not yet support vector types with
5717 the <tt>fcmp</tt> instruction.</p>
5721 <!-- _______________________________________________________________________ -->
5723 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
5730 <result> = phi <ty> [ <val0>, <label0>], ...
5734 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
5735 SSA graph representing the function.</p>
5738 <p>The type of the incoming values is specified with the first type field. After
5739 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5740 one pair for each predecessor basic block of the current block. Only values
5741 of <a href="#t_firstclass">first class</a> type may be used as the value
5742 arguments to the PHI node. Only labels may be used as the label
5745 <p>There must be no non-phi instructions between the start of a basic block and
5746 the PHI instructions: i.e. PHI instructions must be first in a basic
5749 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5750 occur on the edge from the corresponding predecessor block to the current
5751 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5752 value on the same edge).</p>
5755 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
5756 specified by the pair corresponding to the predecessor basic block that
5757 executed just prior to the current block.</p>
5761 Loop: ; Infinite loop that counts from 0 on up...
5762 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5763 %nextindvar = add i32 %indvar, 1
5769 <!-- _______________________________________________________________________ -->
5771 <a name="i_select">'<tt>select</tt>' Instruction</a>
5778 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5780 <i>selty</i> is either i1 or {<N x i1>}
5784 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
5785 condition, without branching.</p>
5789 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5790 values indicating the condition, and two values of the
5791 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
5792 vectors and the condition is a scalar, then entire vectors are selected, not
5793 individual elements.</p>
5796 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
5797 first value argument; otherwise, it returns the second value argument.</p>
5799 <p>If the condition is a vector of i1, then the value arguments must be vectors
5800 of the same size, and the selection is done element by element.</p>
5804 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
5807 <p>Note that the code generator does not yet support conditions
5808 with vector type.</p>
5812 <!-- _______________________________________________________________________ -->
5814 <a name="i_call">'<tt>call</tt>' Instruction</a>
5821 <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>]
5825 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
5828 <p>This instruction requires several arguments:</p>
5831 <li>The optional "tail" marker indicates that the callee function does not
5832 access any allocas or varargs in the caller. Note that calls may be
5833 marked "tail" even if they do not occur before
5834 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
5835 present, the function call is eligible for tail call optimization,
5836 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
5837 optimized into a jump</a>. The code generator may optimize calls marked
5838 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
5839 sibling call optimization</a> when the caller and callee have
5840 matching signatures, or 2) forced tail call optimization when the
5841 following extra requirements are met:
5843 <li>Caller and callee both have the calling
5844 convention <tt>fastcc</tt>.</li>
5845 <li>The call is in tail position (ret immediately follows call and ret
5846 uses value of call or is void).</li>
5847 <li>Option <tt>-tailcallopt</tt> is enabled,
5848 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
5849 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
5850 constraints are met.</a></li>
5854 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5855 convention</a> the call should use. If none is specified, the call
5856 defaults to using C calling conventions. The calling convention of the
5857 call must match the calling convention of the target function, or else the
5858 behavior is undefined.</li>
5860 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5861 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5862 '<tt>inreg</tt>' attributes are valid here.</li>
5864 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5865 type of the return value. Functions that return no value are marked
5866 <tt><a href="#t_void">void</a></tt>.</li>
5868 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5869 being invoked. The argument types must match the types implied by this
5870 signature. This type can be omitted if the function is not varargs and if
5871 the function type does not return a pointer to a function.</li>
5873 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5874 be invoked. In most cases, this is a direct function invocation, but
5875 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5876 to function value.</li>
5878 <li>'<tt>function args</tt>': argument list whose types match the function
5879 signature argument types and parameter attributes. All arguments must be
5880 of <a href="#t_firstclass">first class</a> type. If the function
5881 signature indicates the function accepts a variable number of arguments,
5882 the extra arguments can be specified.</li>
5884 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5885 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5886 '<tt>readnone</tt>' attributes are valid here.</li>
5890 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5891 a specified function, with its incoming arguments bound to the specified
5892 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5893 function, control flow continues with the instruction after the function
5894 call, and the return value of the function is bound to the result
5899 %retval = call i32 @test(i32 %argc)
5900 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
5901 %X = tail call i32 @foo() <i>; yields i32</i>
5902 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5903 call void %foo(i8 97 signext)
5905 %struct.A = type { i32, i8 }
5906 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
5907 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5908 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
5909 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
5910 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
5913 <p>llvm treats calls to some functions with names and arguments that match the
5914 standard C99 library as being the C99 library functions, and may perform
5915 optimizations or generate code for them under that assumption. This is
5916 something we'd like to change in the future to provide better support for
5917 freestanding environments and non-C-based languages.</p>
5921 <!-- _______________________________________________________________________ -->
5923 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5930 <resultval> = va_arg <va_list*> <arglist>, <argty>
5934 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
5935 the "variable argument" area of a function call. It is used to implement the
5936 <tt>va_arg</tt> macro in C.</p>
5939 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
5940 argument. It returns a value of the specified argument type and increments
5941 the <tt>va_list</tt> to point to the next argument. The actual type
5942 of <tt>va_list</tt> is target specific.</p>
5945 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
5946 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
5947 to the next argument. For more information, see the variable argument
5948 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
5950 <p>It is legal for this instruction to be called in a function which does not
5951 take a variable number of arguments, for example, the <tt>vfprintf</tt>
5954 <p><tt>va_arg</tt> is an LLVM instruction instead of
5955 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
5959 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
5961 <p>Note that the code generator does not yet fully support va_arg on many
5962 targets. Also, it does not currently support va_arg with aggregate types on
5971 <!-- *********************************************************************** -->
5972 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
5973 <!-- *********************************************************************** -->
5977 <p>LLVM supports the notion of an "intrinsic function". These functions have
5978 well known names and semantics and are required to follow certain
5979 restrictions. Overall, these intrinsics represent an extension mechanism for
5980 the LLVM language that does not require changing all of the transformations
5981 in LLVM when adding to the language (or the bitcode reader/writer, the
5982 parser, etc...).</p>
5984 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
5985 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5986 begin with this prefix. Intrinsic functions must always be external
5987 functions: you cannot define the body of intrinsic functions. Intrinsic
5988 functions may only be used in call or invoke instructions: it is illegal to
5989 take the address of an intrinsic function. Additionally, because intrinsic
5990 functions are part of the LLVM language, it is required if any are added that
5991 they be documented here.</p>
5993 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5994 family of functions that perform the same operation but on different data
5995 types. Because LLVM can represent over 8 million different integer types,
5996 overloading is used commonly to allow an intrinsic function to operate on any
5997 integer type. One or more of the argument types or the result type can be
5998 overloaded to accept any integer type. Argument types may also be defined as
5999 exactly matching a previous argument's type or the result type. This allows
6000 an intrinsic function which accepts multiple arguments, but needs all of them
6001 to be of the same type, to only be overloaded with respect to a single
6002 argument or the result.</p>
6004 <p>Overloaded intrinsics will have the names of its overloaded argument types
6005 encoded into its function name, each preceded by a period. Only those types
6006 which are overloaded result in a name suffix. Arguments whose type is matched
6007 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6008 can take an integer of any width and returns an integer of exactly the same
6009 integer width. This leads to a family of functions such as
6010 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6011 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6012 suffix is required. Because the argument's type is matched against the return
6013 type, it does not require its own name suffix.</p>
6015 <p>To learn how to add an intrinsic function, please see the
6016 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6018 <!-- ======================================================================= -->
6020 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6025 <p>Variable argument support is defined in LLVM with
6026 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6027 intrinsic functions. These functions are related to the similarly named
6028 macros defined in the <tt><stdarg.h></tt> header file.</p>
6030 <p>All of these functions operate on arguments that use a target-specific value
6031 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6032 not define what this type is, so all transformations should be prepared to
6033 handle these functions regardless of the type used.</p>
6035 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6036 instruction and the variable argument handling intrinsic functions are
6039 <pre class="doc_code">
6040 define i32 @test(i32 %X, ...) {
6041 ; Initialize variable argument processing
6043 %ap2 = bitcast i8** %ap to i8*
6044 call void @llvm.va_start(i8* %ap2)
6046 ; Read a single integer argument
6047 %tmp = va_arg i8** %ap, i32
6049 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6051 %aq2 = bitcast i8** %aq to i8*
6052 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6053 call void @llvm.va_end(i8* %aq2)
6055 ; Stop processing of arguments.
6056 call void @llvm.va_end(i8* %ap2)
6060 declare void @llvm.va_start(i8*)
6061 declare void @llvm.va_copy(i8*, i8*)
6062 declare void @llvm.va_end(i8*)
6065 <!-- _______________________________________________________________________ -->
6067 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6075 declare void %llvm.va_start(i8* <arglist>)
6079 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6080 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6083 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6086 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6087 macro available in C. In a target-dependent way, it initializes
6088 the <tt>va_list</tt> element to which the argument points, so that the next
6089 call to <tt>va_arg</tt> will produce the first variable argument passed to
6090 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6091 need to know the last argument of the function as the compiler can figure
6096 <!-- _______________________________________________________________________ -->
6098 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6105 declare void @llvm.va_end(i8* <arglist>)
6109 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6110 which has been initialized previously
6111 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6112 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6115 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6118 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6119 macro available in C. In a target-dependent way, it destroys
6120 the <tt>va_list</tt> element to which the argument points. Calls
6121 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6122 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6123 with calls to <tt>llvm.va_end</tt>.</p>
6127 <!-- _______________________________________________________________________ -->
6129 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6136 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6140 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6141 from the source argument list to the destination argument list.</p>
6144 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6145 The second argument is a pointer to a <tt>va_list</tt> element to copy
6149 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6150 macro available in C. In a target-dependent way, it copies the
6151 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6152 element. This intrinsic is necessary because
6153 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6154 arbitrarily complex and require, for example, memory allocation.</p>
6160 <!-- ======================================================================= -->
6162 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6167 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6168 Collection</a> (GC) requires the implementation and generation of these
6169 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6170 roots on the stack</a>, as well as garbage collector implementations that
6171 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6172 barriers. Front-ends for type-safe garbage collected languages should generate
6173 these intrinsics to make use of the LLVM garbage collectors. For more details,
6174 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6177 <p>The garbage collection intrinsics only operate on objects in the generic
6178 address space (address space zero).</p>
6180 <!-- _______________________________________________________________________ -->
6182 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6189 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6193 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6194 the code generator, and allows some metadata to be associated with it.</p>
6197 <p>The first argument specifies the address of a stack object that contains the
6198 root pointer. The second pointer (which must be either a constant or a
6199 global value address) contains the meta-data to be associated with the
6203 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6204 location. At compile-time, the code generator generates information to allow
6205 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6206 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6211 <!-- _______________________________________________________________________ -->
6213 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6220 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6224 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6225 locations, allowing garbage collector implementations that require read
6229 <p>The second argument is the address to read from, which should be an address
6230 allocated from the garbage collector. The first object is a pointer to the
6231 start of the referenced object, if needed by the language runtime (otherwise
6235 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6236 instruction, but may be replaced with substantially more complex code by the
6237 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6238 may only be used in a function which <a href="#gc">specifies a GC
6243 <!-- _______________________________________________________________________ -->
6245 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6252 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6256 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6257 locations, allowing garbage collector implementations that require write
6258 barriers (such as generational or reference counting collectors).</p>
6261 <p>The first argument is the reference to store, the second is the start of the
6262 object to store it to, and the third is the address of the field of Obj to
6263 store to. If the runtime does not require a pointer to the object, Obj may
6267 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6268 instruction, but may be replaced with substantially more complex code by the
6269 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6270 may only be used in a function which <a href="#gc">specifies a GC
6277 <!-- ======================================================================= -->
6279 <a name="int_codegen">Code Generator Intrinsics</a>
6284 <p>These intrinsics are provided by LLVM to expose special features that may
6285 only be implemented with code generator support.</p>
6287 <!-- _______________________________________________________________________ -->
6289 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6296 declare i8 *@llvm.returnaddress(i32 <level>)
6300 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6301 target-specific value indicating the return address of the current function
6302 or one of its callers.</p>
6305 <p>The argument to this intrinsic indicates which function to return the address
6306 for. Zero indicates the calling function, one indicates its caller, etc.
6307 The argument is <b>required</b> to be a constant integer value.</p>
6310 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6311 indicating the return address of the specified call frame, or zero if it
6312 cannot be identified. The value returned by this intrinsic is likely to be
6313 incorrect or 0 for arguments other than zero, so it should only be used for
6314 debugging purposes.</p>
6316 <p>Note that calling this intrinsic does not prevent function inlining or other
6317 aggressive transformations, so the value returned may not be that of the
6318 obvious source-language caller.</p>
6322 <!-- _______________________________________________________________________ -->
6324 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6331 declare i8* @llvm.frameaddress(i32 <level>)
6335 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6336 target-specific frame pointer value for the specified stack frame.</p>
6339 <p>The argument to this intrinsic indicates which function to return the frame
6340 pointer for. Zero indicates the calling function, one indicates its caller,
6341 etc. The argument is <b>required</b> to be a constant integer value.</p>
6344 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6345 indicating the frame address of the specified call frame, or zero if it
6346 cannot be identified. The value returned by this intrinsic is likely to be
6347 incorrect or 0 for arguments other than zero, so it should only be used for
6348 debugging purposes.</p>
6350 <p>Note that calling this intrinsic does not prevent function inlining or other
6351 aggressive transformations, so the value returned may not be that of the
6352 obvious source-language caller.</p>
6356 <!-- _______________________________________________________________________ -->
6358 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6365 declare i8* @llvm.stacksave()
6369 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6370 of the function stack, for use
6371 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6372 useful for implementing language features like scoped automatic variable
6373 sized arrays in C99.</p>
6376 <p>This intrinsic returns a opaque pointer value that can be passed
6377 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6378 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6379 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6380 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6381 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6382 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6386 <!-- _______________________________________________________________________ -->
6388 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6395 declare void @llvm.stackrestore(i8* %ptr)
6399 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6400 the function stack to the state it was in when the
6401 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6402 executed. This is useful for implementing language features like scoped
6403 automatic variable sized arrays in C99.</p>
6406 <p>See the description
6407 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6411 <!-- _______________________________________________________________________ -->
6413 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6420 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6424 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6425 insert a prefetch instruction if supported; otherwise, it is a noop.
6426 Prefetches have no effect on the behavior of the program but can change its
6427 performance characteristics.</p>
6430 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6431 specifier determining if the fetch should be for a read (0) or write (1),
6432 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6433 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6434 specifies whether the prefetch is performed on the data (1) or instruction (0)
6435 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6436 must be constant integers.</p>
6439 <p>This intrinsic does not modify the behavior of the program. In particular,
6440 prefetches cannot trap and do not produce a value. On targets that support
6441 this intrinsic, the prefetch can provide hints to the processor cache for
6442 better performance.</p>
6446 <!-- _______________________________________________________________________ -->
6448 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
6455 declare void @llvm.pcmarker(i32 <id>)
6459 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
6460 Counter (PC) in a region of code to simulators and other tools. The method
6461 is target specific, but it is expected that the marker will use exported
6462 symbols to transmit the PC of the marker. The marker makes no guarantees
6463 that it will remain with any specific instruction after optimizations. It is
6464 possible that the presence of a marker will inhibit optimizations. The
6465 intended use is to be inserted after optimizations to allow correlations of
6466 simulation runs.</p>
6469 <p><tt>id</tt> is a numerical id identifying the marker.</p>
6472 <p>This intrinsic does not modify the behavior of the program. Backends that do
6473 not support this intrinsic may ignore it.</p>
6477 <!-- _______________________________________________________________________ -->
6479 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
6486 declare i64 @llvm.readcyclecounter()
6490 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
6491 counter register (or similar low latency, high accuracy clocks) on those
6492 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6493 should map to RPCC. As the backing counters overflow quickly (on the order
6494 of 9 seconds on alpha), this should only be used for small timings.</p>
6497 <p>When directly supported, reading the cycle counter should not modify any
6498 memory. Implementations are allowed to either return a application specific
6499 value or a system wide value. On backends without support, this is lowered
6500 to a constant 0.</p>
6506 <!-- ======================================================================= -->
6508 <a name="int_libc">Standard C Library Intrinsics</a>
6513 <p>LLVM provides intrinsics for a few important standard C library functions.
6514 These intrinsics allow source-language front-ends to pass information about
6515 the alignment of the pointer arguments to the code generator, providing
6516 opportunity for more efficient code generation.</p>
6518 <!-- _______________________________________________________________________ -->
6520 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
6526 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
6527 integer bit width and for different address spaces. Not all targets support
6528 all bit widths however.</p>
6531 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6532 i32 <len>, i32 <align>, i1 <isvolatile>)
6533 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6534 i64 <len>, i32 <align>, i1 <isvolatile>)
6538 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6539 source location to the destination location.</p>
6541 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
6542 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6543 and the pointers can be in specified address spaces.</p>
6547 <p>The first argument is a pointer to the destination, the second is a pointer
6548 to the source. The third argument is an integer argument specifying the
6549 number of bytes to copy, the fourth argument is the alignment of the
6550 source and destination locations, and the fifth is a boolean indicating a
6551 volatile access.</p>
6553 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6554 then the caller guarantees that both the source and destination pointers are
6555 aligned to that boundary.</p>
6557 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6558 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
6559 The detailed access behavior is not very cleanly specified and it is unwise
6560 to depend on it.</p>
6564 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6565 source location to the destination location, which are not allowed to
6566 overlap. It copies "len" bytes of memory over. If the argument is known to
6567 be aligned to some boundary, this can be specified as the fourth argument,
6568 otherwise it should be set to 0 or 1.</p>
6572 <!-- _______________________________________________________________________ -->
6574 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
6580 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6581 width and for different address space. Not all targets support all bit
6585 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6586 i32 <len>, i32 <align>, i1 <isvolatile>)
6587 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6588 i64 <len>, i32 <align>, i1 <isvolatile>)
6592 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
6593 source location to the destination location. It is similar to the
6594 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
6597 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
6598 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6599 and the pointers can be in specified address spaces.</p>
6603 <p>The first argument is a pointer to the destination, the second is a pointer
6604 to the source. The third argument is an integer argument specifying the
6605 number of bytes to copy, the fourth argument is the alignment of the
6606 source and destination locations, and the fifth is a boolean indicating a
6607 volatile access.</p>
6609 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6610 then the caller guarantees that the source and destination pointers are
6611 aligned to that boundary.</p>
6613 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6614 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
6615 The detailed access behavior is not very cleanly specified and it is unwise
6616 to depend on it.</p>
6620 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
6621 source location to the destination location, which may overlap. It copies
6622 "len" bytes of memory over. If the argument is known to be aligned to some
6623 boundary, this can be specified as the fourth argument, otherwise it should
6624 be set to 0 or 1.</p>
6628 <!-- _______________________________________________________________________ -->
6630 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
6636 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6637 width and for different address spaces. However, not all targets support all
6641 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6642 i32 <len>, i32 <align>, i1 <isvolatile>)
6643 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6644 i64 <len>, i32 <align>, i1 <isvolatile>)
6648 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
6649 particular byte value.</p>
6651 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
6652 intrinsic does not return a value and takes extra alignment/volatile
6653 arguments. Also, the destination can be in an arbitrary address space.</p>
6656 <p>The first argument is a pointer to the destination to fill, the second is the
6657 byte value with which to fill it, the third argument is an integer argument
6658 specifying the number of bytes to fill, and the fourth argument is the known
6659 alignment of the destination location.</p>
6661 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6662 then the caller guarantees that the destination pointer is aligned to that
6665 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6666 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
6667 The detailed access behavior is not very cleanly specified and it is unwise
6668 to depend on it.</p>
6671 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
6672 at the destination location. If the argument is known to be aligned to some
6673 boundary, this can be specified as the fourth argument, otherwise it should
6674 be set to 0 or 1.</p>
6678 <!-- _______________________________________________________________________ -->
6680 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
6686 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
6687 floating point or vector of floating point type. Not all targets support all
6691 declare float @llvm.sqrt.f32(float %Val)
6692 declare double @llvm.sqrt.f64(double %Val)
6693 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6694 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6695 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6699 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
6700 returning the same value as the libm '<tt>sqrt</tt>' functions would.
6701 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
6702 behavior for negative numbers other than -0.0 (which allows for better
6703 optimization, because there is no need to worry about errno being
6704 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
6707 <p>The argument and return value are floating point numbers of the same
6711 <p>This function returns the sqrt of the specified operand if it is a
6712 nonnegative floating point number.</p>
6716 <!-- _______________________________________________________________________ -->
6718 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
6724 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
6725 floating point or vector of floating point type. Not all targets support all
6729 declare float @llvm.powi.f32(float %Val, i32 %power)
6730 declare double @llvm.powi.f64(double %Val, i32 %power)
6731 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6732 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6733 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6737 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
6738 specified (positive or negative) power. The order of evaluation of
6739 multiplications is not defined. When a vector of floating point type is
6740 used, the second argument remains a scalar integer value.</p>
6743 <p>The second argument is an integer power, and the first is a value to raise to
6747 <p>This function returns the first value raised to the second power with an
6748 unspecified sequence of rounding operations.</p>
6752 <!-- _______________________________________________________________________ -->
6754 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
6760 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
6761 floating point or vector of floating point type. Not all targets support all
6765 declare float @llvm.sin.f32(float %Val)
6766 declare double @llvm.sin.f64(double %Val)
6767 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6768 declare fp128 @llvm.sin.f128(fp128 %Val)
6769 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6773 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
6776 <p>The argument and return value are floating point numbers of the same
6780 <p>This function returns the sine of the specified operand, returning the same
6781 values as the libm <tt>sin</tt> functions would, and handles error conditions
6782 in the same way.</p>
6786 <!-- _______________________________________________________________________ -->
6788 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
6794 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
6795 floating point or vector of floating point type. Not all targets support all
6799 declare float @llvm.cos.f32(float %Val)
6800 declare double @llvm.cos.f64(double %Val)
6801 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6802 declare fp128 @llvm.cos.f128(fp128 %Val)
6803 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6807 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
6810 <p>The argument and return value are floating point numbers of the same
6814 <p>This function returns the cosine of the specified operand, returning the same
6815 values as the libm <tt>cos</tt> functions would, and handles error conditions
6816 in the same way.</p>
6820 <!-- _______________________________________________________________________ -->
6822 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
6828 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
6829 floating point or vector of floating point type. Not all targets support all
6833 declare float @llvm.pow.f32(float %Val, float %Power)
6834 declare double @llvm.pow.f64(double %Val, double %Power)
6835 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6836 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6837 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6841 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
6842 specified (positive or negative) power.</p>
6845 <p>The second argument is a floating point power, and the first is a value to
6846 raise to that power.</p>
6849 <p>This function returns the first value raised to the second power, returning
6850 the same values as the libm <tt>pow</tt> functions would, and handles error
6851 conditions in the same way.</p>
6857 <!-- _______________________________________________________________________ -->
6859 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
6865 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
6866 floating point or vector of floating point type. Not all targets support all
6870 declare float @llvm.exp.f32(float %Val)
6871 declare double @llvm.exp.f64(double %Val)
6872 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
6873 declare fp128 @llvm.exp.f128(fp128 %Val)
6874 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
6878 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
6881 <p>The argument and return value are floating point numbers of the same
6885 <p>This function returns the same values as the libm <tt>exp</tt> functions
6886 would, and handles error conditions in the same way.</p>
6890 <!-- _______________________________________________________________________ -->
6892 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
6898 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
6899 floating point or vector of floating point type. Not all targets support all
6903 declare float @llvm.log.f32(float %Val)
6904 declare double @llvm.log.f64(double %Val)
6905 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
6906 declare fp128 @llvm.log.f128(fp128 %Val)
6907 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
6911 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
6914 <p>The argument and return value are floating point numbers of the same
6918 <p>This function returns the same values as the libm <tt>log</tt> functions
6919 would, and handles error conditions in the same way.</p>
6922 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
6928 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
6929 floating point or vector of floating point type. Not all targets support all
6933 declare float @llvm.fma.f32(float %a, float %b, float %c)
6934 declare double @llvm.fma.f64(double %a, double %b, double %c)
6935 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
6936 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
6937 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
6941 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
6945 <p>The argument and return value are floating point numbers of the same
6949 <p>This function returns the same values as the libm <tt>fma</tt> functions
6954 <!-- ======================================================================= -->
6956 <a name="int_manip">Bit Manipulation Intrinsics</a>
6961 <p>LLVM provides intrinsics for a few important bit manipulation operations.
6962 These allow efficient code generation for some algorithms.</p>
6964 <!-- _______________________________________________________________________ -->
6966 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
6972 <p>This is an overloaded intrinsic function. You can use bswap on any integer
6973 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
6976 declare i16 @llvm.bswap.i16(i16 <id>)
6977 declare i32 @llvm.bswap.i32(i32 <id>)
6978 declare i64 @llvm.bswap.i64(i64 <id>)
6982 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
6983 values with an even number of bytes (positive multiple of 16 bits). These
6984 are useful for performing operations on data that is not in the target's
6985 native byte order.</p>
6988 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
6989 and low byte of the input i16 swapped. Similarly,
6990 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
6991 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
6992 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
6993 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
6994 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
6995 more, respectively).</p>
6999 <!-- _______________________________________________________________________ -->
7001 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7007 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7008 width, or on any vector with integer elements. Not all targets support all
7009 bit widths or vector types, however.</p>
7012 declare i8 @llvm.ctpop.i8(i8 <src>)
7013 declare i16 @llvm.ctpop.i16(i16 <src>)
7014 declare i32 @llvm.ctpop.i32(i32 <src>)
7015 declare i64 @llvm.ctpop.i64(i64 <src>)
7016 declare i256 @llvm.ctpop.i256(i256 <src>)
7017 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7021 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7025 <p>The only argument is the value to be counted. The argument may be of any
7026 integer type, or a vector with integer elements.
7027 The return type must match the argument type.</p>
7030 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7031 element of a vector.</p>
7035 <!-- _______________________________________________________________________ -->
7037 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7043 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7044 integer bit width, or any vector whose elements are integers. Not all
7045 targets support all bit widths or vector types, however.</p>
7048 declare i8 @llvm.ctlz.i8 (i8 <src>)
7049 declare i16 @llvm.ctlz.i16(i16 <src>)
7050 declare i32 @llvm.ctlz.i32(i32 <src>)
7051 declare i64 @llvm.ctlz.i64(i64 <src>)
7052 declare i256 @llvm.ctlz.i256(i256 <src>)
7053 declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src;gt)
7057 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7058 leading zeros in a variable.</p>
7061 <p>The only argument is the value to be counted. The argument may be of any
7062 integer type, or any vector type with integer element type.
7063 The return type must match the argument type.</p>
7066 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7067 zeros in a variable, or within each element of the vector if the operation
7068 is of vector type. If the src == 0 then the result is the size in bits of
7069 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7073 <!-- _______________________________________________________________________ -->
7075 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7081 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7082 integer bit width, or any vector of integer elements. Not all targets
7083 support all bit widths or vector types, however.</p>
7086 declare i8 @llvm.cttz.i8 (i8 <src>)
7087 declare i16 @llvm.cttz.i16(i16 <src>)
7088 declare i32 @llvm.cttz.i32(i32 <src>)
7089 declare i64 @llvm.cttz.i64(i64 <src>)
7090 declare i256 @llvm.cttz.i256(i256 <src>)
7091 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>)
7095 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7099 <p>The only argument is the value to be counted. The argument may be of any
7100 integer type, or a vectory with integer element type.. The return type
7101 must match the argument type.</p>
7104 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7105 zeros in a variable, or within each element of a vector.
7106 If the src == 0 then the result is the size in bits of
7107 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7113 <!-- ======================================================================= -->
7115 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7120 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7122 <!-- _______________________________________________________________________ -->
7124 <a name="int_sadd_overflow">
7125 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7132 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7133 on any integer bit width.</p>
7136 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7137 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7138 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7142 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7143 a signed addition of the two arguments, and indicate whether an overflow
7144 occurred during the signed summation.</p>
7147 <p>The arguments (%a and %b) and the first element of the result structure may
7148 be of integer types of any bit width, but they must have the same bit
7149 width. The second element of the result structure must be of
7150 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7151 undergo signed addition.</p>
7154 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7155 a signed addition of the two variables. They return a structure — the
7156 first element of which is the signed summation, and the second element of
7157 which is a bit specifying if the signed summation resulted in an
7162 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7163 %sum = extractvalue {i32, i1} %res, 0
7164 %obit = extractvalue {i32, i1} %res, 1
7165 br i1 %obit, label %overflow, label %normal
7170 <!-- _______________________________________________________________________ -->
7172 <a name="int_uadd_overflow">
7173 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7180 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7181 on any integer bit width.</p>
7184 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7185 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7186 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7190 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7191 an unsigned addition of the two arguments, and indicate whether a carry
7192 occurred during the unsigned summation.</p>
7195 <p>The arguments (%a and %b) and the first element of the result structure may
7196 be of integer types of any bit width, but they must have the same bit
7197 width. The second element of the result structure must be of
7198 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7199 undergo unsigned addition.</p>
7202 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7203 an unsigned addition of the two arguments. They return a structure —
7204 the first element of which is the sum, and the second element of which is a
7205 bit specifying if the unsigned summation resulted in a carry.</p>
7209 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7210 %sum = extractvalue {i32, i1} %res, 0
7211 %obit = extractvalue {i32, i1} %res, 1
7212 br i1 %obit, label %carry, label %normal
7217 <!-- _______________________________________________________________________ -->
7219 <a name="int_ssub_overflow">
7220 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7227 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7228 on any integer bit width.</p>
7231 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7232 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7233 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7237 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7238 a signed subtraction of the two arguments, and indicate whether an overflow
7239 occurred during the signed subtraction.</p>
7242 <p>The arguments (%a and %b) and the first element of the result structure may
7243 be of integer types of any bit width, but they must have the same bit
7244 width. The second element of the result structure must be of
7245 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7246 undergo signed subtraction.</p>
7249 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7250 a signed subtraction of the two arguments. They return a structure —
7251 the first element of which is the subtraction, and the second element of
7252 which is a bit specifying if the signed subtraction resulted in an
7257 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7258 %sum = extractvalue {i32, i1} %res, 0
7259 %obit = extractvalue {i32, i1} %res, 1
7260 br i1 %obit, label %overflow, label %normal
7265 <!-- _______________________________________________________________________ -->
7267 <a name="int_usub_overflow">
7268 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7275 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7276 on any integer bit width.</p>
7279 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7280 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7281 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7285 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7286 an unsigned subtraction of the two arguments, and indicate whether an
7287 overflow occurred during the unsigned subtraction.</p>
7290 <p>The arguments (%a and %b) and the first element of the result structure may
7291 be of integer types of any bit width, but they must have the same bit
7292 width. The second element of the result structure must be of
7293 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7294 undergo unsigned subtraction.</p>
7297 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7298 an unsigned subtraction of the two arguments. They return a structure —
7299 the first element of which is the subtraction, and the second element of
7300 which is a bit specifying if the unsigned subtraction resulted in an
7305 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7306 %sum = extractvalue {i32, i1} %res, 0
7307 %obit = extractvalue {i32, i1} %res, 1
7308 br i1 %obit, label %overflow, label %normal
7313 <!-- _______________________________________________________________________ -->
7315 <a name="int_smul_overflow">
7316 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7323 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7324 on any integer bit width.</p>
7327 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7328 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7329 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7334 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7335 a signed multiplication of the two arguments, and indicate whether an
7336 overflow occurred during the signed multiplication.</p>
7339 <p>The arguments (%a and %b) and the first element of the result structure may
7340 be of integer types of any bit width, but they must have the same bit
7341 width. The second element of the result structure must be of
7342 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7343 undergo signed multiplication.</p>
7346 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7347 a signed multiplication of the two arguments. They return a structure —
7348 the first element of which is the multiplication, and the second element of
7349 which is a bit specifying if the signed multiplication resulted in an
7354 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7355 %sum = extractvalue {i32, i1} %res, 0
7356 %obit = extractvalue {i32, i1} %res, 1
7357 br i1 %obit, label %overflow, label %normal
7362 <!-- _______________________________________________________________________ -->
7364 <a name="int_umul_overflow">
7365 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7372 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
7373 on any integer bit width.</p>
7376 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7377 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7378 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7382 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7383 a unsigned multiplication of the two arguments, and indicate whether an
7384 overflow occurred during the unsigned multiplication.</p>
7387 <p>The arguments (%a and %b) and the first element of the result structure may
7388 be of integer types of any bit width, but they must have the same bit
7389 width. The second element of the result structure must be of
7390 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7391 undergo unsigned multiplication.</p>
7394 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7395 an unsigned multiplication of the two arguments. They return a structure
7396 — the first element of which is the multiplication, and the second
7397 element of which is a bit specifying if the unsigned multiplication resulted
7402 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7403 %sum = extractvalue {i32, i1} %res, 0
7404 %obit = extractvalue {i32, i1} %res, 1
7405 br i1 %obit, label %overflow, label %normal
7412 <!-- ======================================================================= -->
7414 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
7419 <p>Half precision floating point is a storage-only format. This means that it is
7420 a dense encoding (in memory) but does not support computation in the
7423 <p>This means that code must first load the half-precision floating point
7424 value as an i16, then convert it to float with <a
7425 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
7426 Computation can then be performed on the float value (including extending to
7427 double etc). To store the value back to memory, it is first converted to
7428 float if needed, then converted to i16 with
7429 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
7430 storing as an i16 value.</p>
7432 <!-- _______________________________________________________________________ -->
7434 <a name="int_convert_to_fp16">
7435 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
7443 declare i16 @llvm.convert.to.fp16(f32 %a)
7447 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7448 a conversion from single precision floating point format to half precision
7449 floating point format.</p>
7452 <p>The intrinsic function contains single argument - the value to be
7456 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7457 a conversion from single precision floating point format to half precision
7458 floating point format. The return value is an <tt>i16</tt> which
7459 contains the converted number.</p>
7463 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7464 store i16 %res, i16* @x, align 2
7469 <!-- _______________________________________________________________________ -->
7471 <a name="int_convert_from_fp16">
7472 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
7480 declare f32 @llvm.convert.from.fp16(i16 %a)
7484 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
7485 a conversion from half precision floating point format to single precision
7486 floating point format.</p>
7489 <p>The intrinsic function contains single argument - the value to be
7493 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
7494 conversion from half single precision floating point format to single
7495 precision floating point format. The input half-float value is represented by
7496 an <tt>i16</tt> value.</p>
7500 %a = load i16* @x, align 2
7501 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7508 <!-- ======================================================================= -->
7510 <a name="int_debugger">Debugger Intrinsics</a>
7515 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
7516 prefix), are described in
7517 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
7518 Level Debugging</a> document.</p>
7522 <!-- ======================================================================= -->
7524 <a name="int_eh">Exception Handling Intrinsics</a>
7529 <p>The LLVM exception handling intrinsics (which all start with
7530 <tt>llvm.eh.</tt> prefix), are described in
7531 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
7532 Handling</a> document.</p>
7536 <!-- ======================================================================= -->
7538 <a name="int_trampoline">Trampoline Intrinsic</a>
7543 <p>This intrinsic makes it possible to excise one parameter, marked with
7544 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
7545 The result is a callable
7546 function pointer lacking the nest parameter - the caller does not need to
7547 provide a value for it. Instead, the value to use is stored in advance in a
7548 "trampoline", a block of memory usually allocated on the stack, which also
7549 contains code to splice the nest value into the argument list. This is used
7550 to implement the GCC nested function address extension.</p>
7552 <p>For example, if the function is
7553 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
7554 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
7557 <pre class="doc_code">
7558 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7559 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
7560 %p = call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval)
7561 %fp = bitcast i8* %p to i32 (i32, i32)*
7564 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
7565 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
7567 <!-- _______________________________________________________________________ -->
7570 '<tt>llvm.init.trampoline</tt>' Intrinsic
7578 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
7582 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
7583 function pointer suitable for executing it.</p>
7586 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
7587 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
7588 sufficiently aligned block of memory; this memory is written to by the
7589 intrinsic. Note that the size and the alignment are target-specific - LLVM
7590 currently provides no portable way of determining them, so a front-end that
7591 generates this intrinsic needs to have some target-specific knowledge.
7592 The <tt>func</tt> argument must hold a function bitcast to
7593 an <tt>i8*</tt>.</p>
7596 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
7597 dependent code, turning it into a function. A pointer to this function is
7598 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
7599 function pointer type</a> before being called. The new function's signature
7600 is the same as that of <tt>func</tt> with any arguments marked with
7601 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
7602 is allowed, and it must be of pointer type. Calling the new function is
7603 equivalent to calling <tt>func</tt> with the same argument list, but
7604 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
7605 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
7606 by <tt>tramp</tt> is modified, then the effect of any later call to the
7607 returned function pointer is undefined.</p>
7613 <!-- ======================================================================= -->
7615 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
7620 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
7621 hardware constructs for atomic operations and memory synchronization. This
7622 provides an interface to the hardware, not an interface to the programmer. It
7623 is aimed at a low enough level to allow any programming models or APIs
7624 (Application Programming Interfaces) which need atomic behaviors to map
7625 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
7626 hardware provides a "universal IR" for source languages, it also provides a
7627 starting point for developing a "universal" atomic operation and
7628 synchronization IR.</p>
7630 <p>These do <em>not</em> form an API such as high-level threading libraries,
7631 software transaction memory systems, atomic primitives, and intrinsic
7632 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
7633 application libraries. The hardware interface provided by LLVM should allow
7634 a clean implementation of all of these APIs and parallel programming models.
7635 No one model or paradigm should be selected above others unless the hardware
7636 itself ubiquitously does so.</p>
7638 <!-- _______________________________________________________________________ -->
7640 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
7646 declare void @llvm.memory.barrier(i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device>)
7650 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
7651 specific pairs of memory access types.</p>
7654 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
7655 The first four arguments enables a specific barrier as listed below. The
7656 fifth argument specifies that the barrier applies to io or device or uncached
7660 <li><tt>ll</tt>: load-load barrier</li>
7661 <li><tt>ls</tt>: load-store barrier</li>
7662 <li><tt>sl</tt>: store-load barrier</li>
7663 <li><tt>ss</tt>: store-store barrier</li>
7664 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
7668 <p>This intrinsic causes the system to enforce some ordering constraints upon
7669 the loads and stores of the program. This barrier does not
7670 indicate <em>when</em> any events will occur, it only enforces
7671 an <em>order</em> in which they occur. For any of the specified pairs of load
7672 and store operations (f.ex. load-load, or store-load), all of the first
7673 operations preceding the barrier will complete before any of the second
7674 operations succeeding the barrier begin. Specifically the semantics for each
7675 pairing is as follows:</p>
7678 <li><tt>ll</tt>: All loads before the barrier must complete before any load
7679 after the barrier begins.</li>
7680 <li><tt>ls</tt>: All loads before the barrier must complete before any
7681 store after the barrier begins.</li>
7682 <li><tt>ss</tt>: All stores before the barrier must complete before any
7683 store after the barrier begins.</li>
7684 <li><tt>sl</tt>: All stores before the barrier must complete before any
7685 load after the barrier begins.</li>
7688 <p>These semantics are applied with a logical "and" behavior when more than one
7689 is enabled in a single memory barrier intrinsic.</p>
7691 <p>Backends may implement stronger barriers than those requested when they do
7692 not support as fine grained a barrier as requested. Some architectures do
7693 not need all types of barriers and on such architectures, these become
7698 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7699 %ptr = bitcast i8* %mallocP to i32*
7702 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
7703 call void @llvm.memory.barrier(i1 false, i1 true, i1 false, i1 false, i1 true)
7704 <i>; guarantee the above finishes</i>
7705 store i32 8, %ptr <i>; before this begins</i>
7710 <!-- _______________________________________________________________________ -->
7712 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
7718 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
7719 any integer bit width and for different address spaces. Not all targets
7720 support all bit widths however.</p>
7723 declare i8 @llvm.atomic.cmp.swap.i8.p0i8(i8* <ptr>, i8 <cmp>, i8 <val>)
7724 declare i16 @llvm.atomic.cmp.swap.i16.p0i16(i16* <ptr>, i16 <cmp>, i16 <val>)
7725 declare i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* <ptr>, i32 <cmp>, i32 <val>)
7726 declare i64 @llvm.atomic.cmp.swap.i64.p0i64(i64* <ptr>, i64 <cmp>, i64 <val>)
7730 <p>This loads a value in memory and compares it to a given value. If they are
7731 equal, it stores a new value into the memory.</p>
7734 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
7735 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
7736 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
7737 this integer type. While any bit width integer may be used, targets may only
7738 lower representations they support in hardware.</p>
7741 <p>This entire intrinsic must be executed atomically. It first loads the value
7742 in memory pointed to by <tt>ptr</tt> and compares it with the
7743 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
7744 memory. The loaded value is yielded in all cases. This provides the
7745 equivalent of an atomic compare-and-swap operation within the SSA
7750 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7751 %ptr = bitcast i8* %mallocP to i32*
7754 %val1 = add i32 4, 4
7755 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 4, %val1)
7756 <i>; yields {i32}:result1 = 4</i>
7757 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7758 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7760 %val2 = add i32 1, 1
7761 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 5, %val2)
7762 <i>; yields {i32}:result2 = 8</i>
7763 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
7765 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
7770 <!-- _______________________________________________________________________ -->
7772 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
7778 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
7779 integer bit width. Not all targets support all bit widths however.</p>
7782 declare i8 @llvm.atomic.swap.i8.p0i8(i8* <ptr>, i8 <val>)
7783 declare i16 @llvm.atomic.swap.i16.p0i16(i16* <ptr>, i16 <val>)
7784 declare i32 @llvm.atomic.swap.i32.p0i32(i32* <ptr>, i32 <val>)
7785 declare i64 @llvm.atomic.swap.i64.p0i64(i64* <ptr>, i64 <val>)
7789 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
7790 the value from memory. It then stores the value in <tt>val</tt> in the memory
7791 at <tt>ptr</tt>.</p>
7794 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
7795 the <tt>val</tt> argument and the result must be integers of the same bit
7796 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
7797 integer type. The targets may only lower integer representations they
7801 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
7802 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
7803 equivalent of an atomic swap operation within the SSA framework.</p>
7807 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7808 %ptr = bitcast i8* %mallocP to i32*
7811 %val1 = add i32 4, 4
7812 %result1 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val1)
7813 <i>; yields {i32}:result1 = 4</i>
7814 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7815 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7817 %val2 = add i32 1, 1
7818 %result2 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val2)
7819 <i>; yields {i32}:result2 = 8</i>
7821 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
7822 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
7827 <!-- _______________________________________________________________________ -->
7829 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
7835 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
7836 any integer bit width. Not all targets support all bit widths however.</p>
7839 declare i8 @llvm.atomic.load.add.i8.p0i8(i8* <ptr>, i8 <delta>)
7840 declare i16 @llvm.atomic.load.add.i16.p0i16(i16* <ptr>, i16 <delta>)
7841 declare i32 @llvm.atomic.load.add.i32.p0i32(i32* <ptr>, i32 <delta>)
7842 declare i64 @llvm.atomic.load.add.i64.p0i64(i64* <ptr>, i64 <delta>)
7846 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
7847 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7850 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7851 and the second an integer value. The result is also an integer value. These
7852 integer types can have any bit width, but they must all have the same bit
7853 width. The targets may only lower integer representations they support.</p>
7856 <p>This intrinsic does a series of operations atomically. It first loads the
7857 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
7858 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
7862 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7863 %ptr = bitcast i8* %mallocP to i32*
7865 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 4)
7866 <i>; yields {i32}:result1 = 4</i>
7867 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 2)
7868 <i>; yields {i32}:result2 = 8</i>
7869 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 5)
7870 <i>; yields {i32}:result3 = 10</i>
7871 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
7876 <!-- _______________________________________________________________________ -->
7878 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
7884 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
7885 any integer bit width and for different address spaces. Not all targets
7886 support all bit widths however.</p>
7889 declare i8 @llvm.atomic.load.sub.i8.p0i32(i8* <ptr>, i8 <delta>)
7890 declare i16 @llvm.atomic.load.sub.i16.p0i32(i16* <ptr>, i16 <delta>)
7891 declare i32 @llvm.atomic.load.sub.i32.p0i32(i32* <ptr>, i32 <delta>)
7892 declare i64 @llvm.atomic.load.sub.i64.p0i32(i64* <ptr>, i64 <delta>)
7896 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
7897 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7900 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7901 and the second an integer value. The result is also an integer value. These
7902 integer types can have any bit width, but they must all have the same bit
7903 width. The targets may only lower integer representations they support.</p>
7906 <p>This intrinsic does a series of operations atomically. It first loads the
7907 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
7908 result to <tt>ptr</tt>. It yields the original value stored
7909 at <tt>ptr</tt>.</p>
7913 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7914 %ptr = bitcast i8* %mallocP to i32*
7916 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 4)
7917 <i>; yields {i32}:result1 = 8</i>
7918 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 2)
7919 <i>; yields {i32}:result2 = 4</i>
7920 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 5)
7921 <i>; yields {i32}:result3 = 2</i>
7922 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
7927 <!-- _______________________________________________________________________ -->
7929 <a name="int_atomic_load_and">
7930 '<tt>llvm.atomic.load.and.*</tt>' Intrinsic
7933 <a name="int_atomic_load_nand">
7934 '<tt>llvm.atomic.load.nand.*</tt>' Intrinsic
7937 <a name="int_atomic_load_or">
7938 '<tt>llvm.atomic.load.or.*</tt>' Intrinsic
7941 <a name="int_atomic_load_xor">
7942 '<tt>llvm.atomic.load.xor.*</tt>' Intrinsic
7949 <p>These are overloaded intrinsics. You can
7950 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
7951 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
7952 bit width and for different address spaces. Not all targets support all bit
7956 declare i8 @llvm.atomic.load.and.i8.p0i8(i8* <ptr>, i8 <delta>)
7957 declare i16 @llvm.atomic.load.and.i16.p0i16(i16* <ptr>, i16 <delta>)
7958 declare i32 @llvm.atomic.load.and.i32.p0i32(i32* <ptr>, i32 <delta>)
7959 declare i64 @llvm.atomic.load.and.i64.p0i64(i64* <ptr>, i64 <delta>)
7963 declare i8 @llvm.atomic.load.or.i8.p0i8(i8* <ptr>, i8 <delta>)
7964 declare i16 @llvm.atomic.load.or.i16.p0i16(i16* <ptr>, i16 <delta>)
7965 declare i32 @llvm.atomic.load.or.i32.p0i32(i32* <ptr>, i32 <delta>)
7966 declare i64 @llvm.atomic.load.or.i64.p0i64(i64* <ptr>, i64 <delta>)
7970 declare i8 @llvm.atomic.load.nand.i8.p0i32(i8* <ptr>, i8 <delta>)
7971 declare i16 @llvm.atomic.load.nand.i16.p0i32(i16* <ptr>, i16 <delta>)
7972 declare i32 @llvm.atomic.load.nand.i32.p0i32(i32* <ptr>, i32 <delta>)
7973 declare i64 @llvm.atomic.load.nand.i64.p0i32(i64* <ptr>, i64 <delta>)
7977 declare i8 @llvm.atomic.load.xor.i8.p0i32(i8* <ptr>, i8 <delta>)
7978 declare i16 @llvm.atomic.load.xor.i16.p0i32(i16* <ptr>, i16 <delta>)
7979 declare i32 @llvm.atomic.load.xor.i32.p0i32(i32* <ptr>, i32 <delta>)
7980 declare i64 @llvm.atomic.load.xor.i64.p0i32(i64* <ptr>, i64 <delta>)
7984 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
7985 the value stored in memory at <tt>ptr</tt>. It yields the original value
7986 at <tt>ptr</tt>.</p>
7989 <p>These intrinsics take two arguments, the first a pointer to an integer value
7990 and the second an integer value. The result is also an integer value. These
7991 integer types can have any bit width, but they must all have the same bit
7992 width. The targets may only lower integer representations they support.</p>
7995 <p>These intrinsics does a series of operations atomically. They first load the
7996 value stored at <tt>ptr</tt>. They then do the bitwise
7997 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
7998 original value stored at <tt>ptr</tt>.</p>
8002 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8003 %ptr = bitcast i8* %mallocP to i32*
8004 store i32 0x0F0F, %ptr
8005 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32(i32* %ptr, i32 0xFF)
8006 <i>; yields {i32}:result0 = 0x0F0F</i>
8007 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32(i32* %ptr, i32 0xFF)
8008 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
8009 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32(i32* %ptr, i32 0F)
8010 <i>; yields {i32}:result2 = 0xF0</i>
8011 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32(i32* %ptr, i32 0F)
8012 <i>; yields {i32}:result3 = FF</i>
8013 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
8018 <!-- _______________________________________________________________________ -->
8020 <a name="int_atomic_load_max">
8021 '<tt>llvm.atomic.load.max.*</tt>' Intrinsic
8024 <a name="int_atomic_load_min">
8025 '<tt>llvm.atomic.load.min.*</tt>' Intrinsic
8028 <a name="int_atomic_load_umax">
8029 '<tt>llvm.atomic.load.umax.*</tt>' Intrinsic
8032 <a name="int_atomic_load_umin">
8033 '<tt>llvm.atomic.load.umin.*</tt>' Intrinsic
8040 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
8041 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
8042 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
8043 address spaces. Not all targets support all bit widths however.</p>
8046 declare i8 @llvm.atomic.load.max.i8.p0i8(i8* <ptr>, i8 <delta>)
8047 declare i16 @llvm.atomic.load.max.i16.p0i16(i16* <ptr>, i16 <delta>)
8048 declare i32 @llvm.atomic.load.max.i32.p0i32(i32* <ptr>, i32 <delta>)
8049 declare i64 @llvm.atomic.load.max.i64.p0i64(i64* <ptr>, i64 <delta>)
8053 declare i8 @llvm.atomic.load.min.i8.p0i8(i8* <ptr>, i8 <delta>)
8054 declare i16 @llvm.atomic.load.min.i16.p0i16(i16* <ptr>, i16 <delta>)
8055 declare i32 @llvm.atomic.load.min.i32.p0i32(i32* <ptr>, i32 <delta>)
8056 declare i64 @llvm.atomic.load.min.i64.p0i64(i64* <ptr>, i64 <delta>)
8060 declare i8 @llvm.atomic.load.umax.i8.p0i8(i8* <ptr>, i8 <delta>)
8061 declare i16 @llvm.atomic.load.umax.i16.p0i16(i16* <ptr>, i16 <delta>)
8062 declare i32 @llvm.atomic.load.umax.i32.p0i32(i32* <ptr>, i32 <delta>)
8063 declare i64 @llvm.atomic.load.umax.i64.p0i64(i64* <ptr>, i64 <delta>)
8067 declare i8 @llvm.atomic.load.umin.i8.p0i8(i8* <ptr>, i8 <delta>)
8068 declare i16 @llvm.atomic.load.umin.i16.p0i16(i16* <ptr>, i16 <delta>)
8069 declare i32 @llvm.atomic.load.umin.i32.p0i32(i32* <ptr>, i32 <delta>)
8070 declare i64 @llvm.atomic.load.umin.i64.p0i64(i64* <ptr>, i64 <delta>)
8074 <p>These intrinsics takes the signed or unsigned minimum or maximum of
8075 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
8076 original value at <tt>ptr</tt>.</p>
8079 <p>These intrinsics take two arguments, the first a pointer to an integer value
8080 and the second an integer value. The result is also an integer value. These
8081 integer types can have any bit width, but they must all have the same bit
8082 width. The targets may only lower integer representations they support.</p>
8085 <p>These intrinsics does a series of operations atomically. They first load the
8086 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
8087 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
8088 yield the original value stored at <tt>ptr</tt>.</p>
8092 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8093 %ptr = bitcast i8* %mallocP to i32*
8095 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32(i32* %ptr, i32 -2)
8096 <i>; yields {i32}:result0 = 7</i>
8097 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32(i32* %ptr, i32 8)
8098 <i>; yields {i32}:result1 = -2</i>
8099 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32(i32* %ptr, i32 10)
8100 <i>; yields {i32}:result2 = 8</i>
8101 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32(i32* %ptr, i32 30)
8102 <i>; yields {i32}:result3 = 8</i>
8103 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
8110 <!-- ======================================================================= -->
8112 <a name="int_memorymarkers">Memory Use Markers</a>
8117 <p>This class of intrinsics exists to information about the lifetime of memory
8118 objects and ranges where variables are immutable.</p>
8120 <!-- _______________________________________________________________________ -->
8122 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8129 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8133 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8134 object's lifetime.</p>
8137 <p>The first argument is a constant integer representing the size of the
8138 object, or -1 if it is variable sized. The second argument is a pointer to
8142 <p>This intrinsic indicates that before this point in the code, the value of the
8143 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8144 never be used and has an undefined value. A load from the pointer that
8145 precedes this intrinsic can be replaced with
8146 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8150 <!-- _______________________________________________________________________ -->
8152 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8159 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8163 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8164 object's lifetime.</p>
8167 <p>The first argument is a constant integer representing the size of the
8168 object, or -1 if it is variable sized. The second argument is a pointer to
8172 <p>This intrinsic indicates that after this point in the code, the value of the
8173 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8174 never be used and has an undefined value. Any stores into the memory object
8175 following this intrinsic may be removed as dead.
8179 <!-- _______________________________________________________________________ -->
8181 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8188 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8192 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8193 a memory object will not change.</p>
8196 <p>The first argument is a constant integer representing the size of the
8197 object, or -1 if it is variable sized. The second argument is a pointer to
8201 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8202 the return value, the referenced memory location is constant and
8207 <!-- _______________________________________________________________________ -->
8209 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8216 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8220 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8221 a memory object are mutable.</p>
8224 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8225 The second argument is a constant integer representing the size of the
8226 object, or -1 if it is variable sized and the third argument is a pointer
8230 <p>This intrinsic indicates that the memory is mutable again.</p>
8236 <!-- ======================================================================= -->
8238 <a name="int_general">General Intrinsics</a>
8243 <p>This class of intrinsics is designed to be generic and has no specific
8246 <!-- _______________________________________________________________________ -->
8248 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8255 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8259 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8262 <p>The first argument is a pointer to a value, the second is a pointer to a
8263 global string, the third is a pointer to a global string which is the source
8264 file name, and the last argument is the line number.</p>
8267 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8268 This can be useful for special purpose optimizations that want to look for
8269 these annotations. These have no other defined use, they are ignored by code
8270 generation and optimization.</p>
8274 <!-- _______________________________________________________________________ -->
8276 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8282 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8283 any integer bit width.</p>
8286 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8287 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8288 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8289 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8290 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8294 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8297 <p>The first argument is an integer value (result of some expression), the
8298 second is a pointer to a global string, the third is a pointer to a global
8299 string which is the source file name, and the last argument is the line
8300 number. It returns the value of the first argument.</p>
8303 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8304 arbitrary strings. This can be useful for special purpose optimizations that
8305 want to look for these annotations. These have no other defined use, they
8306 are ignored by code generation and optimization.</p>
8310 <!-- _______________________________________________________________________ -->
8312 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8319 declare void @llvm.trap()
8323 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8329 <p>This intrinsics is lowered to the target dependent trap instruction. If the
8330 target does not have a trap instruction, this intrinsic will be lowered to
8331 the call of the <tt>abort()</tt> function.</p>
8335 <!-- _______________________________________________________________________ -->
8337 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8344 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8348 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8349 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8350 ensure that it is placed on the stack before local variables.</p>
8353 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8354 arguments. The first argument is the value loaded from the stack
8355 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8356 that has enough space to hold the value of the guard.</p>
8359 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8360 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8361 stack. This is to ensure that if a local variable on the stack is
8362 overwritten, it will destroy the value of the guard. When the function exits,
8363 the guard on the stack is checked against the original guard. If they are
8364 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8369 <!-- _______________________________________________________________________ -->
8371 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8378 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <type>)
8379 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <type>)
8383 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8384 the optimizers to determine at compile time whether a) an operation (like
8385 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8386 runtime check for overflow isn't necessary. An object in this context means
8387 an allocation of a specific class, structure, array, or other object.</p>
8390 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8391 argument is a pointer to or into the <tt>object</tt>. The second argument
8392 is a boolean 0 or 1. This argument determines whether you want the
8393 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
8394 1, variables are not allowed.</p>
8397 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
8398 representing the size of the object concerned, or <tt>i32/i64 -1 or 0</tt>,
8399 depending on the <tt>type</tt> argument, if the size cannot be determined at
8408 <!-- *********************************************************************** -->
8411 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8412 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8413 <a href="http://validator.w3.org/check/referer"><img
8414 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8416 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8417 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8418 Last modified: $Date$