1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <h1>LLVM Language Reference Manual</h1>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a>
25 <li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
26 <li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
27 <li><a href="#linkage_linker_private_weak">'<tt>linker_private_weak</tt>' Linkage</a></li>
28 <li><a href="#linkage_linker_private_weak_def_auto">'<tt>linker_private_weak_def_auto</tt>' Linkage</a></li>
29 <li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
30 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
31 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
32 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
33 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
34 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
35 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
36 <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
37 <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
38 <li><a href="#linkage_external">'<tt>externally visible</tt>' Linkage</a></li>
39 <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
40 <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
43 <li><a href="#callingconv">Calling Conventions</a></li>
44 <li><a href="#namedtypes">Named Types</a></li>
45 <li><a href="#globalvars">Global Variables</a></li>
46 <li><a href="#functionstructure">Functions</a></li>
47 <li><a href="#aliasstructure">Aliases</a></li>
48 <li><a href="#namedmetadatastructure">Named Metadata</a></li>
49 <li><a href="#paramattrs">Parameter Attributes</a></li>
50 <li><a href="#fnattrs">Function Attributes</a></li>
51 <li><a href="#gc">Garbage Collector Names</a></li>
52 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
53 <li><a href="#datalayout">Data Layout</a></li>
54 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
55 <li><a href="#volatile">Volatile Memory Accesses</a></li>
56 <li><a href="#memmodel">Memory Model for Concurrent Operations</a></li>
57 <li><a href="#ordering">Atomic Memory Ordering Constraints</a></li>
60 <li><a href="#typesystem">Type System</a>
62 <li><a href="#t_classifications">Type Classifications</a></li>
63 <li><a href="#t_primitive">Primitive Types</a>
65 <li><a href="#t_integer">Integer Type</a></li>
66 <li><a href="#t_floating">Floating Point Types</a></li>
67 <li><a href="#t_x86mmx">X86mmx Type</a></li>
68 <li><a href="#t_void">Void Type</a></li>
69 <li><a href="#t_label">Label Type</a></li>
70 <li><a href="#t_metadata">Metadata Type</a></li>
73 <li><a href="#t_derived">Derived Types</a>
75 <li><a href="#t_aggregate">Aggregate Types</a>
77 <li><a href="#t_array">Array Type</a></li>
78 <li><a href="#t_struct">Structure Type</a></li>
79 <li><a href="#t_opaque">Opaque Structure Types</a></li>
80 <li><a href="#t_vector">Vector Type</a></li>
83 <li><a href="#t_function">Function Type</a></li>
84 <li><a href="#t_pointer">Pointer Type</a></li>
89 <li><a href="#constants">Constants</a>
91 <li><a href="#simpleconstants">Simple Constants</a></li>
92 <li><a href="#complexconstants">Complex Constants</a></li>
93 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
94 <li><a href="#undefvalues">Undefined Values</a></li>
95 <li><a href="#trapvalues">Trap Values</a></li>
96 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
97 <li><a href="#constantexprs">Constant Expressions</a></li>
100 <li><a href="#othervalues">Other Values</a>
102 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
103 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a></li>
106 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
108 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
109 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
110 Global Variable</a></li>
111 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
112 Global Variable</a></li>
113 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
114 Global Variable</a></li>
117 <li><a href="#instref">Instruction Reference</a>
119 <li><a href="#terminators">Terminator Instructions</a>
121 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
122 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
123 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
124 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
125 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
126 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
127 <li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li>
128 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
131 <li><a href="#binaryops">Binary Operations</a>
133 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
134 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
135 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
136 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
137 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
138 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
139 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
140 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
141 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
142 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
143 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
144 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
147 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
149 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
150 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
151 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
152 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
153 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
154 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
157 <li><a href="#vectorops">Vector Operations</a>
159 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
160 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
161 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
164 <li><a href="#aggregateops">Aggregate Operations</a>
166 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
167 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
170 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
172 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
173 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
174 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
175 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
176 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
177 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
178 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
181 <li><a href="#convertops">Conversion Operations</a>
183 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
184 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
185 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
186 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
187 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
188 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
189 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
190 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
191 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
192 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
193 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
194 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
197 <li><a href="#otherops">Other Operations</a>
199 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
200 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
201 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
202 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
203 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
204 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
205 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
210 <li><a href="#intrinsics">Intrinsic Functions</a>
212 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
214 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
215 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
216 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
219 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
221 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
222 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
223 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
226 <li><a href="#int_codegen">Code Generator Intrinsics</a>
228 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
229 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
230 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
231 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
232 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
233 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
234 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
237 <li><a href="#int_libc">Standard C Library Intrinsics</a>
239 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
240 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
241 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
242 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
243 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
244 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
245 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
246 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
247 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
248 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
249 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
252 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
254 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
255 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
256 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
257 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
260 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
262 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
263 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
264 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
265 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
266 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
267 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
270 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
272 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
273 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
276 <li><a href="#int_debugger">Debugger intrinsics</a></li>
277 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
278 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
280 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
281 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
284 <li><a href="#int_atomics">Atomic intrinsics</a>
286 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
287 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
288 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
289 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
290 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
291 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
292 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
293 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
294 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
295 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
296 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
297 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
298 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
301 <li><a href="#int_memorymarkers">Memory Use Markers</a>
303 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
304 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
305 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
306 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
309 <li><a href="#int_general">General intrinsics</a>
311 <li><a href="#int_var_annotation">
312 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
313 <li><a href="#int_annotation">
314 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
315 <li><a href="#int_trap">
316 '<tt>llvm.trap</tt>' Intrinsic</a></li>
317 <li><a href="#int_stackprotector">
318 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
319 <li><a href="#int_objectsize">
320 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
327 <div class="doc_author">
328 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
329 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
332 <!-- *********************************************************************** -->
333 <h2><a name="abstract">Abstract</a></h2>
334 <!-- *********************************************************************** -->
338 <p>This document is a reference manual for the LLVM assembly language. LLVM is
339 a Static Single Assignment (SSA) based representation that provides type
340 safety, low-level operations, flexibility, and the capability of representing
341 'all' high-level languages cleanly. It is the common code representation
342 used throughout all phases of the LLVM compilation strategy.</p>
346 <!-- *********************************************************************** -->
347 <h2><a name="introduction">Introduction</a></h2>
348 <!-- *********************************************************************** -->
352 <p>The LLVM code representation is designed to be used in three different forms:
353 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
354 for fast loading by a Just-In-Time compiler), and as a human readable
355 assembly language representation. This allows LLVM to provide a powerful
356 intermediate representation for efficient compiler transformations and
357 analysis, while providing a natural means to debug and visualize the
358 transformations. The three different forms of LLVM are all equivalent. This
359 document describes the human readable representation and notation.</p>
361 <p>The LLVM representation aims to be light-weight and low-level while being
362 expressive, typed, and extensible at the same time. It aims to be a
363 "universal IR" of sorts, by being at a low enough level that high-level ideas
364 may be cleanly mapped to it (similar to how microprocessors are "universal
365 IR's", allowing many source languages to be mapped to them). By providing
366 type information, LLVM can be used as the target of optimizations: for
367 example, through pointer analysis, it can be proven that a C automatic
368 variable is never accessed outside of the current function, allowing it to
369 be promoted to a simple SSA value instead of a memory location.</p>
371 <!-- _______________________________________________________________________ -->
373 <a name="wellformed">Well-Formedness</a>
378 <p>It is important to note that this document describes 'well formed' LLVM
379 assembly language. There is a difference between what the parser accepts and
380 what is considered 'well formed'. For example, the following instruction is
381 syntactically okay, but not well formed:</p>
383 <pre class="doc_code">
384 %x = <a href="#i_add">add</a> i32 1, %x
387 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
388 LLVM infrastructure provides a verification pass that may be used to verify
389 that an LLVM module is well formed. This pass is automatically run by the
390 parser after parsing input assembly and by the optimizer before it outputs
391 bitcode. The violations pointed out by the verifier pass indicate bugs in
392 transformation passes or input to the parser.</p>
398 <!-- Describe the typesetting conventions here. -->
400 <!-- *********************************************************************** -->
401 <h2><a name="identifiers">Identifiers</a></h2>
402 <!-- *********************************************************************** -->
406 <p>LLVM identifiers come in two basic types: global and local. Global
407 identifiers (functions, global variables) begin with the <tt>'@'</tt>
408 character. Local identifiers (register names, types) begin with
409 the <tt>'%'</tt> character. Additionally, there are three different formats
410 for identifiers, for different purposes:</p>
413 <li>Named values are represented as a string of characters with their prefix.
414 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
415 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
416 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
417 other characters in their names can be surrounded with quotes. Special
418 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
419 ASCII code for the character in hexadecimal. In this way, any character
420 can be used in a name value, even quotes themselves.</li>
422 <li>Unnamed values are represented as an unsigned numeric value with their
423 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
425 <li>Constants, which are described in a <a href="#constants">section about
426 constants</a>, below.</li>
429 <p>LLVM requires that values start with a prefix for two reasons: Compilers
430 don't need to worry about name clashes with reserved words, and the set of
431 reserved words may be expanded in the future without penalty. Additionally,
432 unnamed identifiers allow a compiler to quickly come up with a temporary
433 variable without having to avoid symbol table conflicts.</p>
435 <p>Reserved words in LLVM are very similar to reserved words in other
436 languages. There are keywords for different opcodes
437 ('<tt><a href="#i_add">add</a></tt>',
438 '<tt><a href="#i_bitcast">bitcast</a></tt>',
439 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
440 ('<tt><a href="#t_void">void</a></tt>',
441 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
442 reserved words cannot conflict with variable names, because none of them
443 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
445 <p>Here is an example of LLVM code to multiply the integer variable
446 '<tt>%X</tt>' by 8:</p>
450 <pre class="doc_code">
451 %result = <a href="#i_mul">mul</a> i32 %X, 8
454 <p>After strength reduction:</p>
456 <pre class="doc_code">
457 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
460 <p>And the hard way:</p>
462 <pre class="doc_code">
463 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
464 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
465 %result = <a href="#i_add">add</a> i32 %1, %1
468 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
469 lexical features of LLVM:</p>
472 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
475 <li>Unnamed temporaries are created when the result of a computation is not
476 assigned to a named value.</li>
478 <li>Unnamed temporaries are numbered sequentially</li>
481 <p>It also shows a convention that we follow in this document. When
482 demonstrating instructions, we will follow an instruction with a comment that
483 defines the type and name of value produced. Comments are shown in italic
488 <!-- *********************************************************************** -->
489 <h2><a name="highlevel">High Level Structure</a></h2>
490 <!-- *********************************************************************** -->
492 <!-- ======================================================================= -->
494 <a name="modulestructure">Module Structure</a>
499 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
500 of the input programs. Each module consists of functions, global variables,
501 and symbol table entries. Modules may be combined together with the LLVM
502 linker, which merges function (and global variable) definitions, resolves
503 forward declarations, and merges symbol table entries. Here is an example of
504 the "hello world" module:</p>
506 <pre class="doc_code">
507 <i>; Declare the string constant as a global constant.</i>
508 <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>
510 <i>; External declaration of the puts function</i>
511 <a href="#functionstructure">declare</a> i32 @puts(i8*) <i>; i32 (i8*)* </i>
513 <i>; Definition of main function</i>
514 define i32 @main() { <i>; i32()* </i>
515 <i>; Convert [13 x i8]* to i8 *...</i>
516 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8*</i>
518 <i>; Call puts function to write out the string to stdout.</i>
519 <a href="#i_call">call</a> i32 @puts(i8* %cast210) <i>; i32</i>
520 <a href="#i_ret">ret</a> i32 0
523 <i>; Named metadata</i>
524 !1 = metadata !{i32 41}
528 <p>This example is made up of a <a href="#globalvars">global variable</a> named
529 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
530 a <a href="#functionstructure">function definition</a> for
531 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
534 <p>In general, a module is made up of a list of global values, where both
535 functions and global variables are global values. Global values are
536 represented by a pointer to a memory location (in this case, a pointer to an
537 array of char, and a pointer to a function), and have one of the
538 following <a href="#linkage">linkage types</a>.</p>
542 <!-- ======================================================================= -->
544 <a name="linkage">Linkage Types</a>
549 <p>All Global Variables and Functions have one of the following types of
553 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
554 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
555 by objects in the current module. In particular, linking code into a
556 module with an private global value may cause the private to be renamed as
557 necessary to avoid collisions. Because the symbol is private to the
558 module, all references can be updated. This doesn't show up in any symbol
559 table in the object file.</dd>
561 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
562 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
563 assembler and evaluated by the linker. Unlike normal strong symbols, they
564 are removed by the linker from the final linked image (executable or
565 dynamic library).</dd>
567 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
568 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
569 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
570 linker. The symbols are removed by the linker from the final linked image
571 (executable or dynamic library).</dd>
573 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
574 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
575 of the object is not taken. For instance, functions that had an inline
576 definition, but the compiler decided not to inline it. Note,
577 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
578 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
579 visibility. The symbols are removed by the linker from the final linked
580 image (executable or dynamic library).</dd>
582 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
583 <dd>Similar to private, but the value shows as a local symbol
584 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
585 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
587 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
588 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
589 into the object file corresponding to the LLVM module. They exist to
590 allow inlining and other optimizations to take place given knowledge of
591 the definition of the global, which is known to be somewhere outside the
592 module. Globals with <tt>available_externally</tt> linkage are allowed to
593 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
594 This linkage type is only allowed on definitions, not declarations.</dd>
596 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
597 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
598 the same name when linkage occurs. This can be used to implement
599 some forms of inline functions, templates, or other code which must be
600 generated in each translation unit that uses it, but where the body may
601 be overridden with a more definitive definition later. Unreferenced
602 <tt>linkonce</tt> globals are allowed to be discarded. Note that
603 <tt>linkonce</tt> linkage does not actually allow the optimizer to
604 inline the body of this function into callers because it doesn't know if
605 this definition of the function is the definitive definition within the
606 program or whether it will be overridden by a stronger definition.
607 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
610 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
611 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
612 <tt>linkonce</tt> linkage, except that unreferenced globals with
613 <tt>weak</tt> linkage may not be discarded. This is used for globals that
614 are declared "weak" in C source code.</dd>
616 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
617 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
618 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
620 Symbols with "<tt>common</tt>" linkage are merged in the same way as
621 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
622 <tt>common</tt> symbols may not have an explicit section,
623 must have a zero initializer, and may not be marked '<a
624 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
625 have common linkage.</dd>
628 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
629 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
630 pointer to array type. When two global variables with appending linkage
631 are linked together, the two global arrays are appended together. This is
632 the LLVM, typesafe, equivalent of having the system linker append together
633 "sections" with identical names when .o files are linked.</dd>
635 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
636 <dd>The semantics of this linkage follow the ELF object file model: the symbol
637 is weak until linked, if not linked, the symbol becomes null instead of
638 being an undefined reference.</dd>
640 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
641 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
642 <dd>Some languages allow differing globals to be merged, such as two functions
643 with different semantics. Other languages, such as <tt>C++</tt>, ensure
644 that only equivalent globals are ever merged (the "one definition rule"
645 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
646 and <tt>weak_odr</tt> linkage types to indicate that the global will only
647 be merged with equivalent globals. These linkage types are otherwise the
648 same as their non-<tt>odr</tt> versions.</dd>
650 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
651 <dd>If none of the above identifiers are used, the global is externally
652 visible, meaning that it participates in linkage and can be used to
653 resolve external symbol references.</dd>
656 <p>The next two types of linkage are targeted for Microsoft Windows platform
657 only. They are designed to support importing (exporting) symbols from (to)
658 DLLs (Dynamic Link Libraries).</p>
661 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
662 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
663 or variable via a global pointer to a pointer that is set up by the DLL
664 exporting the symbol. On Microsoft Windows targets, the pointer name is
665 formed by combining <code>__imp_</code> and the function or variable
668 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
669 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
670 pointer to a pointer in a DLL, so that it can be referenced with the
671 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
672 name is formed by combining <code>__imp_</code> and the function or
676 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
677 another module defined a "<tt>.LC0</tt>" variable and was linked with this
678 one, one of the two would be renamed, preventing a collision. Since
679 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
680 declarations), they are accessible outside of the current module.</p>
682 <p>It is illegal for a function <i>declaration</i> to have any linkage type
683 other than "externally visible", <tt>dllimport</tt>
684 or <tt>extern_weak</tt>.</p>
686 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
687 or <tt>weak_odr</tt> linkages.</p>
691 <!-- ======================================================================= -->
693 <a name="callingconv">Calling Conventions</a>
698 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
699 and <a href="#i_invoke">invokes</a> can all have an optional calling
700 convention specified for the call. The calling convention of any pair of
701 dynamic caller/callee must match, or the behavior of the program is
702 undefined. The following calling conventions are supported by LLVM, and more
703 may be added in the future:</p>
706 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
707 <dd>This calling convention (the default if no other calling convention is
708 specified) matches the target C calling conventions. This calling
709 convention supports varargs function calls and tolerates some mismatch in
710 the declared prototype and implemented declaration of the function (as
713 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
714 <dd>This calling convention attempts to make calls as fast as possible
715 (e.g. by passing things in registers). This calling convention allows the
716 target to use whatever tricks it wants to produce fast code for the
717 target, without having to conform to an externally specified ABI
718 (Application Binary Interface).
719 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
720 when this or the GHC convention is used.</a> This calling convention
721 does not support varargs and requires the prototype of all callees to
722 exactly match the prototype of the function definition.</dd>
724 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
725 <dd>This calling convention attempts to make code in the caller as efficient
726 as possible under the assumption that the call is not commonly executed.
727 As such, these calls often preserve all registers so that the call does
728 not break any live ranges in the caller side. This calling convention
729 does not support varargs and requires the prototype of all callees to
730 exactly match the prototype of the function definition.</dd>
732 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
733 <dd>This calling convention has been implemented specifically for use by the
734 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
735 It passes everything in registers, going to extremes to achieve this by
736 disabling callee save registers. This calling convention should not be
737 used lightly but only for specific situations such as an alternative to
738 the <em>register pinning</em> performance technique often used when
739 implementing functional programming languages.At the moment only X86
740 supports this convention and it has the following limitations:
742 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
743 floating point types are supported.</li>
744 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
745 6 floating point parameters.</li>
747 This calling convention supports
748 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
749 requires both the caller and callee are using it.
752 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
753 <dd>Any calling convention may be specified by number, allowing
754 target-specific calling conventions to be used. Target specific calling
755 conventions start at 64.</dd>
758 <p>More calling conventions can be added/defined on an as-needed basis, to
759 support Pascal conventions or any other well-known target-independent
764 <!-- ======================================================================= -->
766 <a name="visibility">Visibility Styles</a>
771 <p>All Global Variables and Functions have one of the following visibility
775 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
776 <dd>On targets that use the ELF object file format, default visibility means
777 that the declaration is visible to other modules and, in shared libraries,
778 means that the declared entity may be overridden. On Darwin, default
779 visibility means that the declaration is visible to other modules. Default
780 visibility corresponds to "external linkage" in the language.</dd>
782 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
783 <dd>Two declarations of an object with hidden visibility refer to the same
784 object if they are in the same shared object. Usually, hidden visibility
785 indicates that the symbol will not be placed into the dynamic symbol
786 table, so no other module (executable or shared library) can reference it
789 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
790 <dd>On ELF, protected visibility indicates that the symbol will be placed in
791 the dynamic symbol table, but that references within the defining module
792 will bind to the local symbol. That is, the symbol cannot be overridden by
798 <!-- ======================================================================= -->
800 <a name="namedtypes">Named Types</a>
805 <p>LLVM IR allows you to specify name aliases for certain types. This can make
806 it easier to read the IR and make the IR more condensed (particularly when
807 recursive types are involved). An example of a name specification is:</p>
809 <pre class="doc_code">
810 %mytype = type { %mytype*, i32 }
813 <p>You may give a name to any <a href="#typesystem">type</a> except
814 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
815 is expected with the syntax "%mytype".</p>
817 <p>Note that type names are aliases for the structural type that they indicate,
818 and that you can therefore specify multiple names for the same type. This
819 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
820 uses structural typing, the name is not part of the type. When printing out
821 LLVM IR, the printer will pick <em>one name</em> to render all types of a
822 particular shape. This means that if you have code where two different
823 source types end up having the same LLVM type, that the dumper will sometimes
824 print the "wrong" or unexpected type. This is an important design point and
825 isn't going to change.</p>
829 <!-- ======================================================================= -->
831 <a name="globalvars">Global Variables</a>
836 <p>Global variables define regions of memory allocated at compilation time
837 instead of run-time. Global variables may optionally be initialized, may
838 have an explicit section to be placed in, and may have an optional explicit
839 alignment specified. A variable may be defined as "thread_local", which
840 means that it will not be shared by threads (each thread will have a
841 separated copy of the variable). A variable may be defined as a global
842 "constant," which indicates that the contents of the variable
843 will <b>never</b> be modified (enabling better optimization, allowing the
844 global data to be placed in the read-only section of an executable, etc).
845 Note that variables that need runtime initialization cannot be marked
846 "constant" as there is a store to the variable.</p>
848 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
849 constant, even if the final definition of the global is not. This capability
850 can be used to enable slightly better optimization of the program, but
851 requires the language definition to guarantee that optimizations based on the
852 'constantness' are valid for the translation units that do not include the
855 <p>As SSA values, global variables define pointer values that are in scope
856 (i.e. they dominate) all basic blocks in the program. Global variables
857 always define a pointer to their "content" type because they describe a
858 region of memory, and all memory objects in LLVM are accessed through
861 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
862 that the address is not significant, only the content. Constants marked
863 like this can be merged with other constants if they have the same
864 initializer. Note that a constant with significant address <em>can</em>
865 be merged with a <tt>unnamed_addr</tt> constant, the result being a
866 constant whose address is significant.</p>
868 <p>A global variable may be declared to reside in a target-specific numbered
869 address space. For targets that support them, address spaces may affect how
870 optimizations are performed and/or what target instructions are used to
871 access the variable. The default address space is zero. The address space
872 qualifier must precede any other attributes.</p>
874 <p>LLVM allows an explicit section to be specified for globals. If the target
875 supports it, it will emit globals to the section specified.</p>
877 <p>An explicit alignment may be specified for a global, which must be a power
878 of 2. If not present, or if the alignment is set to zero, the alignment of
879 the global is set by the target to whatever it feels convenient. If an
880 explicit alignment is specified, the global is forced to have exactly that
881 alignment. Targets and optimizers are not allowed to over-align the global
882 if the global has an assigned section. In this case, the extra alignment
883 could be observable: for example, code could assume that the globals are
884 densely packed in their section and try to iterate over them as an array,
885 alignment padding would break this iteration.</p>
887 <p>For example, the following defines a global in a numbered address space with
888 an initializer, section, and alignment:</p>
890 <pre class="doc_code">
891 @G = addrspace(5) constant float 1.0, section "foo", align 4
897 <!-- ======================================================================= -->
899 <a name="functionstructure">Functions</a>
904 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
905 optional <a href="#linkage">linkage type</a>, an optional
906 <a href="#visibility">visibility style</a>, an optional
907 <a href="#callingconv">calling convention</a>,
908 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
909 <a href="#paramattrs">parameter attribute</a> for the return type, a function
910 name, a (possibly empty) argument list (each with optional
911 <a href="#paramattrs">parameter attributes</a>), optional
912 <a href="#fnattrs">function attributes</a>, an optional section, an optional
913 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
914 curly brace, a list of basic blocks, and a closing curly brace.</p>
916 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
917 optional <a href="#linkage">linkage type</a>, an optional
918 <a href="#visibility">visibility style</a>, an optional
919 <a href="#callingconv">calling convention</a>,
920 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
921 <a href="#paramattrs">parameter attribute</a> for the return type, a function
922 name, a possibly empty list of arguments, an optional alignment, and an
923 optional <a href="#gc">garbage collector name</a>.</p>
925 <p>A function definition contains a list of basic blocks, forming the CFG
926 (Control Flow Graph) for the function. Each basic block may optionally start
927 with a label (giving the basic block a symbol table entry), contains a list
928 of instructions, and ends with a <a href="#terminators">terminator</a>
929 instruction (such as a branch or function return).</p>
931 <p>The first basic block in a function is special in two ways: it is immediately
932 executed on entrance to the function, and it is not allowed to have
933 predecessor basic blocks (i.e. there can not be any branches to the entry
934 block of a function). Because the block can have no predecessors, it also
935 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
937 <p>LLVM allows an explicit section to be specified for functions. If the target
938 supports it, it will emit functions to the section specified.</p>
940 <p>An explicit alignment may be specified for a function. If not present, or if
941 the alignment is set to zero, the alignment of the function is set by the
942 target to whatever it feels convenient. If an explicit alignment is
943 specified, the function is forced to have at least that much alignment. All
944 alignments must be a power of 2.</p>
946 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
947 be significant and two identical functions can be merged</p>.
950 <pre class="doc_code">
951 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
952 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
953 <ResultType> @<FunctionName> ([argument list])
954 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
955 [<a href="#gc">gc</a>] { ... }
960 <!-- ======================================================================= -->
962 <a name="aliasstructure">Aliases</a>
967 <p>Aliases act as "second name" for the aliasee value (which can be either
968 function, global variable, another alias or bitcast of global value). Aliases
969 may have an optional <a href="#linkage">linkage type</a>, and an
970 optional <a href="#visibility">visibility style</a>.</p>
973 <pre class="doc_code">
974 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
979 <!-- ======================================================================= -->
981 <a name="namedmetadatastructure">Named Metadata</a>
986 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
987 nodes</a> (but not metadata strings) are the only valid operands for
988 a named metadata.</p>
991 <pre class="doc_code">
992 ; Some unnamed metadata nodes, which are referenced by the named metadata.
993 !0 = metadata !{metadata !"zero"}
994 !1 = metadata !{metadata !"one"}
995 !2 = metadata !{metadata !"two"}
997 !name = !{!0, !1, !2}
1002 <!-- ======================================================================= -->
1004 <a name="paramattrs">Parameter Attributes</a>
1009 <p>The return type and each parameter of a function type may have a set of
1010 <i>parameter attributes</i> associated with them. Parameter attributes are
1011 used to communicate additional information about the result or parameters of
1012 a function. Parameter attributes are considered to be part of the function,
1013 not of the function type, so functions with different parameter attributes
1014 can have the same function type.</p>
1016 <p>Parameter attributes are simple keywords that follow the type specified. If
1017 multiple parameter attributes are needed, they are space separated. For
1020 <pre class="doc_code">
1021 declare i32 @printf(i8* noalias nocapture, ...)
1022 declare i32 @atoi(i8 zeroext)
1023 declare signext i8 @returns_signed_char()
1026 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1027 <tt>readonly</tt>) come immediately after the argument list.</p>
1029 <p>Currently, only the following parameter attributes are defined:</p>
1032 <dt><tt><b>zeroext</b></tt></dt>
1033 <dd>This indicates to the code generator that the parameter or return value
1034 should be zero-extended to the extent required by the target's ABI (which
1035 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1036 parameter) or the callee (for a return value).</dd>
1038 <dt><tt><b>signext</b></tt></dt>
1039 <dd>This indicates to the code generator that the parameter or return value
1040 should be sign-extended to the extent required by the target's ABI (which
1041 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1044 <dt><tt><b>inreg</b></tt></dt>
1045 <dd>This indicates that this parameter or return value should be treated in a
1046 special target-dependent fashion during while emitting code for a function
1047 call or return (usually, by putting it in a register as opposed to memory,
1048 though some targets use it to distinguish between two different kinds of
1049 registers). Use of this attribute is target-specific.</dd>
1051 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1052 <dd><p>This indicates that the pointer parameter should really be passed by
1053 value to the function. The attribute implies that a hidden copy of the
1055 is made between the caller and the callee, so the callee is unable to
1056 modify the value in the callee. This attribute is only valid on LLVM
1057 pointer arguments. It is generally used to pass structs and arrays by
1058 value, but is also valid on pointers to scalars. The copy is considered
1059 to belong to the caller not the callee (for example,
1060 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1061 <tt>byval</tt> parameters). This is not a valid attribute for return
1064 <p>The byval attribute also supports specifying an alignment with
1065 the align attribute. It indicates the alignment of the stack slot to
1066 form and the known alignment of the pointer specified to the call site. If
1067 the alignment is not specified, then the code generator makes a
1068 target-specific assumption.</p></dd>
1070 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1071 <dd>This indicates that the pointer parameter specifies the address of a
1072 structure that is the return value of the function in the source program.
1073 This pointer must be guaranteed by the caller to be valid: loads and
1074 stores to the structure may be assumed by the callee to not to trap. This
1075 may only be applied to the first parameter. This is not a valid attribute
1076 for return values. </dd>
1078 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1079 <dd>This indicates that pointer values
1080 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1081 value do not alias pointer values which are not <i>based</i> on it,
1082 ignoring certain "irrelevant" dependencies.
1083 For a call to the parent function, dependencies between memory
1084 references from before or after the call and from those during the call
1085 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1086 return value used in that call.
1087 The caller shares the responsibility with the callee for ensuring that
1088 these requirements are met.
1089 For further details, please see the discussion of the NoAlias response in
1090 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1092 Note that this definition of <tt>noalias</tt> is intentionally
1093 similar to the definition of <tt>restrict</tt> in C99 for function
1094 arguments, though it is slightly weaker.
1096 For function return values, C99's <tt>restrict</tt> is not meaningful,
1097 while LLVM's <tt>noalias</tt> is.
1100 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1101 <dd>This indicates that the callee does not make any copies of the pointer
1102 that outlive the callee itself. This is not a valid attribute for return
1105 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1106 <dd>This indicates that the pointer parameter can be excised using the
1107 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1108 attribute for return values.</dd>
1113 <!-- ======================================================================= -->
1115 <a name="gc">Garbage Collector Names</a>
1120 <p>Each function may specify a garbage collector name, which is simply a
1123 <pre class="doc_code">
1124 define void @f() gc "name" { ... }
1127 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1128 collector which will cause the compiler to alter its output in order to
1129 support the named garbage collection algorithm.</p>
1133 <!-- ======================================================================= -->
1135 <a name="fnattrs">Function Attributes</a>
1140 <p>Function attributes are set to communicate additional information about a
1141 function. Function attributes are considered to be part of the function, not
1142 of the function type, so functions with different parameter attributes can
1143 have the same function type.</p>
1145 <p>Function attributes are simple keywords that follow the type specified. If
1146 multiple attributes are needed, they are space separated. For example:</p>
1148 <pre class="doc_code">
1149 define void @f() noinline { ... }
1150 define void @f() alwaysinline { ... }
1151 define void @f() alwaysinline optsize { ... }
1152 define void @f() optsize { ... }
1156 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1157 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1158 the backend should forcibly align the stack pointer. Specify the
1159 desired alignment, which must be a power of two, in parentheses.
1161 <dt><tt><b>alwaysinline</b></tt></dt>
1162 <dd>This attribute indicates that the inliner should attempt to inline this
1163 function into callers whenever possible, ignoring any active inlining size
1164 threshold for this caller.</dd>
1166 <dt><tt><b>hotpatch</b></tt></dt>
1167 <dd>This attribute indicates that the function should be 'hotpatchable',
1168 meaning the function can be patched and/or hooked even while it is
1169 loaded into memory. On x86, the function prologue will be preceded
1170 by six bytes of padding and will begin with a two-byte instruction.
1171 Most of the functions in the Windows system DLLs in Windows XP SP2 or
1172 higher were compiled in this fashion.</dd>
1174 <dt><tt><b>nonlazybind</b></tt></dt>
1175 <dd>This attribute suppresses lazy symbol binding for the function. This
1176 may make calls to the function faster, at the cost of extra program
1177 startup time if the function is not called during program startup.</dd>
1179 <dt><tt><b>inlinehint</b></tt></dt>
1180 <dd>This attribute indicates that the source code contained a hint that inlining
1181 this function is desirable (such as the "inline" keyword in C/C++). It
1182 is just a hint; it imposes no requirements on the inliner.</dd>
1184 <dt><tt><b>naked</b></tt></dt>
1185 <dd>This attribute disables prologue / epilogue emission for the function.
1186 This can have very system-specific consequences.</dd>
1188 <dt><tt><b>noimplicitfloat</b></tt></dt>
1189 <dd>This attributes disables implicit floating point instructions.</dd>
1191 <dt><tt><b>noinline</b></tt></dt>
1192 <dd>This attribute indicates that the inliner should never inline this
1193 function in any situation. This attribute may not be used together with
1194 the <tt>alwaysinline</tt> attribute.</dd>
1196 <dt><tt><b>noredzone</b></tt></dt>
1197 <dd>This attribute indicates that the code generator should not use a red
1198 zone, even if the target-specific ABI normally permits it.</dd>
1200 <dt><tt><b>noreturn</b></tt></dt>
1201 <dd>This function attribute indicates that the function never returns
1202 normally. This produces undefined behavior at runtime if the function
1203 ever does dynamically return.</dd>
1205 <dt><tt><b>nounwind</b></tt></dt>
1206 <dd>This function attribute indicates that the function never returns with an
1207 unwind or exceptional control flow. If the function does unwind, its
1208 runtime behavior is undefined.</dd>
1210 <dt><tt><b>optsize</b></tt></dt>
1211 <dd>This attribute suggests that optimization passes and code generator passes
1212 make choices that keep the code size of this function low, and otherwise
1213 do optimizations specifically to reduce code size.</dd>
1215 <dt><tt><b>readnone</b></tt></dt>
1216 <dd>This attribute indicates that the function computes its result (or decides
1217 to unwind an exception) based strictly on its arguments, without
1218 dereferencing any pointer arguments or otherwise accessing any mutable
1219 state (e.g. memory, control registers, etc) visible to caller functions.
1220 It does not write through any pointer arguments
1221 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1222 changes any state visible to callers. This means that it cannot unwind
1223 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1224 could use the <tt>unwind</tt> instruction.</dd>
1226 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1227 <dd>This attribute indicates that the function does not write through any
1228 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1229 arguments) or otherwise modify any state (e.g. memory, control registers,
1230 etc) visible to caller functions. It may dereference pointer arguments
1231 and read state that may be set in the caller. A readonly function always
1232 returns the same value (or unwinds an exception identically) when called
1233 with the same set of arguments and global state. It cannot unwind an
1234 exception by calling the <tt>C++</tt> exception throwing methods, but may
1235 use the <tt>unwind</tt> instruction.</dd>
1237 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1238 <dd>This attribute indicates that the function should emit a stack smashing
1239 protector. It is in the form of a "canary"—a random value placed on
1240 the stack before the local variables that's checked upon return from the
1241 function to see if it has been overwritten. A heuristic is used to
1242 determine if a function needs stack protectors or not.<br>
1244 If a function that has an <tt>ssp</tt> attribute is inlined into a
1245 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1246 function will have an <tt>ssp</tt> attribute.</dd>
1248 <dt><tt><b>sspreq</b></tt></dt>
1249 <dd>This attribute indicates that the function should <em>always</em> emit a
1250 stack smashing protector. This overrides
1251 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1253 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1254 function that doesn't have an <tt>sspreq</tt> attribute or which has
1255 an <tt>ssp</tt> attribute, then the resulting function will have
1256 an <tt>sspreq</tt> attribute.</dd>
1258 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1259 <dd>This attribute indicates that the ABI being targeted requires that
1260 an unwind table entry be produce for this function even if we can
1261 show that no exceptions passes by it. This is normally the case for
1262 the ELF x86-64 abi, but it can be disabled for some compilation
1269 <!-- ======================================================================= -->
1271 <a name="moduleasm">Module-Level Inline Assembly</a>
1276 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1277 the GCC "file scope inline asm" blocks. These blocks are internally
1278 concatenated by LLVM and treated as a single unit, but may be separated in
1279 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1281 <pre class="doc_code">
1282 module asm "inline asm code goes here"
1283 module asm "more can go here"
1286 <p>The strings can contain any character by escaping non-printable characters.
1287 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1290 <p>The inline asm code is simply printed to the machine code .s file when
1291 assembly code is generated.</p>
1295 <!-- ======================================================================= -->
1297 <a name="datalayout">Data Layout</a>
1302 <p>A module may specify a target specific data layout string that specifies how
1303 data is to be laid out in memory. The syntax for the data layout is
1306 <pre class="doc_code">
1307 target datalayout = "<i>layout specification</i>"
1310 <p>The <i>layout specification</i> consists of a list of specifications
1311 separated by the minus sign character ('-'). Each specification starts with
1312 a letter and may include other information after the letter to define some
1313 aspect of the data layout. The specifications accepted are as follows:</p>
1317 <dd>Specifies that the target lays out data in big-endian form. That is, the
1318 bits with the most significance have the lowest address location.</dd>
1321 <dd>Specifies that the target lays out data in little-endian form. That is,
1322 the bits with the least significance have the lowest address
1325 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1326 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1327 <i>preferred</i> alignments. All sizes are in bits. Specifying
1328 the <i>pref</i> alignment is optional. If omitted, the
1329 preceding <tt>:</tt> should be omitted too.</dd>
1331 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1332 <dd>This specifies the alignment for an integer type of a given bit
1333 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1335 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1336 <dd>This specifies the alignment for a vector type of a given bit
1339 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1340 <dd>This specifies the alignment for a floating point type of a given bit
1341 <i>size</i>. Only values of <i>size</i> that are supported by the target
1342 will work. 32 (float) and 64 (double) are supported on all targets;
1343 80 or 128 (different flavors of long double) are also supported on some
1346 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1347 <dd>This specifies the alignment for an aggregate type of a given bit
1350 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1351 <dd>This specifies the alignment for a stack object of a given bit
1354 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1355 <dd>This specifies a set of native integer widths for the target CPU
1356 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1357 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1358 this set are considered to support most general arithmetic
1359 operations efficiently.</dd>
1362 <p>When constructing the data layout for a given target, LLVM starts with a
1363 default set of specifications which are then (possibly) overridden by the
1364 specifications in the <tt>datalayout</tt> keyword. The default specifications
1365 are given in this list:</p>
1368 <li><tt>E</tt> - big endian</li>
1369 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1370 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1371 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1372 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1373 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1374 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1375 alignment of 64-bits</li>
1376 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1377 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1378 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1379 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1380 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1381 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1384 <p>When LLVM is determining the alignment for a given type, it uses the
1385 following rules:</p>
1388 <li>If the type sought is an exact match for one of the specifications, that
1389 specification is used.</li>
1391 <li>If no match is found, and the type sought is an integer type, then the
1392 smallest integer type that is larger than the bitwidth of the sought type
1393 is used. If none of the specifications are larger than the bitwidth then
1394 the the largest integer type is used. For example, given the default
1395 specifications above, the i7 type will use the alignment of i8 (next
1396 largest) while both i65 and i256 will use the alignment of i64 (largest
1399 <li>If no match is found, and the type sought is a vector type, then the
1400 largest vector type that is smaller than the sought vector type will be
1401 used as a fall back. This happens because <128 x double> can be
1402 implemented in terms of 64 <2 x double>, for example.</li>
1407 <!-- ======================================================================= -->
1409 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1414 <p>Any memory access must be done through a pointer value associated
1415 with an address range of the memory access, otherwise the behavior
1416 is undefined. Pointer values are associated with address ranges
1417 according to the following rules:</p>
1420 <li>A pointer value is associated with the addresses associated with
1421 any value it is <i>based</i> on.
1422 <li>An address of a global variable is associated with the address
1423 range of the variable's storage.</li>
1424 <li>The result value of an allocation instruction is associated with
1425 the address range of the allocated storage.</li>
1426 <li>A null pointer in the default address-space is associated with
1428 <li>An integer constant other than zero or a pointer value returned
1429 from a function not defined within LLVM may be associated with address
1430 ranges allocated through mechanisms other than those provided by
1431 LLVM. Such ranges shall not overlap with any ranges of addresses
1432 allocated by mechanisms provided by LLVM.</li>
1435 <p>A pointer value is <i>based</i> on another pointer value according
1436 to the following rules:</p>
1439 <li>A pointer value formed from a
1440 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1441 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1442 <li>The result value of a
1443 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1444 of the <tt>bitcast</tt>.</li>
1445 <li>A pointer value formed by an
1446 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1447 pointer values that contribute (directly or indirectly) to the
1448 computation of the pointer's value.</li>
1449 <li>The "<i>based</i> on" relationship is transitive.</li>
1452 <p>Note that this definition of <i>"based"</i> is intentionally
1453 similar to the definition of <i>"based"</i> in C99, though it is
1454 slightly weaker.</p>
1456 <p>LLVM IR does not associate types with memory. The result type of a
1457 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1458 alignment of the memory from which to load, as well as the
1459 interpretation of the value. The first operand type of a
1460 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1461 and alignment of the store.</p>
1463 <p>Consequently, type-based alias analysis, aka TBAA, aka
1464 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1465 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1466 additional information which specialized optimization passes may use
1467 to implement type-based alias analysis.</p>
1471 <!-- ======================================================================= -->
1473 <a name="volatile">Volatile Memory Accesses</a>
1478 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1479 href="#i_store"><tt>store</tt></a>s, and <a
1480 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1481 The optimizers must not change the number of volatile operations or change their
1482 order of execution relative to other volatile operations. The optimizers
1483 <i>may</i> change the order of volatile operations relative to non-volatile
1484 operations. This is not Java's "volatile" and has no cross-thread
1485 synchronization behavior.</p>
1489 <!-- ======================================================================= -->
1491 <a name="memmodel">Memory Model for Concurrent Operations</a>
1496 <p>The LLVM IR does not define any way to start parallel threads of execution
1497 or to register signal handlers. Nonetheless, there are platform-specific
1498 ways to create them, and we define LLVM IR's behavior in their presence. This
1499 model is inspired by the C++0x memory model.</p>
1501 <p>For a more informal introduction to this model, see the
1502 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1504 <p>We define a <i>happens-before</i> partial order as the least partial order
1507 <li>Is a superset of single-thread program order, and</li>
1508 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1509 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1510 by platform-specific techniques, like pthread locks, thread
1511 creation, thread joining, etc., and by atomic instructions.
1512 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1516 <p>Note that program order does not introduce <i>happens-before</i> edges
1517 between a thread and signals executing inside that thread.</p>
1519 <p>Every (defined) read operation (load instructions, memcpy, atomic
1520 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1521 (defined) write operations (store instructions, atomic
1522 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1523 initialized globals are considered to have a write of the initializer which is
1524 atomic and happens before any other read or write of the memory in question.
1525 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1526 any write to the same byte, except:</p>
1529 <li>If <var>write<sub>1</sub></var> happens before
1530 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1531 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1532 does not see <var>write<sub>1</sub></var>.
1533 <li>If <var>R<sub>byte</sub></var> happens before
1534 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1535 see <var>write<sub>3</sub></var>.
1538 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1540 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1541 is supposed to give guarantees which can support
1542 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1543 addresses which do not behave like normal memory. It does not generally
1544 provide cross-thread synchronization.)
1545 <li>Otherwise, if there is no write to the same byte that happens before
1546 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1547 <tt>undef</tt> for that byte.
1548 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1549 <var>R<sub>byte</sub></var> returns the value written by that
1551 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1552 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1553 values written. See the <a href="#ordering">Atomic Memory Ordering
1554 Constraints</a> section for additional constraints on how the choice
1556 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1559 <p><var>R</var> returns the value composed of the series of bytes it read.
1560 This implies that some bytes within the value may be <tt>undef</tt>
1561 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1562 defines the semantics of the operation; it doesn't mean that targets will
1563 emit more than one instruction to read the series of bytes.</p>
1565 <p>Note that in cases where none of the atomic intrinsics are used, this model
1566 places only one restriction on IR transformations on top of what is required
1567 for single-threaded execution: introducing a store to a byte which might not
1568 otherwise be stored is not allowed in general. (Specifically, in the case
1569 where another thread might write to and read from an address, introducing a
1570 store can change a load that may see exactly one write into a load that may
1571 see multiple writes.)</p>
1573 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1574 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1575 none of the backends currently in the tree fall into this category; however,
1576 there might be targets which care. If there are, we want a paragraph
1579 Targets may specify that stores narrower than a certain width are not
1580 available; on such a target, for the purposes of this model, treat any
1581 non-atomic write with an alignment or width less than the minimum width
1582 as if it writes to the relevant surrounding bytes.
1587 <!-- ======================================================================= -->
1589 <a name="ordering">Atomic Memory Ordering Constraints</a>
1594 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1595 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1596 <a href="#i_fence"><code>fence</code></a>,
1597 <a href="#i_load"><code>atomic load</code></a>, and
1598 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1599 that determines which other atomic instructions on the same address they
1600 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1601 but are somewhat more colloquial. If these descriptions aren't precise enough,
1602 check those specs (see spec references in the
1603 <a href="Atomic.html#introduction">atomics guide</a>).
1604 <a href="#i_fence"><code>fence</code></a> instructions
1605 treat these orderings somewhat differently since they don't take an address.
1606 See that instruction's documentation for details.</p>
1608 <p>For a simpler introduction to the ordering constraints, see the
1609 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1612 <dt><code>unordered</code></dt>
1613 <dd>The set of values that can be read is governed by the happens-before
1614 partial order. A value cannot be read unless some operation wrote it.
1615 This is intended to provide a guarantee strong enough to model Java's
1616 non-volatile shared variables. This ordering cannot be specified for
1617 read-modify-write operations; it is not strong enough to make them atomic
1618 in any interesting way.</dd>
1619 <dt><code>monotonic</code></dt>
1620 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1621 total order for modifications by <code>monotonic</code> operations on each
1622 address. All modification orders must be compatible with the happens-before
1623 order. There is no guarantee that the modification orders can be combined to
1624 a global total order for the whole program (and this often will not be
1625 possible). The read in an atomic read-modify-write operation
1626 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1627 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1628 reads the value in the modification order immediately before the value it
1629 writes. If one atomic read happens before another atomic read of the same
1630 address, the later read must see the same value or a later value in the
1631 address's modification order. This disallows reordering of
1632 <code>monotonic</code> (or stronger) operations on the same address. If an
1633 address is written <code>monotonic</code>ally by one thread, and other threads
1634 <code>monotonic</code>ally read that address repeatedly, the other threads must
1635 eventually see the write. This corresponds to the C++0x/C1x
1636 <code>memory_order_relaxed</code>.</dd>
1637 <dt><code>acquire</code></dt>
1638 <dd>In addition to the guarantees of <code>monotonic</code>,
1639 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1640 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1641 <dt><code>release</code></dt>
1642 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1643 writes a value which is subsequently read by an <code>acquire</code> operation,
1644 it <i>synchronizes-with</i> that operation. (This isn't a complete
1645 description; see the C++0x definition of a release sequence.) This corresponds
1646 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1647 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1648 <code>acquire</code> and <code>release</code> operation on its address.
1649 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1650 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1651 <dd>In addition to the guarantees of <code>acq_rel</code>
1652 (<code>acquire</code> for an operation which only reads, <code>release</code>
1653 for an operation which only writes), there is a global total order on all
1654 sequentially-consistent operations on all addresses, which is consistent with
1655 the <i>happens-before</i> partial order and with the modification orders of
1656 all the affected addresses. Each sequentially-consistent read sees the last
1657 preceding write to the same address in this global order. This corresponds
1658 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1661 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1662 it only <i>synchronizes with</i> or participates in modification and seq_cst
1663 total orderings with other operations running in the same thread (for example,
1664 in signal handlers).</p>
1670 <!-- *********************************************************************** -->
1671 <h2><a name="typesystem">Type System</a></h2>
1672 <!-- *********************************************************************** -->
1676 <p>The LLVM type system is one of the most important features of the
1677 intermediate representation. Being typed enables a number of optimizations
1678 to be performed on the intermediate representation directly, without having
1679 to do extra analyses on the side before the transformation. A strong type
1680 system makes it easier to read the generated code and enables novel analyses
1681 and transformations that are not feasible to perform on normal three address
1682 code representations.</p>
1684 <!-- ======================================================================= -->
1686 <a name="t_classifications">Type Classifications</a>
1691 <p>The types fall into a few useful classifications:</p>
1693 <table border="1" cellspacing="0" cellpadding="4">
1695 <tr><th>Classification</th><th>Types</th></tr>
1697 <td><a href="#t_integer">integer</a></td>
1698 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1701 <td><a href="#t_floating">floating point</a></td>
1702 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1705 <td><a name="t_firstclass">first class</a></td>
1706 <td><a href="#t_integer">integer</a>,
1707 <a href="#t_floating">floating point</a>,
1708 <a href="#t_pointer">pointer</a>,
1709 <a href="#t_vector">vector</a>,
1710 <a href="#t_struct">structure</a>,
1711 <a href="#t_array">array</a>,
1712 <a href="#t_label">label</a>,
1713 <a href="#t_metadata">metadata</a>.
1717 <td><a href="#t_primitive">primitive</a></td>
1718 <td><a href="#t_label">label</a>,
1719 <a href="#t_void">void</a>,
1720 <a href="#t_integer">integer</a>,
1721 <a href="#t_floating">floating point</a>,
1722 <a href="#t_x86mmx">x86mmx</a>,
1723 <a href="#t_metadata">metadata</a>.</td>
1726 <td><a href="#t_derived">derived</a></td>
1727 <td><a href="#t_array">array</a>,
1728 <a href="#t_function">function</a>,
1729 <a href="#t_pointer">pointer</a>,
1730 <a href="#t_struct">structure</a>,
1731 <a href="#t_vector">vector</a>,
1732 <a href="#t_opaque">opaque</a>.
1738 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1739 important. Values of these types are the only ones which can be produced by
1744 <!-- ======================================================================= -->
1746 <a name="t_primitive">Primitive Types</a>
1751 <p>The primitive types are the fundamental building blocks of the LLVM
1754 <!-- _______________________________________________________________________ -->
1756 <a name="t_integer">Integer Type</a>
1762 <p>The integer type is a very simple type that simply specifies an arbitrary
1763 bit width for the integer type desired. Any bit width from 1 bit to
1764 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1771 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1775 <table class="layout">
1777 <td class="left"><tt>i1</tt></td>
1778 <td class="left">a single-bit integer.</td>
1781 <td class="left"><tt>i32</tt></td>
1782 <td class="left">a 32-bit integer.</td>
1785 <td class="left"><tt>i1942652</tt></td>
1786 <td class="left">a really big integer of over 1 million bits.</td>
1792 <!-- _______________________________________________________________________ -->
1794 <a name="t_floating">Floating Point Types</a>
1801 <tr><th>Type</th><th>Description</th></tr>
1802 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1803 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1804 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1805 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1806 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1812 <!-- _______________________________________________________________________ -->
1814 <a name="t_x86mmx">X86mmx Type</a>
1820 <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>
1829 <!-- _______________________________________________________________________ -->
1831 <a name="t_void">Void Type</a>
1837 <p>The void type does not represent any value and has no size.</p>
1846 <!-- _______________________________________________________________________ -->
1848 <a name="t_label">Label Type</a>
1854 <p>The label type represents code labels.</p>
1863 <!-- _______________________________________________________________________ -->
1865 <a name="t_metadata">Metadata Type</a>
1871 <p>The metadata type represents embedded metadata. No derived types may be
1872 created from metadata except for <a href="#t_function">function</a>
1884 <!-- ======================================================================= -->
1886 <a name="t_derived">Derived Types</a>
1891 <p>The real power in LLVM comes from the derived types in the system. This is
1892 what allows a programmer to represent arrays, functions, pointers, and other
1893 useful types. Each of these types contain one or more element types which
1894 may be a primitive type, or another derived type. For example, it is
1895 possible to have a two dimensional array, using an array as the element type
1896 of another array.</p>
1901 <!-- _______________________________________________________________________ -->
1903 <a name="t_aggregate">Aggregate Types</a>
1908 <p>Aggregate Types are a subset of derived types that can contain multiple
1909 member types. <a href="#t_array">Arrays</a>,
1910 <a href="#t_struct">structs</a>, and <a href="#t_vector">vectors</a> are
1911 aggregate types.</p>
1915 <!-- _______________________________________________________________________ -->
1917 <a name="t_array">Array Type</a>
1923 <p>The array type is a very simple derived type that arranges elements
1924 sequentially in memory. The array type requires a size (number of elements)
1925 and an underlying data type.</p>
1929 [<# elements> x <elementtype>]
1932 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1933 be any type with a size.</p>
1936 <table class="layout">
1938 <td class="left"><tt>[40 x i32]</tt></td>
1939 <td class="left">Array of 40 32-bit integer values.</td>
1942 <td class="left"><tt>[41 x i32]</tt></td>
1943 <td class="left">Array of 41 32-bit integer values.</td>
1946 <td class="left"><tt>[4 x i8]</tt></td>
1947 <td class="left">Array of 4 8-bit integer values.</td>
1950 <p>Here are some examples of multidimensional arrays:</p>
1951 <table class="layout">
1953 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1954 <td class="left">3x4 array of 32-bit integer values.</td>
1957 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1958 <td class="left">12x10 array of single precision floating point values.</td>
1961 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1962 <td class="left">2x3x4 array of 16-bit integer values.</td>
1966 <p>There is no restriction on indexing beyond the end of the array implied by
1967 a static type (though there are restrictions on indexing beyond the bounds
1968 of an allocated object in some cases). This means that single-dimension
1969 'variable sized array' addressing can be implemented in LLVM with a zero
1970 length array type. An implementation of 'pascal style arrays' in LLVM could
1971 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1975 <!-- _______________________________________________________________________ -->
1977 <a name="t_function">Function Type</a>
1983 <p>The function type can be thought of as a function signature. It consists of
1984 a return type and a list of formal parameter types. The return type of a
1985 function type is a first class type or a void type.</p>
1989 <returntype> (<parameter list>)
1992 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1993 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1994 which indicates that the function takes a variable number of arguments.
1995 Variable argument functions can access their arguments with
1996 the <a href="#int_varargs">variable argument handling intrinsic</a>
1997 functions. '<tt><returntype></tt>' is any type except
1998 <a href="#t_label">label</a>.</p>
2001 <table class="layout">
2003 <td class="left"><tt>i32 (i32)</tt></td>
2004 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2006 </tr><tr class="layout">
2007 <td class="left"><tt>float (i16, i32 *) *
2009 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2010 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2011 returning <tt>float</tt>.
2013 </tr><tr class="layout">
2014 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2015 <td class="left">A vararg function that takes at least one
2016 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2017 which returns an integer. This is the signature for <tt>printf</tt> in
2020 </tr><tr class="layout">
2021 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2022 <td class="left">A function taking an <tt>i32</tt>, returning a
2023 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2030 <!-- _______________________________________________________________________ -->
2032 <a name="t_struct">Structure Type</a>
2038 <p>The structure type is used to represent a collection of data members together
2039 in memory. The elements of a structure may be any type that has a size.</p>
2041 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2042 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2043 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2044 Structures in registers are accessed using the
2045 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2046 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2048 <p>Structures may optionally be "packed" structures, which indicate that the
2049 alignment of the struct is one byte, and that there is no padding between
2050 the elements. In non-packed structs, padding between field types is inserted
2051 as defined by the TargetData string in the module, which is required to match
2052 what the underlying processor expects.</p>
2054 <p>Structures can either be "literal" or "identified". A literal structure is
2055 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2056 types are always defined at the top level with a name. Literal types are
2057 uniqued by their contents and can never be recursive or opaque since there is
2058 no way to write one. Identified types can be recursive, can be opaqued, and are
2064 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2065 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2069 <table class="layout">
2071 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2072 <td class="left">A triple of three <tt>i32</tt> values</td>
2075 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2076 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2077 second element is a <a href="#t_pointer">pointer</a> to a
2078 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2079 an <tt>i32</tt>.</td>
2082 <td class="left"><tt><{ i8, i32 }></tt></td>
2083 <td class="left">A packed struct known to be 5 bytes in size.</td>
2089 <!-- _______________________________________________________________________ -->
2091 <a name="t_opaque">Opaque Structure Types</a>
2097 <p>Opaque structure types are used to represent named structure types that do
2098 not have a body specified. This corresponds (for example) to the C notion of
2099 a forward declared structure.</p>
2108 <table class="layout">
2110 <td class="left"><tt>opaque</tt></td>
2111 <td class="left">An opaque type.</td>
2119 <!-- _______________________________________________________________________ -->
2121 <a name="t_pointer">Pointer Type</a>
2127 <p>The pointer type is used to specify memory locations.
2128 Pointers are commonly used to reference objects in memory.</p>
2130 <p>Pointer types may have an optional address space attribute defining the
2131 numbered address space where the pointed-to object resides. The default
2132 address space is number zero. The semantics of non-zero address
2133 spaces are target-specific.</p>
2135 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2136 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2144 <table class="layout">
2146 <td class="left"><tt>[4 x i32]*</tt></td>
2147 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2148 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2151 <td class="left"><tt>i32 (i32*) *</tt></td>
2152 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2153 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2157 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2158 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2159 that resides in address space #5.</td>
2165 <!-- _______________________________________________________________________ -->
2167 <a name="t_vector">Vector Type</a>
2173 <p>A vector type is a simple derived type that represents a vector of elements.
2174 Vector types are used when multiple primitive data are operated in parallel
2175 using a single instruction (SIMD). A vector type requires a size (number of
2176 elements) and an underlying primitive data type. Vector types are considered
2177 <a href="#t_firstclass">first class</a>.</p>
2181 < <# elements> x <elementtype> >
2184 <p>The number of elements is a constant integer value larger than 0; elementtype
2185 may be any integer or floating point type. Vectors of size zero are not
2186 allowed, and pointers are not allowed as the element type.</p>
2189 <table class="layout">
2191 <td class="left"><tt><4 x i32></tt></td>
2192 <td class="left">Vector of 4 32-bit integer values.</td>
2195 <td class="left"><tt><8 x float></tt></td>
2196 <td class="left">Vector of 8 32-bit floating-point values.</td>
2199 <td class="left"><tt><2 x i64></tt></td>
2200 <td class="left">Vector of 2 64-bit integer values.</td>
2208 <!-- *********************************************************************** -->
2209 <h2><a name="constants">Constants</a></h2>
2210 <!-- *********************************************************************** -->
2214 <p>LLVM has several different basic types of constants. This section describes
2215 them all and their syntax.</p>
2217 <!-- ======================================================================= -->
2219 <a name="simpleconstants">Simple Constants</a>
2225 <dt><b>Boolean constants</b></dt>
2226 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2227 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2229 <dt><b>Integer constants</b></dt>
2230 <dd>Standard integers (such as '4') are constants of
2231 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2232 with integer types.</dd>
2234 <dt><b>Floating point constants</b></dt>
2235 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2236 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2237 notation (see below). The assembler requires the exact decimal value of a
2238 floating-point constant. For example, the assembler accepts 1.25 but
2239 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2240 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2242 <dt><b>Null pointer constants</b></dt>
2243 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2244 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2247 <p>The one non-intuitive notation for constants is the hexadecimal form of
2248 floating point constants. For example, the form '<tt>double
2249 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2250 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2251 constants are required (and the only time that they are generated by the
2252 disassembler) is when a floating point constant must be emitted but it cannot
2253 be represented as a decimal floating point number in a reasonable number of
2254 digits. For example, NaN's, infinities, and other special values are
2255 represented in their IEEE hexadecimal format so that assembly and disassembly
2256 do not cause any bits to change in the constants.</p>
2258 <p>When using the hexadecimal form, constants of types float and double are
2259 represented using the 16-digit form shown above (which matches the IEEE754
2260 representation for double); float values must, however, be exactly
2261 representable as IEE754 single precision. Hexadecimal format is always used
2262 for long double, and there are three forms of long double. The 80-bit format
2263 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2264 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2265 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2266 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2267 currently supported target uses this format. Long doubles will only work if
2268 they match the long double format on your target. All hexadecimal formats
2269 are big-endian (sign bit at the left).</p>
2271 <p>There are no constants of type x86mmx.</p>
2274 <!-- ======================================================================= -->
2276 <a name="aggregateconstants"></a> <!-- old anchor -->
2277 <a name="complexconstants">Complex Constants</a>
2282 <p>Complex constants are a (potentially recursive) combination of simple
2283 constants and smaller complex constants.</p>
2286 <dt><b>Structure constants</b></dt>
2287 <dd>Structure constants are represented with notation similar to structure
2288 type definitions (a comma separated list of elements, surrounded by braces
2289 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2290 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2291 Structure constants must have <a href="#t_struct">structure type</a>, and
2292 the number and types of elements must match those specified by the
2295 <dt><b>Array constants</b></dt>
2296 <dd>Array constants are represented with notation similar to array type
2297 definitions (a comma separated list of elements, surrounded by square
2298 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2299 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2300 the number and types of elements must match those specified by the
2303 <dt><b>Vector constants</b></dt>
2304 <dd>Vector constants are represented with notation similar to vector type
2305 definitions (a comma separated list of elements, surrounded by
2306 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2307 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2308 have <a href="#t_vector">vector type</a>, and the number and types of
2309 elements must match those specified by the type.</dd>
2311 <dt><b>Zero initialization</b></dt>
2312 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2313 value to zero of <em>any</em> type, including scalar and
2314 <a href="#t_aggregate">aggregate</a> types.
2315 This is often used to avoid having to print large zero initializers
2316 (e.g. for large arrays) and is always exactly equivalent to using explicit
2317 zero initializers.</dd>
2319 <dt><b>Metadata node</b></dt>
2320 <dd>A metadata node is a structure-like constant with
2321 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2322 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2323 be interpreted as part of the instruction stream, metadata is a place to
2324 attach additional information such as debug info.</dd>
2329 <!-- ======================================================================= -->
2331 <a name="globalconstants">Global Variable and Function Addresses</a>
2336 <p>The addresses of <a href="#globalvars">global variables</a>
2337 and <a href="#functionstructure">functions</a> are always implicitly valid
2338 (link-time) constants. These constants are explicitly referenced when
2339 the <a href="#identifiers">identifier for the global</a> is used and always
2340 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2341 legal LLVM file:</p>
2343 <pre class="doc_code">
2346 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2351 <!-- ======================================================================= -->
2353 <a name="undefvalues">Undefined Values</a>
2358 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2359 indicates that the user of the value may receive an unspecified bit-pattern.
2360 Undefined values may be of any type (other than '<tt>label</tt>'
2361 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2363 <p>Undefined values are useful because they indicate to the compiler that the
2364 program is well defined no matter what value is used. This gives the
2365 compiler more freedom to optimize. Here are some examples of (potentially
2366 surprising) transformations that are valid (in pseudo IR):</p>
2369 <pre class="doc_code">
2379 <p>This is safe because all of the output bits are affected by the undef bits.
2380 Any output bit can have a zero or one depending on the input bits.</p>
2382 <pre class="doc_code">
2393 <p>These logical operations have bits that are not always affected by the input.
2394 For example, if <tt>%X</tt> has a zero bit, then the output of the
2395 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2396 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2397 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2398 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2399 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2400 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2401 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2403 <pre class="doc_code">
2404 %A = select undef, %X, %Y
2405 %B = select undef, 42, %Y
2406 %C = select %X, %Y, undef
2417 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2418 branch) conditions can go <em>either way</em>, but they have to come from one
2419 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2420 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2421 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2422 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2423 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2426 <pre class="doc_code">
2427 %A = xor undef, undef
2445 <p>This example points out that two '<tt>undef</tt>' operands are not
2446 necessarily the same. This can be surprising to people (and also matches C
2447 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2448 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2449 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2450 its value over its "live range". This is true because the variable doesn't
2451 actually <em>have a live range</em>. Instead, the value is logically read
2452 from arbitrary registers that happen to be around when needed, so the value
2453 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2454 need to have the same semantics or the core LLVM "replace all uses with"
2455 concept would not hold.</p>
2457 <pre class="doc_code">
2465 <p>These examples show the crucial difference between an <em>undefined
2466 value</em> and <em>undefined behavior</em>. An undefined value (like
2467 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2468 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2469 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2470 defined on SNaN's. However, in the second example, we can make a more
2471 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2472 arbitrary value, we are allowed to assume that it could be zero. Since a
2473 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2474 the operation does not execute at all. This allows us to delete the divide and
2475 all code after it. Because the undefined operation "can't happen", the
2476 optimizer can assume that it occurs in dead code.</p>
2478 <pre class="doc_code">
2479 a: store undef -> %X
2480 b: store %X -> undef
2486 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2487 undefined value can be assumed to not have any effect; we can assume that the
2488 value is overwritten with bits that happen to match what was already there.
2489 However, a store <em>to</em> an undefined location could clobber arbitrary
2490 memory, therefore, it has undefined behavior.</p>
2494 <!-- ======================================================================= -->
2496 <a name="trapvalues">Trap Values</a>
2501 <p>Trap values are similar to <a href="#undefvalues">undef values</a>, however
2502 instead of representing an unspecified bit pattern, they represent the
2503 fact that an instruction or constant expression which cannot evoke side
2504 effects has nevertheless detected a condition which results in undefined
2507 <p>There is currently no way of representing a trap value in the IR; they
2508 only exist when produced by operations such as
2509 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2511 <p>Trap value behavior is defined in terms of value <i>dependence</i>:</p>
2514 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2515 their operands.</li>
2517 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2518 to their dynamic predecessor basic block.</li>
2520 <li>Function arguments depend on the corresponding actual argument values in
2521 the dynamic callers of their functions.</li>
2523 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2524 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2525 control back to them.</li>
2527 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2528 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_unwind"><tt>unwind</tt></a>,
2529 or exception-throwing call instructions that dynamically transfer control
2532 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2533 referenced memory addresses, following the order in the IR
2534 (including loads and stores implied by intrinsics such as
2535 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2537 <!-- TODO: In the case of multiple threads, this only applies if the store
2538 "happens-before" the load or store. -->
2540 <!-- TODO: floating-point exception state -->
2542 <li>An instruction with externally visible side effects depends on the most
2543 recent preceding instruction with externally visible side effects, following
2544 the order in the IR. (This includes
2545 <a href="#volatile">volatile operations</a>.)</li>
2547 <li>An instruction <i>control-depends</i> on a
2548 <a href="#terminators">terminator instruction</a>
2549 if the terminator instruction has multiple successors and the instruction
2550 is always executed when control transfers to one of the successors, and
2551 may not be executed when control is transferred to another.</li>
2553 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2554 instruction if the set of instructions it otherwise depends on would be
2555 different if the terminator had transferred control to a different
2558 <li>Dependence is transitive.</li>
2562 <p>Whenever a trap value is generated, all values which depend on it evaluate
2563 to trap. If they have side effects, the evoke their side effects as if each
2564 operand with a trap value were undef. If they have externally-visible side
2565 effects, the behavior is undefined.</p>
2567 <p>Here are some examples:</p>
2569 <pre class="doc_code">
2571 %trap = sub nuw i32 0, 1 ; Results in a trap value.
2572 %still_trap = and i32 %trap, 0 ; Whereas (and i32 undef, 0) would return 0.
2573 %trap_yet_again = getelementptr i32* @h, i32 %still_trap
2574 store i32 0, i32* %trap_yet_again ; undefined behavior
2576 store i32 %trap, i32* @g ; Trap value conceptually stored to memory.
2577 %trap2 = load i32* @g ; Returns a trap value, not just undef.
2579 volatile store i32 %trap, i32* @g ; External observation; undefined behavior.
2581 %narrowaddr = bitcast i32* @g to i16*
2582 %wideaddr = bitcast i32* @g to i64*
2583 %trap3 = load i16* %narrowaddr ; Returns a trap value.
2584 %trap4 = load i64* %wideaddr ; Returns a trap value.
2586 %cmp = icmp slt i32 %trap, 0 ; Returns a trap value.
2587 br i1 %cmp, label %true, label %end ; Branch to either destination.
2590 volatile store i32 0, i32* @g ; This is control-dependent on %cmp, so
2591 ; it has undefined behavior.
2595 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2596 ; Both edges into this PHI are
2597 ; control-dependent on %cmp, so this
2598 ; always results in a trap value.
2600 volatile store i32 0, i32* @g ; This would depend on the store in %true
2601 ; if %cmp is true, or the store in %entry
2602 ; otherwise, so this is undefined behavior.
2604 br i1 %cmp, label %second_true, label %second_end
2605 ; The same branch again, but this time the
2606 ; true block doesn't have side effects.
2613 volatile store i32 0, i32* @g ; This time, the instruction always depends
2614 ; on the store in %end. Also, it is
2615 ; control-equivalent to %end, so this is
2616 ; well-defined (again, ignoring earlier
2617 ; undefined behavior in this example).
2622 <!-- ======================================================================= -->
2624 <a name="blockaddress">Addresses of Basic Blocks</a>
2629 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2631 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2632 basic block in the specified function, and always has an i8* type. Taking
2633 the address of the entry block is illegal.</p>
2635 <p>This value only has defined behavior when used as an operand to the
2636 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2637 comparisons against null. Pointer equality tests between labels addresses
2638 results in undefined behavior — though, again, comparison against null
2639 is ok, and no label is equal to the null pointer. This may be passed around
2640 as an opaque pointer sized value as long as the bits are not inspected. This
2641 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2642 long as the original value is reconstituted before the <tt>indirectbr</tt>
2645 <p>Finally, some targets may provide defined semantics when using the value as
2646 the operand to an inline assembly, but that is target specific.</p>
2651 <!-- ======================================================================= -->
2653 <a name="constantexprs">Constant Expressions</a>
2658 <p>Constant expressions are used to allow expressions involving other constants
2659 to be used as constants. Constant expressions may be of
2660 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2661 operation that does not have side effects (e.g. load and call are not
2662 supported). The following is the syntax for constant expressions:</p>
2665 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2666 <dd>Truncate a constant to another type. The bit size of CST must be larger
2667 than the bit size of TYPE. Both types must be integers.</dd>
2669 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2670 <dd>Zero extend a constant to another type. The bit size of CST must be
2671 smaller than the bit size of TYPE. Both types must be integers.</dd>
2673 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2674 <dd>Sign extend a constant to another type. The bit size of CST must be
2675 smaller than the bit size of TYPE. Both types must be integers.</dd>
2677 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2678 <dd>Truncate a floating point constant to another floating point type. The
2679 size of CST must be larger than the size of TYPE. Both types must be
2680 floating point.</dd>
2682 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2683 <dd>Floating point extend a constant to another type. The size of CST must be
2684 smaller or equal to the size of TYPE. Both types must be floating
2687 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2688 <dd>Convert a floating point constant to the corresponding unsigned integer
2689 constant. TYPE must be a scalar or vector integer type. CST must be of
2690 scalar or vector floating point type. Both CST and TYPE must be scalars,
2691 or vectors of the same number of elements. If the value won't fit in the
2692 integer type, the results are undefined.</dd>
2694 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2695 <dd>Convert a floating point constant to the corresponding signed integer
2696 constant. TYPE must be a scalar or vector integer type. CST must be of
2697 scalar or vector floating point type. Both CST and TYPE must be scalars,
2698 or vectors of the same number of elements. If the value won't fit in the
2699 integer type, the results are undefined.</dd>
2701 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2702 <dd>Convert an unsigned integer constant to the corresponding floating point
2703 constant. TYPE must be a scalar or vector floating point type. CST must be
2704 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2705 vectors of the same number of elements. If the value won't fit in the
2706 floating point type, the results are undefined.</dd>
2708 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2709 <dd>Convert a signed integer constant to the corresponding floating point
2710 constant. TYPE must be a scalar or vector floating point type. CST must be
2711 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2712 vectors of the same number of elements. If the value won't fit in the
2713 floating point type, the results are undefined.</dd>
2715 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2716 <dd>Convert a pointer typed constant to the corresponding integer constant
2717 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2718 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2719 make it fit in <tt>TYPE</tt>.</dd>
2721 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2722 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2723 type. CST must be of integer type. The CST value is zero extended,
2724 truncated, or unchanged to make it fit in a pointer size. This one is
2725 <i>really</i> dangerous!</dd>
2727 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2728 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2729 are the same as those for the <a href="#i_bitcast">bitcast
2730 instruction</a>.</dd>
2732 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2733 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2734 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2735 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2736 instruction, the index list may have zero or more indexes, which are
2737 required to make sense for the type of "CSTPTR".</dd>
2739 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2740 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2742 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2743 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2745 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2746 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2748 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2749 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2752 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2753 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2756 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2757 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2760 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2761 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2762 constants. The index list is interpreted in a similar manner as indices in
2763 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2764 index value must be specified.</dd>
2766 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2767 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2768 constants. The index list is interpreted in a similar manner as indices in
2769 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2770 index value must be specified.</dd>
2772 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2773 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2774 be any of the <a href="#binaryops">binary</a>
2775 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2776 on operands are the same as those for the corresponding instruction
2777 (e.g. no bitwise operations on floating point values are allowed).</dd>
2784 <!-- *********************************************************************** -->
2785 <h2><a name="othervalues">Other Values</a></h2>
2786 <!-- *********************************************************************** -->
2788 <!-- ======================================================================= -->
2790 <a name="inlineasm">Inline Assembler Expressions</a>
2795 <p>LLVM supports inline assembler expressions (as opposed
2796 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2797 a special value. This value represents the inline assembler as a string
2798 (containing the instructions to emit), a list of operand constraints (stored
2799 as a string), a flag that indicates whether or not the inline asm
2800 expression has side effects, and a flag indicating whether the function
2801 containing the asm needs to align its stack conservatively. An example
2802 inline assembler expression is:</p>
2804 <pre class="doc_code">
2805 i32 (i32) asm "bswap $0", "=r,r"
2808 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2809 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2812 <pre class="doc_code">
2813 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2816 <p>Inline asms with side effects not visible in the constraint list must be
2817 marked as having side effects. This is done through the use of the
2818 '<tt>sideeffect</tt>' keyword, like so:</p>
2820 <pre class="doc_code">
2821 call void asm sideeffect "eieio", ""()
2824 <p>In some cases inline asms will contain code that will not work unless the
2825 stack is aligned in some way, such as calls or SSE instructions on x86,
2826 yet will not contain code that does that alignment within the asm.
2827 The compiler should make conservative assumptions about what the asm might
2828 contain and should generate its usual stack alignment code in the prologue
2829 if the '<tt>alignstack</tt>' keyword is present:</p>
2831 <pre class="doc_code">
2832 call void asm alignstack "eieio", ""()
2835 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2838 <p>TODO: The format of the asm and constraints string still need to be
2839 documented here. Constraints on what can be done (e.g. duplication, moving,
2840 etc need to be documented). This is probably best done by reference to
2841 another document that covers inline asm from a holistic perspective.</p>
2844 <a name="inlineasm_md">Inline Asm Metadata</a>
2849 <p>The call instructions that wrap inline asm nodes may have a "!srcloc" MDNode
2850 attached to it that contains a list of constant integers. If present, the
2851 code generator will use the integer as the location cookie value when report
2852 errors through the LLVMContext error reporting mechanisms. This allows a
2853 front-end to correlate backend errors that occur with inline asm back to the
2854 source code that produced it. For example:</p>
2856 <pre class="doc_code">
2857 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2859 !42 = !{ i32 1234567 }
2862 <p>It is up to the front-end to make sense of the magic numbers it places in the
2863 IR. If the MDNode contains multiple constants, the code generator will use
2864 the one that corresponds to the line of the asm that the error occurs on.</p>
2870 <!-- ======================================================================= -->
2872 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2877 <p>LLVM IR allows metadata to be attached to instructions in the program that
2878 can convey extra information about the code to the optimizers and code
2879 generator. One example application of metadata is source-level debug
2880 information. There are two metadata primitives: strings and nodes. All
2881 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2882 preceding exclamation point ('<tt>!</tt>').</p>
2884 <p>A metadata string is a string surrounded by double quotes. It can contain
2885 any character by escaping non-printable characters with "\xx" where "xx" is
2886 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2888 <p>Metadata nodes are represented with notation similar to structure constants
2889 (a comma separated list of elements, surrounded by braces and preceded by an
2890 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2891 10}</tt>". Metadata nodes can have any values as their operand.</p>
2893 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2894 metadata nodes, which can be looked up in the module symbol table. For
2895 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2897 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2898 function is using two metadata arguments.</p>
2900 <div class="doc_code">
2902 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2906 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2907 attached with <tt>add</tt> instruction using <tt>!dbg</tt> identifier.</p>
2909 <div class="doc_code">
2911 %indvar.next = add i64 %indvar, 1, !dbg !21
2919 <!-- *********************************************************************** -->
2921 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2923 <!-- *********************************************************************** -->
2925 <p>LLVM has a number of "magic" global variables that contain data that affect
2926 code generation or other IR semantics. These are documented here. All globals
2927 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2928 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2931 <!-- ======================================================================= -->
2933 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2938 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2939 href="#linkage_appending">appending linkage</a>. This array contains a list of
2940 pointers to global variables and functions which may optionally have a pointer
2941 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2947 @llvm.used = appending global [2 x i8*] [
2949 i8* bitcast (i32* @Y to i8*)
2950 ], section "llvm.metadata"
2953 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2954 compiler, assembler, and linker are required to treat the symbol as if there is
2955 a reference to the global that it cannot see. For example, if a variable has
2956 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2957 list, it cannot be deleted. This is commonly used to represent references from
2958 inline asms and other things the compiler cannot "see", and corresponds to
2959 "attribute((used))" in GNU C.</p>
2961 <p>On some targets, the code generator must emit a directive to the assembler or
2962 object file to prevent the assembler and linker from molesting the symbol.</p>
2966 <!-- ======================================================================= -->
2968 <a name="intg_compiler_used">
2969 The '<tt>llvm.compiler.used</tt>' Global Variable
2975 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2976 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2977 touching the symbol. On targets that support it, this allows an intelligent
2978 linker to optimize references to the symbol without being impeded as it would be
2979 by <tt>@llvm.used</tt>.</p>
2981 <p>This is a rare construct that should only be used in rare circumstances, and
2982 should not be exposed to source languages.</p>
2986 <!-- ======================================================================= -->
2988 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2993 %0 = type { i32, void ()* }
2994 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2996 <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.
3001 <!-- ======================================================================= -->
3003 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3008 %0 = type { i32, void ()* }
3009 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3012 <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.
3019 <!-- *********************************************************************** -->
3020 <h2><a name="instref">Instruction Reference</a></h2>
3021 <!-- *********************************************************************** -->
3025 <p>The LLVM instruction set consists of several different classifications of
3026 instructions: <a href="#terminators">terminator
3027 instructions</a>, <a href="#binaryops">binary instructions</a>,
3028 <a href="#bitwiseops">bitwise binary instructions</a>,
3029 <a href="#memoryops">memory instructions</a>, and
3030 <a href="#otherops">other instructions</a>.</p>
3032 <!-- ======================================================================= -->
3034 <a name="terminators">Terminator Instructions</a>
3039 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3040 in a program ends with a "Terminator" instruction, which indicates which
3041 block should be executed after the current block is finished. These
3042 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3043 control flow, not values (the one exception being the
3044 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3046 <p>The terminator instructions are:
3047 '<a href="#i_ret"><tt>ret</tt></a>',
3048 '<a href="#i_br"><tt>br</tt></a>',
3049 '<a href="#i_switch"><tt>switch</tt></a>',
3050 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3051 '<a href="#i_invoke"><tt>invoke</tt></a>',
3052 '<a href="#i_unwind"><tt>unwind</tt></a>',
3053 '<a href="#i_resume"><tt>resume</tt></a>', and
3054 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3056 <!-- _______________________________________________________________________ -->
3058 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3065 ret <type> <value> <i>; Return a value from a non-void function</i>
3066 ret void <i>; Return from void function</i>
3070 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3071 a value) from a function back to the caller.</p>
3073 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3074 value and then causes control flow, and one that just causes control flow to
3078 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3079 return value. The type of the return value must be a
3080 '<a href="#t_firstclass">first class</a>' type.</p>
3082 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3083 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3084 value or a return value with a type that does not match its type, or if it
3085 has a void return type and contains a '<tt>ret</tt>' instruction with a
3089 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3090 the calling function's context. If the caller is a
3091 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3092 instruction after the call. If the caller was an
3093 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3094 the beginning of the "normal" destination block. If the instruction returns
3095 a value, that value shall set the call or invoke instruction's return
3100 ret i32 5 <i>; Return an integer value of 5</i>
3101 ret void <i>; Return from a void function</i>
3102 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3106 <!-- _______________________________________________________________________ -->
3108 <a name="i_br">'<tt>br</tt>' Instruction</a>
3115 br i1 <cond>, label <iftrue>, label <iffalse>
3116 br label <dest> <i>; Unconditional branch</i>
3120 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3121 different basic block in the current function. There are two forms of this
3122 instruction, corresponding to a conditional branch and an unconditional
3126 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3127 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3128 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3132 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3133 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3134 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3135 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3140 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3141 br i1 %cond, label %IfEqual, label %IfUnequal
3143 <a href="#i_ret">ret</a> i32 1
3145 <a href="#i_ret">ret</a> i32 0
3150 <!-- _______________________________________________________________________ -->
3152 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3159 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3163 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3164 several different places. It is a generalization of the '<tt>br</tt>'
3165 instruction, allowing a branch to occur to one of many possible
3169 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3170 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3171 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3172 The table is not allowed to contain duplicate constant entries.</p>
3175 <p>The <tt>switch</tt> instruction specifies a table of values and
3176 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3177 is searched for the given value. If the value is found, control flow is
3178 transferred to the corresponding destination; otherwise, control flow is
3179 transferred to the default destination.</p>
3181 <h5>Implementation:</h5>
3182 <p>Depending on properties of the target machine and the particular
3183 <tt>switch</tt> instruction, this instruction may be code generated in
3184 different ways. For example, it could be generated as a series of chained
3185 conditional branches or with a lookup table.</p>
3189 <i>; Emulate a conditional br instruction</i>
3190 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3191 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3193 <i>; Emulate an unconditional br instruction</i>
3194 switch i32 0, label %dest [ ]
3196 <i>; Implement a jump table:</i>
3197 switch i32 %val, label %otherwise [ i32 0, label %onzero
3199 i32 2, label %ontwo ]
3205 <!-- _______________________________________________________________________ -->
3207 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3214 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3219 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3220 within the current function, whose address is specified by
3221 "<tt>address</tt>". Address must be derived from a <a
3222 href="#blockaddress">blockaddress</a> constant.</p>
3226 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3227 rest of the arguments indicate the full set of possible destinations that the
3228 address may point to. Blocks are allowed to occur multiple times in the
3229 destination list, though this isn't particularly useful.</p>
3231 <p>This destination list is required so that dataflow analysis has an accurate
3232 understanding of the CFG.</p>
3236 <p>Control transfers to the block specified in the address argument. All
3237 possible destination blocks must be listed in the label list, otherwise this
3238 instruction has undefined behavior. This implies that jumps to labels
3239 defined in other functions have undefined behavior as well.</p>
3241 <h5>Implementation:</h5>
3243 <p>This is typically implemented with a jump through a register.</p>
3247 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3253 <!-- _______________________________________________________________________ -->
3255 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3262 <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>]
3263 to label <normal label> unwind label <exception label>
3267 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3268 function, with the possibility of control flow transfer to either the
3269 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3270 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3271 control flow will return to the "normal" label. If the callee (or any
3272 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
3273 instruction, control is interrupted and continued at the dynamically nearest
3274 "exception" label.</p>
3276 <p>The '<tt>exception</tt>' label is a
3277 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3278 exception. As such, '<tt>exception</tt>' label is required to have the
3279 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3280 the information about about the behavior of the program after unwinding
3281 happens, as its first non-PHI instruction. The restrictions on the
3282 "<tt>landingpad</tt>" instruction's tightly couples it to the
3283 "<tt>invoke</tt>" instruction, so that the important information contained
3284 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3288 <p>This instruction requires several arguments:</p>
3291 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3292 convention</a> the call should use. If none is specified, the call
3293 defaults to using C calling conventions.</li>
3295 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3296 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3297 '<tt>inreg</tt>' attributes are valid here.</li>
3299 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3300 function value being invoked. In most cases, this is a direct function
3301 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3302 off an arbitrary pointer to function value.</li>
3304 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3305 function to be invoked. </li>
3307 <li>'<tt>function args</tt>': argument list whose types match the function
3308 signature argument types and parameter attributes. All arguments must be
3309 of <a href="#t_firstclass">first class</a> type. If the function
3310 signature indicates the function accepts a variable number of arguments,
3311 the extra arguments can be specified.</li>
3313 <li>'<tt>normal label</tt>': the label reached when the called function
3314 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3316 <li>'<tt>exception label</tt>': the label reached when a callee returns with
3317 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
3319 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3320 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3321 '<tt>readnone</tt>' attributes are valid here.</li>
3325 <p>This instruction is designed to operate as a standard
3326 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3327 primary difference is that it establishes an association with a label, which
3328 is used by the runtime library to unwind the stack.</p>
3330 <p>This instruction is used in languages with destructors to ensure that proper
3331 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3332 exception. Additionally, this is important for implementation of
3333 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3335 <p>For the purposes of the SSA form, the definition of the value returned by the
3336 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3337 block to the "normal" label. If the callee unwinds then no return value is
3340 <p>Note that the code generator does not yet completely support unwind, and
3341 that the invoke/unwind semantics are likely to change in future versions.</p>
3345 %retval = invoke i32 @Test(i32 15) to label %Continue
3346 unwind label %TestCleanup <i>; {i32}:retval set</i>
3347 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3348 unwind label %TestCleanup <i>; {i32}:retval set</i>
3353 <!-- _______________________________________________________________________ -->
3356 <a name="i_unwind">'<tt>unwind</tt>' Instruction</a>
3367 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
3368 at the first callee in the dynamic call stack which used
3369 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
3370 This is primarily used to implement exception handling.</p>
3373 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
3374 immediately halt. The dynamic call stack is then searched for the
3375 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
3376 Once found, execution continues at the "exceptional" destination block
3377 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
3378 instruction in the dynamic call chain, undefined behavior results.</p>
3380 <p>Note that the code generator does not yet completely support unwind, and
3381 that the invoke/unwind semantics are likely to change in future versions.</p>
3385 <!-- _______________________________________________________________________ -->
3388 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3395 resume <type> <value>
3399 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3403 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3404 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3408 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3409 (in-flight) exception whose unwinding was interrupted with
3410 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3414 resume { i8*, i32 } %exn
3419 <!-- _______________________________________________________________________ -->
3422 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3433 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3434 instruction is used to inform the optimizer that a particular portion of the
3435 code is not reachable. This can be used to indicate that the code after a
3436 no-return function cannot be reached, and other facts.</p>
3439 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3445 <!-- ======================================================================= -->
3447 <a name="binaryops">Binary Operations</a>
3452 <p>Binary operators are used to do most of the computation in a program. They
3453 require two operands of the same type, execute an operation on them, and
3454 produce a single value. The operands might represent multiple data, as is
3455 the case with the <a href="#t_vector">vector</a> data type. The result value
3456 has the same type as its operands.</p>
3458 <p>There are several different binary operators:</p>
3460 <!-- _______________________________________________________________________ -->
3462 <a name="i_add">'<tt>add</tt>' Instruction</a>
3469 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3470 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3471 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3472 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3476 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3479 <p>The two arguments to the '<tt>add</tt>' instruction must
3480 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3481 integer values. Both arguments must have identical types.</p>
3484 <p>The value produced is the integer sum of the two operands.</p>
3486 <p>If the sum has unsigned overflow, the result returned is the mathematical
3487 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3489 <p>Because LLVM integers use a two's complement representation, this instruction
3490 is appropriate for both signed and unsigned integers.</p>
3492 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3493 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3494 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3495 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3496 respectively, occurs.</p>
3500 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3505 <!-- _______________________________________________________________________ -->
3507 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3514 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3518 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3521 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3522 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3523 floating point values. Both arguments must have identical types.</p>
3526 <p>The value produced is the floating point sum of the two operands.</p>
3530 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3535 <!-- _______________________________________________________________________ -->
3537 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3544 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3545 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3546 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3547 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3551 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3554 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3555 '<tt>neg</tt>' instruction present in most other intermediate
3556 representations.</p>
3559 <p>The two arguments to the '<tt>sub</tt>' instruction must
3560 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3561 integer values. Both arguments must have identical types.</p>
3564 <p>The value produced is the integer difference of the two operands.</p>
3566 <p>If the difference has unsigned overflow, the result returned is the
3567 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3570 <p>Because LLVM integers use a two's complement representation, this instruction
3571 is appropriate for both signed and unsigned integers.</p>
3573 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3574 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3575 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3576 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3577 respectively, occurs.</p>
3581 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3582 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3587 <!-- _______________________________________________________________________ -->
3589 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3596 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3600 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3603 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3604 '<tt>fneg</tt>' instruction present in most other intermediate
3605 representations.</p>
3608 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3609 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3610 floating point values. Both arguments must have identical types.</p>
3613 <p>The value produced is the floating point difference of the two operands.</p>
3617 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3618 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3623 <!-- _______________________________________________________________________ -->
3625 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3632 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3633 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3634 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3635 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3639 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3642 <p>The two arguments to the '<tt>mul</tt>' instruction must
3643 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3644 integer values. Both arguments must have identical types.</p>
3647 <p>The value produced is the integer product of the two operands.</p>
3649 <p>If the result of the multiplication has unsigned overflow, the result
3650 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3651 width of the result.</p>
3653 <p>Because LLVM integers use a two's complement representation, and the result
3654 is the same width as the operands, this instruction returns the correct
3655 result for both signed and unsigned integers. If a full product
3656 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3657 be sign-extended or zero-extended as appropriate to the width of the full
3660 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3661 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3662 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3663 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3664 respectively, occurs.</p>
3668 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3673 <!-- _______________________________________________________________________ -->
3675 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3682 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3686 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3689 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3690 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3691 floating point values. Both arguments must have identical types.</p>
3694 <p>The value produced is the floating point product of the two operands.</p>
3698 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3703 <!-- _______________________________________________________________________ -->
3705 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
3712 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3713 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3717 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3720 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3721 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3722 values. Both arguments must have identical types.</p>
3725 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3727 <p>Note that unsigned integer division and signed integer division are distinct
3728 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3730 <p>Division by zero leads to undefined behavior.</p>
3732 <p>If the <tt>exact</tt> keyword is present, the result value of the
3733 <tt>udiv</tt> is a <a href="#trapvalues">trap value</a> if %op1 is not a
3734 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
3739 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3744 <!-- _______________________________________________________________________ -->
3746 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
3753 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3754 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3758 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3761 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3762 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3763 values. Both arguments must have identical types.</p>
3766 <p>The value produced is the signed integer quotient of the two operands rounded
3769 <p>Note that signed integer division and unsigned integer division are distinct
3770 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3772 <p>Division by zero leads to undefined behavior. Overflow also leads to
3773 undefined behavior; this is a rare case, but can occur, for example, by doing
3774 a 32-bit division of -2147483648 by -1.</p>
3776 <p>If the <tt>exact</tt> keyword is present, the result value of the
3777 <tt>sdiv</tt> is a <a href="#trapvalues">trap value</a> if the result would
3782 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3787 <!-- _______________________________________________________________________ -->
3789 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
3796 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3800 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3803 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3804 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3805 floating point values. Both arguments must have identical types.</p>
3808 <p>The value produced is the floating point quotient of the two operands.</p>
3812 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3817 <!-- _______________________________________________________________________ -->
3819 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3826 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3830 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3831 division of its two arguments.</p>
3834 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3835 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3836 values. Both arguments must have identical types.</p>
3839 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3840 This instruction always performs an unsigned division to get the
3843 <p>Note that unsigned integer remainder and signed integer remainder are
3844 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3846 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3850 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3855 <!-- _______________________________________________________________________ -->
3857 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3864 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3868 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3869 division of its two operands. This instruction can also take
3870 <a href="#t_vector">vector</a> versions of the values in which case the
3871 elements must be integers.</p>
3874 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3875 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3876 values. Both arguments must have identical types.</p>
3879 <p>This instruction returns the <i>remainder</i> of a division (where the result
3880 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
3881 <i>modulo</i> operator (where the result is either zero or has the same sign
3882 as the divisor, <tt>op2</tt>) of a value.
3883 For more information about the difference,
3884 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3885 Math Forum</a>. For a table of how this is implemented in various languages,
3886 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3887 Wikipedia: modulo operation</a>.</p>
3889 <p>Note that signed integer remainder and unsigned integer remainder are
3890 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3892 <p>Taking the remainder of a division by zero leads to undefined behavior.
3893 Overflow also leads to undefined behavior; this is a rare case, but can
3894 occur, for example, by taking the remainder of a 32-bit division of
3895 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3896 lets srem be implemented using instructions that return both the result of
3897 the division and the remainder.)</p>
3901 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3906 <!-- _______________________________________________________________________ -->
3908 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
3915 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3919 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3920 its two operands.</p>
3923 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3924 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3925 floating point values. Both arguments must have identical types.</p>
3928 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3929 has the same sign as the dividend.</p>
3933 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3940 <!-- ======================================================================= -->
3942 <a name="bitwiseops">Bitwise Binary Operations</a>
3947 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3948 program. They are generally very efficient instructions and can commonly be
3949 strength reduced from other instructions. They require two operands of the
3950 same type, execute an operation on them, and produce a single value. The
3951 resulting value is the same type as its operands.</p>
3953 <!-- _______________________________________________________________________ -->
3955 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
3962 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3963 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3964 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3965 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3969 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3970 a specified number of bits.</p>
3973 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3974 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3975 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3978 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3979 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3980 is (statically or dynamically) negative or equal to or larger than the number
3981 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3982 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3983 shift amount in <tt>op2</tt>.</p>
3985 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
3986 <a href="#trapvalues">trap value</a> if it shifts out any non-zero bits. If
3987 the <tt>nsw</tt> keyword is present, then the shift produces a
3988 <a href="#trapvalues">trap value</a> if it shifts out any bits that disagree
3989 with the resultant sign bit. As such, NUW/NSW have the same semantics as
3990 they would if the shift were expressed as a mul instruction with the same
3991 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
3995 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3996 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3997 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3998 <result> = shl i32 1, 32 <i>; undefined</i>
3999 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4004 <!-- _______________________________________________________________________ -->
4006 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4013 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4014 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4018 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4019 operand shifted to the right a specified number of bits with zero fill.</p>
4022 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4023 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4024 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4027 <p>This instruction always performs a logical shift right operation. The most
4028 significant bits of the result will be filled with zero bits after the shift.
4029 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4030 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4031 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4032 shift amount in <tt>op2</tt>.</p>
4034 <p>If the <tt>exact</tt> keyword is present, the result value of the
4035 <tt>lshr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
4036 shifted out are non-zero.</p>
4041 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4042 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4043 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4044 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4045 <result> = lshr i32 1, 32 <i>; undefined</i>
4046 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4051 <!-- _______________________________________________________________________ -->
4053 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4060 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4061 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4065 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4066 operand shifted to the right a specified number of bits with sign
4070 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4071 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4072 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4075 <p>This instruction always performs an arithmetic shift right operation, The
4076 most significant bits of the result will be filled with the sign bit
4077 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4078 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4079 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4080 the corresponding shift amount in <tt>op2</tt>.</p>
4082 <p>If the <tt>exact</tt> keyword is present, the result value of the
4083 <tt>ashr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
4084 shifted out are non-zero.</p>
4088 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4089 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4090 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4091 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4092 <result> = ashr i32 1, 32 <i>; undefined</i>
4093 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4098 <!-- _______________________________________________________________________ -->
4100 <a name="i_and">'<tt>and</tt>' Instruction</a>
4107 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4111 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4115 <p>The two arguments to the '<tt>and</tt>' instruction must be
4116 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4117 values. Both arguments must have identical types.</p>
4120 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4122 <table border="1" cellspacing="0" cellpadding="4">
4154 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4155 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4156 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4159 <!-- _______________________________________________________________________ -->
4161 <a name="i_or">'<tt>or</tt>' Instruction</a>
4168 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4172 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4176 <p>The two arguments to the '<tt>or</tt>' instruction must be
4177 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4178 values. Both arguments must have identical types.</p>
4181 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4183 <table border="1" cellspacing="0" cellpadding="4">
4215 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4216 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4217 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4222 <!-- _______________________________________________________________________ -->
4224 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4231 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4235 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4236 its two operands. The <tt>xor</tt> is used to implement the "one's
4237 complement" operation, which is the "~" operator in C.</p>
4240 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4241 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4242 values. Both arguments must have identical types.</p>
4245 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4247 <table border="1" cellspacing="0" cellpadding="4">
4279 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4280 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4281 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4282 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4289 <!-- ======================================================================= -->
4291 <a name="vectorops">Vector Operations</a>
4296 <p>LLVM supports several instructions to represent vector operations in a
4297 target-independent manner. These instructions cover the element-access and
4298 vector-specific operations needed to process vectors effectively. While LLVM
4299 does directly support these vector operations, many sophisticated algorithms
4300 will want to use target-specific intrinsics to take full advantage of a
4301 specific target.</p>
4303 <!-- _______________________________________________________________________ -->
4305 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4312 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4316 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4317 from a vector at a specified index.</p>
4321 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4322 of <a href="#t_vector">vector</a> type. The second operand is an index
4323 indicating the position from which to extract the element. The index may be
4327 <p>The result is a scalar of the same type as the element type of
4328 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4329 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4330 results are undefined.</p>
4334 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4339 <!-- _______________________________________________________________________ -->
4341 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4348 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4352 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4353 vector at a specified index.</p>
4356 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4357 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4358 whose type must equal the element type of the first operand. The third
4359 operand is an index indicating the position at which to insert the value.
4360 The index may be a variable.</p>
4363 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4364 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4365 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4366 results are undefined.</p>
4370 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4375 <!-- _______________________________________________________________________ -->
4377 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4384 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4388 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4389 from two input vectors, returning a vector with the same element type as the
4390 input and length that is the same as the shuffle mask.</p>
4393 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4394 with types that match each other. The third argument is a shuffle mask whose
4395 element type is always 'i32'. The result of the instruction is a vector
4396 whose length is the same as the shuffle mask and whose element type is the
4397 same as the element type of the first two operands.</p>
4399 <p>The shuffle mask operand is required to be a constant vector with either
4400 constant integer or undef values.</p>
4403 <p>The elements of the two input vectors are numbered from left to right across
4404 both of the vectors. The shuffle mask operand specifies, for each element of
4405 the result vector, which element of the two input vectors the result element
4406 gets. The element selector may be undef (meaning "don't care") and the
4407 second operand may be undef if performing a shuffle from only one vector.</p>
4411 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4412 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4413 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4414 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4415 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4416 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4417 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4418 <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>
4425 <!-- ======================================================================= -->
4427 <a name="aggregateops">Aggregate Operations</a>
4432 <p>LLVM supports several instructions for working with
4433 <a href="#t_aggregate">aggregate</a> values.</p>
4435 <!-- _______________________________________________________________________ -->
4437 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4444 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4448 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4449 from an <a href="#t_aggregate">aggregate</a> value.</p>
4452 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4453 of <a href="#t_struct">struct</a> or
4454 <a href="#t_array">array</a> type. The operands are constant indices to
4455 specify which value to extract in a similar manner as indices in a
4456 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4457 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4459 <li>Since the value being indexed is not a pointer, the first index is
4460 omitted and assumed to be zero.</li>
4461 <li>At least one index must be specified.</li>
4462 <li>Not only struct indices but also array indices must be in
4467 <p>The result is the value at the position in the aggregate specified by the
4472 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4477 <!-- _______________________________________________________________________ -->
4479 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4486 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4490 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4491 in an <a href="#t_aggregate">aggregate</a> value.</p>
4494 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4495 of <a href="#t_struct">struct</a> or
4496 <a href="#t_array">array</a> type. The second operand is a first-class
4497 value to insert. The following operands are constant indices indicating
4498 the position at which to insert the value in a similar manner as indices in a
4499 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4500 value to insert must have the same type as the value identified by the
4504 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4505 that of <tt>val</tt> except that the value at the position specified by the
4506 indices is that of <tt>elt</tt>.</p>
4510 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4511 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4512 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4519 <!-- ======================================================================= -->
4521 <a name="memoryops">Memory Access and Addressing Operations</a>
4526 <p>A key design point of an SSA-based representation is how it represents
4527 memory. In LLVM, no memory locations are in SSA form, which makes things
4528 very simple. This section describes how to read, write, and allocate
4531 <!-- _______________________________________________________________________ -->
4533 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4540 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4544 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4545 currently executing function, to be automatically released when this function
4546 returns to its caller. The object is always allocated in the generic address
4547 space (address space zero).</p>
4550 <p>The '<tt>alloca</tt>' instruction
4551 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4552 runtime stack, returning a pointer of the appropriate type to the program.
4553 If "NumElements" is specified, it is the number of elements allocated,
4554 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4555 specified, the value result of the allocation is guaranteed to be aligned to
4556 at least that boundary. If not specified, or if zero, the target can choose
4557 to align the allocation on any convenient boundary compatible with the
4560 <p>'<tt>type</tt>' may be any sized type.</p>
4563 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4564 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4565 memory is automatically released when the function returns. The
4566 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4567 variables that must have an address available. When the function returns
4568 (either with the <tt><a href="#i_ret">ret</a></tt>
4569 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
4570 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4574 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4575 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4576 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4577 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4582 <!-- _______________________________________________________________________ -->
4584 <a name="i_load">'<tt>load</tt>' Instruction</a>
4591 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4592 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4593 !<index> = !{ i32 1 }
4597 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4600 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4601 from which to load. The pointer must point to
4602 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4603 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4604 number or order of execution of this <tt>load</tt> with other <a
4605 href="#volatile">volatile operations</a>.</p>
4607 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
4608 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4609 argument. The <code>release</code> and <code>acq_rel</code> orderings are
4610 not valid on <code>load</code> instructions. Atomic loads produce <a
4611 href="#memorymodel">defined</a> results when they may see multiple atomic
4612 stores. The type of the pointee must be an integer type whose bit width
4613 is a power of two greater than or equal to eight and less than or equal
4614 to a target-specific size limit. <code>align</code> must be explicitly
4615 specified on atomic loads, and the load has undefined behavior if the
4616 alignment is not set to a value which is at least the size in bytes of
4617 the pointee. <code>!nontemporal</code> does not have any defined semantics
4618 for atomic loads.</p>
4620 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4621 operation (that is, the alignment of the memory address). A value of 0 or an
4622 omitted <tt>align</tt> argument means that the operation has the preferential
4623 alignment for the target. It is the responsibility of the code emitter to
4624 ensure that the alignment information is correct. Overestimating the
4625 alignment results in undefined behavior. Underestimating the alignment may
4626 produce less efficient code. An alignment of 1 is always safe.</p>
4628 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4629 metatadata name <index> corresponding to a metadata node with
4630 one <tt>i32</tt> entry of value 1. The existence of
4631 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4632 and code generator that this load is not expected to be reused in the cache.
4633 The code generator may select special instructions to save cache bandwidth,
4634 such as the <tt>MOVNT</tt> instruction on x86.</p>
4637 <p>The location of memory pointed to is loaded. If the value being loaded is of
4638 scalar type then the number of bytes read does not exceed the minimum number
4639 of bytes needed to hold all bits of the type. For example, loading an
4640 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4641 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4642 is undefined if the value was not originally written using a store of the
4647 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4648 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4649 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4654 <!-- _______________________________________________________________________ -->
4656 <a name="i_store">'<tt>store</tt>' Instruction</a>
4663 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4664 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
4668 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4671 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4672 and an address at which to store it. The type of the
4673 '<tt><pointer></tt>' operand must be a pointer to
4674 the <a href="#t_firstclass">first class</a> type of the
4675 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
4676 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
4677 order of execution of this <tt>store</tt> with other <a
4678 href="#volatile">volatile operations</a>.</p>
4680 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
4681 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4682 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
4683 valid on <code>store</code> instructions. Atomic loads produce <a
4684 href="#memorymodel">defined</a> results when they may see multiple atomic
4685 stores. The type of the pointee must be an integer type whose bit width
4686 is a power of two greater than or equal to eight and less than or equal
4687 to a target-specific size limit. <code>align</code> must be explicitly
4688 specified on atomic stores, and the store has undefined behavior if the
4689 alignment is not set to a value which is at least the size in bytes of
4690 the pointee. <code>!nontemporal</code> does not have any defined semantics
4691 for atomic stores.</p>
4693 <p>The optional constant "align" argument specifies the alignment of the
4694 operation (that is, the alignment of the memory address). A value of 0 or an
4695 omitted "align" argument means that the operation has the preferential
4696 alignment for the target. It is the responsibility of the code emitter to
4697 ensure that the alignment information is correct. Overestimating the
4698 alignment results in an undefined behavior. Underestimating the alignment may
4699 produce less efficient code. An alignment of 1 is always safe.</p>
4701 <p>The optional !nontemporal metadata must reference a single metatadata
4702 name <index> corresponding to a metadata node with one i32 entry of
4703 value 1. The existence of the !nontemporal metatadata on the
4704 instruction tells the optimizer and code generator that this load is
4705 not expected to be reused in the cache. The code generator may
4706 select special instructions to save cache bandwidth, such as the
4707 MOVNT instruction on x86.</p>
4711 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4712 location specified by the '<tt><pointer></tt>' operand. If
4713 '<tt><value></tt>' is of scalar type then the number of bytes written
4714 does not exceed the minimum number of bytes needed to hold all bits of the
4715 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4716 writing a value of a type like <tt>i20</tt> with a size that is not an
4717 integral number of bytes, it is unspecified what happens to the extra bits
4718 that do not belong to the type, but they will typically be overwritten.</p>
4722 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4723 store i32 3, i32* %ptr <i>; yields {void}</i>
4724 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4729 <!-- _______________________________________________________________________ -->
4731 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
4738 fence [singlethread] <ordering> <i>; yields {void}</i>
4742 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
4743 between operations.</p>
4745 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
4746 href="#ordering">ordering</a> argument which defines what
4747 <i>synchronizes-with</i> edges they add. They can only be given
4748 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
4749 <code>seq_cst</code> orderings.</p>
4752 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
4753 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
4754 <code>acquire</code> ordering semantics if and only if there exist atomic
4755 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
4756 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
4757 <var>X</var> modifies <var>M</var> (either directly or through some side effect
4758 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
4759 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
4760 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
4761 than an explicit <code>fence</code>, one (but not both) of the atomic operations
4762 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
4763 <code>acquire</code> (resp.) ordering constraint and still
4764 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
4765 <i>happens-before</i> edge.</p>
4767 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
4768 having both <code>acquire</code> and <code>release</code> semantics specified
4769 above, participates in the global program order of other <code>seq_cst</code>
4770 operations and/or fences.</p>
4772 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
4773 specifies that the fence only synchronizes with other fences in the same
4774 thread. (This is useful for interacting with signal handlers.)</p>
4778 fence acquire <i>; yields {void}</i>
4779 fence singlethread seq_cst <i>; yields {void}</i>
4784 <!-- _______________________________________________________________________ -->
4786 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
4793 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
4797 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
4798 It loads a value in memory and compares it to a given value. If they are
4799 equal, it stores a new value into the memory.</p>
4802 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
4803 address to operate on, a value to compare to the value currently be at that
4804 address, and a new value to place at that address if the compared values are
4805 equal. The type of '<var><cmp></var>' must be an integer type whose
4806 bit width is a power of two greater than or equal to eight and less than
4807 or equal to a target-specific size limit. '<var><cmp></var>' and
4808 '<var><new></var>' must have the same type, and the type of
4809 '<var><pointer></var>' must be a pointer to that type. If the
4810 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
4811 optimizer is not allowed to modify the number or order of execution
4812 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
4815 <!-- FIXME: Extend allowed types. -->
4817 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
4818 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
4820 <p>The optional "<code>singlethread</code>" argument declares that the
4821 <code>cmpxchg</code> is only atomic with respect to code (usually signal
4822 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
4823 cmpxchg is atomic with respect to all other code in the system.</p>
4825 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
4826 the size in memory of the operand.
4829 <p>The contents of memory at the location specified by the
4830 '<tt><pointer></tt>' operand is read and compared to
4831 '<tt><cmp></tt>'; if the read value is the equal,
4832 '<tt><new></tt>' is written. The original value at the location
4835 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
4836 purpose of identifying <a href="#release_sequence">release sequences</a>. A
4837 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
4838 parameter determined by dropping any <code>release</code> part of the
4839 <code>cmpxchg</code>'s ordering.</p>
4842 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
4843 optimization work on ARM.)
4845 FIXME: Is a weaker ordering constraint on failure helpful in practice?
4851 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
4852 <a href="#i_br">br</a> label %loop
4855 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
4856 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
4857 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
4858 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
4859 <a href="#i_br">br</a> i1 %success, label %done, label %loop
4867 <!-- _______________________________________________________________________ -->
4869 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
4876 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
4880 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
4883 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
4884 operation to apply, an address whose value to modify, an argument to the
4885 operation. The operation must be one of the following keywords:</p>
4900 <p>The type of '<var><value></var>' must be an integer type whose
4901 bit width is a power of two greater than or equal to eight and less than
4902 or equal to a target-specific size limit. The type of the
4903 '<code><pointer></code>' operand must be a pointer to that type.
4904 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
4905 optimizer is not allowed to modify the number or order of execution of this
4906 <code>atomicrmw</code> with other <a href="#volatile">volatile
4909 <!-- FIXME: Extend allowed types. -->
4912 <p>The contents of memory at the location specified by the
4913 '<tt><pointer></tt>' operand are atomically read, modified, and written
4914 back. The original value at the location is returned. The modification is
4915 specified by the <var>operation</var> argument:</p>
4918 <li>xchg: <code>*ptr = val</code></li>
4919 <li>add: <code>*ptr = *ptr + val</code></li>
4920 <li>sub: <code>*ptr = *ptr - val</code></li>
4921 <li>and: <code>*ptr = *ptr & val</code></li>
4922 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
4923 <li>or: <code>*ptr = *ptr | val</code></li>
4924 <li>xor: <code>*ptr = *ptr ^ val</code></li>
4925 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
4926 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
4927 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
4928 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
4933 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
4938 <!-- _______________________________________________________________________ -->
4940 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
4947 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4948 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4952 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4953 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
4954 It performs address calculation only and does not access memory.</p>
4957 <p>The first argument is always a pointer, and forms the basis of the
4958 calculation. The remaining arguments are indices that indicate which of the
4959 elements of the aggregate object are indexed. The interpretation of each
4960 index is dependent on the type being indexed into. The first index always
4961 indexes the pointer value given as the first argument, the second index
4962 indexes a value of the type pointed to (not necessarily the value directly
4963 pointed to, since the first index can be non-zero), etc. The first type
4964 indexed into must be a pointer value, subsequent types can be arrays,
4965 vectors, and structs. Note that subsequent types being indexed into
4966 can never be pointers, since that would require loading the pointer before
4967 continuing calculation.</p>
4969 <p>The type of each index argument depends on the type it is indexing into.
4970 When indexing into a (optionally packed) structure, only <tt>i32</tt>
4971 integer <b>constants</b> are allowed. When indexing into an array, pointer
4972 or vector, integers of any width are allowed, and they are not required to be
4973 constant. These integers are treated as signed values where relevant.</p>
4975 <p>For example, let's consider a C code fragment and how it gets compiled to
4978 <pre class="doc_code">
4990 int *foo(struct ST *s) {
4991 return &s[1].Z.B[5][13];
4995 <p>The LLVM code generated by the GCC frontend is:</p>
4997 <pre class="doc_code">
4998 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4999 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
5001 define i32* @foo(%ST* %s) {
5003 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
5009 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
5010 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
5011 }</tt>' type, a structure. The second index indexes into the third element
5012 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
5013 i8 }</tt>' type, another structure. The third index indexes into the second
5014 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
5015 array. The two dimensions of the array are subscripted into, yielding an
5016 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
5017 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
5019 <p>Note that it is perfectly legal to index partially through a structure,
5020 returning a pointer to an inner element. Because of this, the LLVM code for
5021 the given testcase is equivalent to:</p>
5024 define i32* @foo(%ST* %s) {
5025 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
5026 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
5027 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5028 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5029 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5034 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5035 <tt>getelementptr</tt> is a <a href="#trapvalues">trap value</a> if the
5036 base pointer is not an <i>in bounds</i> address of an allocated object,
5037 or if any of the addresses that would be formed by successive addition of
5038 the offsets implied by the indices to the base address with infinitely
5039 precise signed arithmetic are not an <i>in bounds</i> address of that
5040 allocated object. The <i>in bounds</i> addresses for an allocated object
5041 are all the addresses that point into the object, plus the address one
5042 byte past the end.</p>
5044 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5045 the base address with silently-wrapping two's complement arithmetic. If the
5046 offsets have a different width from the pointer, they are sign-extended or
5047 truncated to the width of the pointer. The result value of the
5048 <tt>getelementptr</tt> may be outside the object pointed to by the base
5049 pointer. The result value may not necessarily be used to access memory
5050 though, even if it happens to point into allocated storage. See the
5051 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5054 <p>The getelementptr instruction is often confusing. For some more insight into
5055 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5059 <i>; yields [12 x i8]*:aptr</i>
5060 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5061 <i>; yields i8*:vptr</i>
5062 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5063 <i>; yields i8*:eptr</i>
5064 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5065 <i>; yields i32*:iptr</i>
5066 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5073 <!-- ======================================================================= -->
5075 <a name="convertops">Conversion Operations</a>
5080 <p>The instructions in this category are the conversion instructions (casting)
5081 which all take a single operand and a type. They perform various bit
5082 conversions on the operand.</p>
5084 <!-- _______________________________________________________________________ -->
5086 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5093 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5097 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5098 type <tt>ty2</tt>.</p>
5101 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5102 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5103 of the same number of integers.
5104 The bit size of the <tt>value</tt> must be larger than
5105 the bit size of the destination type, <tt>ty2</tt>.
5106 Equal sized types are not allowed.</p>
5109 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5110 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5111 source size must be larger than the destination size, <tt>trunc</tt> cannot
5112 be a <i>no-op cast</i>. It will always truncate bits.</p>
5116 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5117 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5118 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5119 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5124 <!-- _______________________________________________________________________ -->
5126 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5133 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5137 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5142 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5143 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5144 of the same number of integers.
5145 The bit size of the <tt>value</tt> must be smaller than
5146 the bit size of the destination type,
5150 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5151 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5153 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5157 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5158 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5159 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5164 <!-- _______________________________________________________________________ -->
5166 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5173 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5177 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5180 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5181 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5182 of the same number of integers.
5183 The bit size of the <tt>value</tt> must be smaller than
5184 the bit size of the destination type,
5188 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5189 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5190 of the type <tt>ty2</tt>.</p>
5192 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5196 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5197 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5198 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5203 <!-- _______________________________________________________________________ -->
5205 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5212 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5216 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5220 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5221 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5222 to cast it to. The size of <tt>value</tt> must be larger than the size of
5223 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5224 <i>no-op cast</i>.</p>
5227 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5228 <a href="#t_floating">floating point</a> type to a smaller
5229 <a href="#t_floating">floating point</a> type. If the value cannot fit
5230 within the destination type, <tt>ty2</tt>, then the results are
5235 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5236 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5241 <!-- _______________________________________________________________________ -->
5243 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5250 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5254 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5255 floating point value.</p>
5258 <p>The '<tt>fpext</tt>' instruction takes a
5259 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5260 a <a href="#t_floating">floating point</a> type to cast it to. The source
5261 type must be smaller than the destination type.</p>
5264 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5265 <a href="#t_floating">floating point</a> type to a larger
5266 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5267 used to make a <i>no-op cast</i> because it always changes bits. Use
5268 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5272 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5273 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5278 <!-- _______________________________________________________________________ -->
5280 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5287 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5291 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5292 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5295 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5296 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5297 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5298 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5299 vector integer type with the same number of elements as <tt>ty</tt></p>
5302 <p>The '<tt>fptoui</tt>' instruction converts its
5303 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5304 towards zero) unsigned integer value. If the value cannot fit
5305 in <tt>ty2</tt>, the results are undefined.</p>
5309 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5310 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5311 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5316 <!-- _______________________________________________________________________ -->
5318 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5325 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5329 <p>The '<tt>fptosi</tt>' instruction converts
5330 <a href="#t_floating">floating point</a> <tt>value</tt> to
5331 type <tt>ty2</tt>.</p>
5334 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5335 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5336 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5337 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5338 vector integer type with the same number of elements as <tt>ty</tt></p>
5341 <p>The '<tt>fptosi</tt>' instruction converts its
5342 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5343 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5344 the results are undefined.</p>
5348 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5349 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5350 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5355 <!-- _______________________________________________________________________ -->
5357 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5364 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5368 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5369 integer and converts that value to the <tt>ty2</tt> type.</p>
5372 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5373 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5374 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5375 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5376 floating point type with the same number of elements as <tt>ty</tt></p>
5379 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5380 integer quantity and converts it to the corresponding floating point
5381 value. If the value cannot fit in the floating point value, the results are
5386 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5387 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5392 <!-- _______________________________________________________________________ -->
5394 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5401 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5405 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5406 and converts that value to the <tt>ty2</tt> type.</p>
5409 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5410 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5411 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5412 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5413 floating point type with the same number of elements as <tt>ty</tt></p>
5416 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5417 quantity and converts it to the corresponding floating point value. If the
5418 value cannot fit in the floating point value, the results are undefined.</p>
5422 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5423 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5428 <!-- _______________________________________________________________________ -->
5430 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5437 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5441 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
5442 the integer type <tt>ty2</tt>.</p>
5445 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5446 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
5447 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
5450 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5451 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5452 truncating or zero extending that value to the size of the integer type. If
5453 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5454 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5455 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5460 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
5461 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
5466 <!-- _______________________________________________________________________ -->
5468 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5475 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5479 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5480 pointer type, <tt>ty2</tt>.</p>
5483 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5484 value to cast, and a type to cast it to, which must be a
5485 <a href="#t_pointer">pointer</a> type.</p>
5488 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5489 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5490 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5491 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5492 than the size of a pointer then a zero extension is done. If they are the
5493 same size, nothing is done (<i>no-op cast</i>).</p>
5497 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5498 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5499 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5504 <!-- _______________________________________________________________________ -->
5506 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5513 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5517 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5518 <tt>ty2</tt> without changing any bits.</p>
5521 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5522 non-aggregate first class value, and a type to cast it to, which must also be
5523 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5524 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5525 identical. If the source type is a pointer, the destination type must also be
5526 a pointer. This instruction supports bitwise conversion of vectors to
5527 integers and to vectors of other types (as long as they have the same
5531 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5532 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5533 this conversion. The conversion is done as if the <tt>value</tt> had been
5534 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
5535 be converted to other pointer types with this instruction. To convert
5536 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5537 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5541 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5542 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5543 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5550 <!-- ======================================================================= -->
5552 <a name="otherops">Other Operations</a>
5557 <p>The instructions in this category are the "miscellaneous" instructions, which
5558 defy better classification.</p>
5560 <!-- _______________________________________________________________________ -->
5562 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5569 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5573 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5574 boolean values based on comparison of its two integer, integer vector, or
5575 pointer operands.</p>
5578 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5579 the condition code indicating the kind of comparison to perform. It is not a
5580 value, just a keyword. The possible condition code are:</p>
5583 <li><tt>eq</tt>: equal</li>
5584 <li><tt>ne</tt>: not equal </li>
5585 <li><tt>ugt</tt>: unsigned greater than</li>
5586 <li><tt>uge</tt>: unsigned greater or equal</li>
5587 <li><tt>ult</tt>: unsigned less than</li>
5588 <li><tt>ule</tt>: unsigned less or equal</li>
5589 <li><tt>sgt</tt>: signed greater than</li>
5590 <li><tt>sge</tt>: signed greater or equal</li>
5591 <li><tt>slt</tt>: signed less than</li>
5592 <li><tt>sle</tt>: signed less or equal</li>
5595 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
5596 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5597 typed. They must also be identical types.</p>
5600 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5601 condition code given as <tt>cond</tt>. The comparison performed always yields
5602 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
5603 result, as follows:</p>
5606 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
5607 <tt>false</tt> otherwise. No sign interpretation is necessary or
5610 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
5611 <tt>false</tt> otherwise. No sign interpretation is necessary or
5614 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
5615 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5617 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
5618 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5619 to <tt>op2</tt>.</li>
5621 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
5622 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5624 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
5625 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5627 <li><tt>sgt</tt>: interprets the operands as signed values and yields
5628 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5630 <li><tt>sge</tt>: interprets the operands as signed values and yields
5631 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5632 to <tt>op2</tt>.</li>
5634 <li><tt>slt</tt>: interprets the operands as signed values and yields
5635 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5637 <li><tt>sle</tt>: interprets the operands as signed values and yields
5638 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5641 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
5642 values are compared as if they were integers.</p>
5644 <p>If the operands are integer vectors, then they are compared element by
5645 element. The result is an <tt>i1</tt> vector with the same number of elements
5646 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
5650 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
5651 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
5652 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
5653 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
5654 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
5655 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
5658 <p>Note that the code generator does not yet support vector types with
5659 the <tt>icmp</tt> instruction.</p>
5663 <!-- _______________________________________________________________________ -->
5665 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
5672 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5676 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
5677 values based on comparison of its operands.</p>
5679 <p>If the operands are floating point scalars, then the result type is a boolean
5680 (<a href="#t_integer"><tt>i1</tt></a>).</p>
5682 <p>If the operands are floating point vectors, then the result type is a vector
5683 of boolean with the same number of elements as the operands being
5687 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
5688 the condition code indicating the kind of comparison to perform. It is not a
5689 value, just a keyword. The possible condition code are:</p>
5692 <li><tt>false</tt>: no comparison, always returns false</li>
5693 <li><tt>oeq</tt>: ordered and equal</li>
5694 <li><tt>ogt</tt>: ordered and greater than </li>
5695 <li><tt>oge</tt>: ordered and greater than or equal</li>
5696 <li><tt>olt</tt>: ordered and less than </li>
5697 <li><tt>ole</tt>: ordered and less than or equal</li>
5698 <li><tt>one</tt>: ordered and not equal</li>
5699 <li><tt>ord</tt>: ordered (no nans)</li>
5700 <li><tt>ueq</tt>: unordered or equal</li>
5701 <li><tt>ugt</tt>: unordered or greater than </li>
5702 <li><tt>uge</tt>: unordered or greater than or equal</li>
5703 <li><tt>ult</tt>: unordered or less than </li>
5704 <li><tt>ule</tt>: unordered or less than or equal</li>
5705 <li><tt>une</tt>: unordered or not equal</li>
5706 <li><tt>uno</tt>: unordered (either nans)</li>
5707 <li><tt>true</tt>: no comparison, always returns true</li>
5710 <p><i>Ordered</i> means that neither operand is a QNAN while
5711 <i>unordered</i> means that either operand may be a QNAN.</p>
5713 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
5714 a <a href="#t_floating">floating point</a> type or
5715 a <a href="#t_vector">vector</a> of floating point type. They must have
5716 identical types.</p>
5719 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
5720 according to the condition code given as <tt>cond</tt>. If the operands are
5721 vectors, then the vectors are compared element by element. Each comparison
5722 performed always yields an <a href="#t_integer">i1</a> result, as
5726 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
5728 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5729 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5731 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5732 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5734 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5735 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5737 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5738 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5740 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5741 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5743 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5744 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5746 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
5748 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
5749 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5751 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
5752 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5754 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
5755 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5757 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
5758 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5760 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
5761 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5763 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
5764 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5766 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
5768 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5773 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
5774 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5775 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5776 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
5779 <p>Note that the code generator does not yet support vector types with
5780 the <tt>fcmp</tt> instruction.</p>
5784 <!-- _______________________________________________________________________ -->
5786 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
5793 <result> = phi <ty> [ <val0>, <label0>], ...
5797 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
5798 SSA graph representing the function.</p>
5801 <p>The type of the incoming values is specified with the first type field. After
5802 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5803 one pair for each predecessor basic block of the current block. Only values
5804 of <a href="#t_firstclass">first class</a> type may be used as the value
5805 arguments to the PHI node. Only labels may be used as the label
5808 <p>There must be no non-phi instructions between the start of a basic block and
5809 the PHI instructions: i.e. PHI instructions must be first in a basic
5812 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5813 occur on the edge from the corresponding predecessor block to the current
5814 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5815 value on the same edge).</p>
5818 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
5819 specified by the pair corresponding to the predecessor basic block that
5820 executed just prior to the current block.</p>
5824 Loop: ; Infinite loop that counts from 0 on up...
5825 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5826 %nextindvar = add i32 %indvar, 1
5832 <!-- _______________________________________________________________________ -->
5834 <a name="i_select">'<tt>select</tt>' Instruction</a>
5841 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5843 <i>selty</i> is either i1 or {<N x i1>}
5847 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
5848 condition, without branching.</p>
5852 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5853 values indicating the condition, and two values of the
5854 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
5855 vectors and the condition is a scalar, then entire vectors are selected, not
5856 individual elements.</p>
5859 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
5860 first value argument; otherwise, it returns the second value argument.</p>
5862 <p>If the condition is a vector of i1, then the value arguments must be vectors
5863 of the same size, and the selection is done element by element.</p>
5867 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
5870 <p>Note that the code generator does not yet support conditions
5871 with vector type.</p>
5875 <!-- _______________________________________________________________________ -->
5877 <a name="i_call">'<tt>call</tt>' Instruction</a>
5884 <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>]
5888 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
5891 <p>This instruction requires several arguments:</p>
5894 <li>The optional "tail" marker indicates that the callee function does not
5895 access any allocas or varargs in the caller. Note that calls may be
5896 marked "tail" even if they do not occur before
5897 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
5898 present, the function call is eligible for tail call optimization,
5899 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
5900 optimized into a jump</a>. The code generator may optimize calls marked
5901 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
5902 sibling call optimization</a> when the caller and callee have
5903 matching signatures, or 2) forced tail call optimization when the
5904 following extra requirements are met:
5906 <li>Caller and callee both have the calling
5907 convention <tt>fastcc</tt>.</li>
5908 <li>The call is in tail position (ret immediately follows call and ret
5909 uses value of call or is void).</li>
5910 <li>Option <tt>-tailcallopt</tt> is enabled,
5911 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
5912 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
5913 constraints are met.</a></li>
5917 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5918 convention</a> the call should use. If none is specified, the call
5919 defaults to using C calling conventions. The calling convention of the
5920 call must match the calling convention of the target function, or else the
5921 behavior is undefined.</li>
5923 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5924 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5925 '<tt>inreg</tt>' attributes are valid here.</li>
5927 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5928 type of the return value. Functions that return no value are marked
5929 <tt><a href="#t_void">void</a></tt>.</li>
5931 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5932 being invoked. The argument types must match the types implied by this
5933 signature. This type can be omitted if the function is not varargs and if
5934 the function type does not return a pointer to a function.</li>
5936 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5937 be invoked. In most cases, this is a direct function invocation, but
5938 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5939 to function value.</li>
5941 <li>'<tt>function args</tt>': argument list whose types match the function
5942 signature argument types and parameter attributes. All arguments must be
5943 of <a href="#t_firstclass">first class</a> type. If the function
5944 signature indicates the function accepts a variable number of arguments,
5945 the extra arguments can be specified.</li>
5947 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5948 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5949 '<tt>readnone</tt>' attributes are valid here.</li>
5953 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5954 a specified function, with its incoming arguments bound to the specified
5955 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5956 function, control flow continues with the instruction after the function
5957 call, and the return value of the function is bound to the result
5962 %retval = call i32 @test(i32 %argc)
5963 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
5964 %X = tail call i32 @foo() <i>; yields i32</i>
5965 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5966 call void %foo(i8 97 signext)
5968 %struct.A = type { i32, i8 }
5969 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
5970 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5971 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
5972 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
5973 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
5976 <p>llvm treats calls to some functions with names and arguments that match the
5977 standard C99 library as being the C99 library functions, and may perform
5978 optimizations or generate code for them under that assumption. This is
5979 something we'd like to change in the future to provide better support for
5980 freestanding environments and non-C-based languages.</p>
5984 <!-- _______________________________________________________________________ -->
5986 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5993 <resultval> = va_arg <va_list*> <arglist>, <argty>
5997 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
5998 the "variable argument" area of a function call. It is used to implement the
5999 <tt>va_arg</tt> macro in C.</p>
6002 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6003 argument. It returns a value of the specified argument type and increments
6004 the <tt>va_list</tt> to point to the next argument. The actual type
6005 of <tt>va_list</tt> is target specific.</p>
6008 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6009 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6010 to the next argument. For more information, see the variable argument
6011 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6013 <p>It is legal for this instruction to be called in a function which does not
6014 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6017 <p><tt>va_arg</tt> is an LLVM instruction instead of
6018 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6022 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6024 <p>Note that the code generator does not yet fully support va_arg on many
6025 targets. Also, it does not currently support va_arg with aggregate types on
6030 <!-- _______________________________________________________________________ -->
6032 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6039 <resultval> = landingpad <somety> personality <type> <pers_fn> <clause>+
6040 <resultval> = landingpad <somety> personality <type> <pers_fn> cleanup <clause>*
6042 <clause> := catch <type> <value>
6043 <clause> := filter <array constant type> <array constant>
6047 <p>The '<tt>landingpad</tt>' instruction is used by
6048 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6049 system</a> to specify that a basic block is a landing pad — one where
6050 the exception lands, and corresponds to the code found in the
6051 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6052 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6053 re-entry to the function. The <tt>resultval</tt> has the
6054 type <tt>somety</tt>.</p>
6057 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6058 function associated with the unwinding mechanism. The optional
6059 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6061 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6062 or <tt>filter</tt> — and contains the global variable representing the
6063 "type" that may be caught or filtered respectively. Unlike the
6064 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6065 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6066 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6067 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6070 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6071 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6072 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6073 calling conventions, how the personality function results are represented in
6074 LLVM IR is target specific.</p>
6076 <p>The clauses are applied in order from top to bottom. If two
6077 <tt>landingpad</tt> instructions are merged together through inlining, the
6078 clauses from the calling function are appended to the list of clauses.</p>
6080 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6083 <li>A landing pad block is a basic block which is the unwind destination of an
6084 '<tt>invoke</tt>' instruction.</li>
6085 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6086 first non-PHI instruction.</li>
6087 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6089 <li>A basic block that is not a landing pad block may not include a
6090 '<tt>landingpad</tt>' instruction.</li>
6091 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6092 personality function.</li>
6097 ;; A landing pad which can catch an integer.
6098 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6100 ;; A landing pad that is a cleanup.
6101 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6103 ;; A landing pad which can catch an integer and can only throw a double.
6104 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6106 filter [1 x i8**] [@_ZTId]
6115 <!-- *********************************************************************** -->
6116 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6117 <!-- *********************************************************************** -->
6121 <p>LLVM supports the notion of an "intrinsic function". These functions have
6122 well known names and semantics and are required to follow certain
6123 restrictions. Overall, these intrinsics represent an extension mechanism for
6124 the LLVM language that does not require changing all of the transformations
6125 in LLVM when adding to the language (or the bitcode reader/writer, the
6126 parser, etc...).</p>
6128 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6129 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6130 begin with this prefix. Intrinsic functions must always be external
6131 functions: you cannot define the body of intrinsic functions. Intrinsic
6132 functions may only be used in call or invoke instructions: it is illegal to
6133 take the address of an intrinsic function. Additionally, because intrinsic
6134 functions are part of the LLVM language, it is required if any are added that
6135 they be documented here.</p>
6137 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6138 family of functions that perform the same operation but on different data
6139 types. Because LLVM can represent over 8 million different integer types,
6140 overloading is used commonly to allow an intrinsic function to operate on any
6141 integer type. One or more of the argument types or the result type can be
6142 overloaded to accept any integer type. Argument types may also be defined as
6143 exactly matching a previous argument's type or the result type. This allows
6144 an intrinsic function which accepts multiple arguments, but needs all of them
6145 to be of the same type, to only be overloaded with respect to a single
6146 argument or the result.</p>
6148 <p>Overloaded intrinsics will have the names of its overloaded argument types
6149 encoded into its function name, each preceded by a period. Only those types
6150 which are overloaded result in a name suffix. Arguments whose type is matched
6151 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6152 can take an integer of any width and returns an integer of exactly the same
6153 integer width. This leads to a family of functions such as
6154 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6155 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6156 suffix is required. Because the argument's type is matched against the return
6157 type, it does not require its own name suffix.</p>
6159 <p>To learn how to add an intrinsic function, please see the
6160 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6162 <!-- ======================================================================= -->
6164 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6169 <p>Variable argument support is defined in LLVM with
6170 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6171 intrinsic functions. These functions are related to the similarly named
6172 macros defined in the <tt><stdarg.h></tt> header file.</p>
6174 <p>All of these functions operate on arguments that use a target-specific value
6175 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6176 not define what this type is, so all transformations should be prepared to
6177 handle these functions regardless of the type used.</p>
6179 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6180 instruction and the variable argument handling intrinsic functions are
6183 <pre class="doc_code">
6184 define i32 @test(i32 %X, ...) {
6185 ; Initialize variable argument processing
6187 %ap2 = bitcast i8** %ap to i8*
6188 call void @llvm.va_start(i8* %ap2)
6190 ; Read a single integer argument
6191 %tmp = va_arg i8** %ap, i32
6193 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6195 %aq2 = bitcast i8** %aq to i8*
6196 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6197 call void @llvm.va_end(i8* %aq2)
6199 ; Stop processing of arguments.
6200 call void @llvm.va_end(i8* %ap2)
6204 declare void @llvm.va_start(i8*)
6205 declare void @llvm.va_copy(i8*, i8*)
6206 declare void @llvm.va_end(i8*)
6209 <!-- _______________________________________________________________________ -->
6211 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6219 declare void %llvm.va_start(i8* <arglist>)
6223 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6224 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6227 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6230 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6231 macro available in C. In a target-dependent way, it initializes
6232 the <tt>va_list</tt> element to which the argument points, so that the next
6233 call to <tt>va_arg</tt> will produce the first variable argument passed to
6234 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6235 need to know the last argument of the function as the compiler can figure
6240 <!-- _______________________________________________________________________ -->
6242 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6249 declare void @llvm.va_end(i8* <arglist>)
6253 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6254 which has been initialized previously
6255 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6256 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6259 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6262 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6263 macro available in C. In a target-dependent way, it destroys
6264 the <tt>va_list</tt> element to which the argument points. Calls
6265 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6266 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6267 with calls to <tt>llvm.va_end</tt>.</p>
6271 <!-- _______________________________________________________________________ -->
6273 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6280 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6284 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6285 from the source argument list to the destination argument list.</p>
6288 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6289 The second argument is a pointer to a <tt>va_list</tt> element to copy
6293 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6294 macro available in C. In a target-dependent way, it copies the
6295 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6296 element. This intrinsic is necessary because
6297 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6298 arbitrarily complex and require, for example, memory allocation.</p>
6306 <!-- ======================================================================= -->
6308 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6313 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6314 Collection</a> (GC) requires the implementation and generation of these
6315 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6316 roots on the stack</a>, as well as garbage collector implementations that
6317 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6318 barriers. Front-ends for type-safe garbage collected languages should generate
6319 these intrinsics to make use of the LLVM garbage collectors. For more details,
6320 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6323 <p>The garbage collection intrinsics only operate on objects in the generic
6324 address space (address space zero).</p>
6326 <!-- _______________________________________________________________________ -->
6328 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6335 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6339 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6340 the code generator, and allows some metadata to be associated with it.</p>
6343 <p>The first argument specifies the address of a stack object that contains the
6344 root pointer. The second pointer (which must be either a constant or a
6345 global value address) contains the meta-data to be associated with the
6349 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6350 location. At compile-time, the code generator generates information to allow
6351 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6352 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6357 <!-- _______________________________________________________________________ -->
6359 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6366 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6370 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6371 locations, allowing garbage collector implementations that require read
6375 <p>The second argument is the address to read from, which should be an address
6376 allocated from the garbage collector. The first object is a pointer to the
6377 start of the referenced object, if needed by the language runtime (otherwise
6381 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6382 instruction, but may be replaced with substantially more complex code by the
6383 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6384 may only be used in a function which <a href="#gc">specifies a GC
6389 <!-- _______________________________________________________________________ -->
6391 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6398 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6402 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6403 locations, allowing garbage collector implementations that require write
6404 barriers (such as generational or reference counting collectors).</p>
6407 <p>The first argument is the reference to store, the second is the start of the
6408 object to store it to, and the third is the address of the field of Obj to
6409 store to. If the runtime does not require a pointer to the object, Obj may
6413 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6414 instruction, but may be replaced with substantially more complex code by the
6415 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6416 may only be used in a function which <a href="#gc">specifies a GC
6423 <!-- ======================================================================= -->
6425 <a name="int_codegen">Code Generator Intrinsics</a>
6430 <p>These intrinsics are provided by LLVM to expose special features that may
6431 only be implemented with code generator support.</p>
6433 <!-- _______________________________________________________________________ -->
6435 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6442 declare i8 *@llvm.returnaddress(i32 <level>)
6446 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6447 target-specific value indicating the return address of the current function
6448 or one of its callers.</p>
6451 <p>The argument to this intrinsic indicates which function to return the address
6452 for. Zero indicates the calling function, one indicates its caller, etc.
6453 The argument is <b>required</b> to be a constant integer value.</p>
6456 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6457 indicating the return address of the specified call frame, or zero if it
6458 cannot be identified. The value returned by this intrinsic is likely to be
6459 incorrect or 0 for arguments other than zero, so it should only be used for
6460 debugging purposes.</p>
6462 <p>Note that calling this intrinsic does not prevent function inlining or other
6463 aggressive transformations, so the value returned may not be that of the
6464 obvious source-language caller.</p>
6468 <!-- _______________________________________________________________________ -->
6470 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6477 declare i8* @llvm.frameaddress(i32 <level>)
6481 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6482 target-specific frame pointer value for the specified stack frame.</p>
6485 <p>The argument to this intrinsic indicates which function to return the frame
6486 pointer for. Zero indicates the calling function, one indicates its caller,
6487 etc. The argument is <b>required</b> to be a constant integer value.</p>
6490 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6491 indicating the frame address of the specified call frame, or zero if it
6492 cannot be identified. The value returned by this intrinsic is likely to be
6493 incorrect or 0 for arguments other than zero, so it should only be used for
6494 debugging purposes.</p>
6496 <p>Note that calling this intrinsic does not prevent function inlining or other
6497 aggressive transformations, so the value returned may not be that of the
6498 obvious source-language caller.</p>
6502 <!-- _______________________________________________________________________ -->
6504 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6511 declare i8* @llvm.stacksave()
6515 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6516 of the function stack, for use
6517 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6518 useful for implementing language features like scoped automatic variable
6519 sized arrays in C99.</p>
6522 <p>This intrinsic returns a opaque pointer value that can be passed
6523 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6524 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6525 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6526 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6527 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6528 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6532 <!-- _______________________________________________________________________ -->
6534 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6541 declare void @llvm.stackrestore(i8* %ptr)
6545 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6546 the function stack to the state it was in when the
6547 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6548 executed. This is useful for implementing language features like scoped
6549 automatic variable sized arrays in C99.</p>
6552 <p>See the description
6553 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6557 <!-- _______________________________________________________________________ -->
6559 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6566 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6570 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6571 insert a prefetch instruction if supported; otherwise, it is a noop.
6572 Prefetches have no effect on the behavior of the program but can change its
6573 performance characteristics.</p>
6576 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6577 specifier determining if the fetch should be for a read (0) or write (1),
6578 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6579 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6580 specifies whether the prefetch is performed on the data (1) or instruction (0)
6581 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6582 must be constant integers.</p>
6585 <p>This intrinsic does not modify the behavior of the program. In particular,
6586 prefetches cannot trap and do not produce a value. On targets that support
6587 this intrinsic, the prefetch can provide hints to the processor cache for
6588 better performance.</p>
6592 <!-- _______________________________________________________________________ -->
6594 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
6601 declare void @llvm.pcmarker(i32 <id>)
6605 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
6606 Counter (PC) in a region of code to simulators and other tools. The method
6607 is target specific, but it is expected that the marker will use exported
6608 symbols to transmit the PC of the marker. The marker makes no guarantees
6609 that it will remain with any specific instruction after optimizations. It is
6610 possible that the presence of a marker will inhibit optimizations. The
6611 intended use is to be inserted after optimizations to allow correlations of
6612 simulation runs.</p>
6615 <p><tt>id</tt> is a numerical id identifying the marker.</p>
6618 <p>This intrinsic does not modify the behavior of the program. Backends that do
6619 not support this intrinsic may ignore it.</p>
6623 <!-- _______________________________________________________________________ -->
6625 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
6632 declare i64 @llvm.readcyclecounter()
6636 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
6637 counter register (or similar low latency, high accuracy clocks) on those
6638 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6639 should map to RPCC. As the backing counters overflow quickly (on the order
6640 of 9 seconds on alpha), this should only be used for small timings.</p>
6643 <p>When directly supported, reading the cycle counter should not modify any
6644 memory. Implementations are allowed to either return a application specific
6645 value or a system wide value. On backends without support, this is lowered
6646 to a constant 0.</p>
6652 <!-- ======================================================================= -->
6654 <a name="int_libc">Standard C Library Intrinsics</a>
6659 <p>LLVM provides intrinsics for a few important standard C library functions.
6660 These intrinsics allow source-language front-ends to pass information about
6661 the alignment of the pointer arguments to the code generator, providing
6662 opportunity for more efficient code generation.</p>
6664 <!-- _______________________________________________________________________ -->
6666 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
6672 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
6673 integer bit width and for different address spaces. Not all targets support
6674 all bit widths however.</p>
6677 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6678 i32 <len>, i32 <align>, i1 <isvolatile>)
6679 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6680 i64 <len>, i32 <align>, i1 <isvolatile>)
6684 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6685 source location to the destination location.</p>
6687 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
6688 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6689 and the pointers can be in specified address spaces.</p>
6693 <p>The first argument is a pointer to the destination, the second is a pointer
6694 to the source. The third argument is an integer argument specifying the
6695 number of bytes to copy, the fourth argument is the alignment of the
6696 source and destination locations, and the fifth is a boolean indicating a
6697 volatile access.</p>
6699 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6700 then the caller guarantees that both the source and destination pointers are
6701 aligned to that boundary.</p>
6703 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6704 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
6705 The detailed access behavior is not very cleanly specified and it is unwise
6706 to depend on it.</p>
6710 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6711 source location to the destination location, which are not allowed to
6712 overlap. It copies "len" bytes of memory over. If the argument is known to
6713 be aligned to some boundary, this can be specified as the fourth argument,
6714 otherwise it should be set to 0 or 1.</p>
6718 <!-- _______________________________________________________________________ -->
6720 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
6726 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6727 width and for different address space. Not all targets support all bit
6731 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6732 i32 <len>, i32 <align>, i1 <isvolatile>)
6733 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6734 i64 <len>, i32 <align>, i1 <isvolatile>)
6738 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
6739 source location to the destination location. It is similar to the
6740 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
6743 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
6744 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6745 and the pointers can be in specified address spaces.</p>
6749 <p>The first argument is a pointer to the destination, the second is a pointer
6750 to the source. The third argument is an integer argument specifying the
6751 number of bytes to copy, the fourth argument is the alignment of the
6752 source and destination locations, and the fifth is a boolean indicating a
6753 volatile access.</p>
6755 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6756 then the caller guarantees that the source and destination pointers are
6757 aligned to that boundary.</p>
6759 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6760 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
6761 The detailed access behavior is not very cleanly specified and it is unwise
6762 to depend on it.</p>
6766 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
6767 source location to the destination location, which may overlap. It copies
6768 "len" bytes of memory over. If the argument is known to be aligned to some
6769 boundary, this can be specified as the fourth argument, otherwise it should
6770 be set to 0 or 1.</p>
6774 <!-- _______________________________________________________________________ -->
6776 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
6782 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6783 width and for different address spaces. However, not all targets support all
6787 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6788 i32 <len>, i32 <align>, i1 <isvolatile>)
6789 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6790 i64 <len>, i32 <align>, i1 <isvolatile>)
6794 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
6795 particular byte value.</p>
6797 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
6798 intrinsic does not return a value and takes extra alignment/volatile
6799 arguments. Also, the destination can be in an arbitrary address space.</p>
6802 <p>The first argument is a pointer to the destination to fill, the second is the
6803 byte value with which to fill it, the third argument is an integer argument
6804 specifying the number of bytes to fill, and the fourth argument is the known
6805 alignment of the destination location.</p>
6807 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6808 then the caller guarantees that the destination pointer is aligned to that
6811 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6812 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
6813 The detailed access behavior is not very cleanly specified and it is unwise
6814 to depend on it.</p>
6817 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
6818 at the destination location. If the argument is known to be aligned to some
6819 boundary, this can be specified as the fourth argument, otherwise it should
6820 be set to 0 or 1.</p>
6824 <!-- _______________________________________________________________________ -->
6826 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
6832 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
6833 floating point or vector of floating point type. Not all targets support all
6837 declare float @llvm.sqrt.f32(float %Val)
6838 declare double @llvm.sqrt.f64(double %Val)
6839 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6840 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6841 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6845 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
6846 returning the same value as the libm '<tt>sqrt</tt>' functions would.
6847 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
6848 behavior for negative numbers other than -0.0 (which allows for better
6849 optimization, because there is no need to worry about errno being
6850 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
6853 <p>The argument and return value are floating point numbers of the same
6857 <p>This function returns the sqrt of the specified operand if it is a
6858 nonnegative floating point number.</p>
6862 <!-- _______________________________________________________________________ -->
6864 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
6870 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
6871 floating point or vector of floating point type. Not all targets support all
6875 declare float @llvm.powi.f32(float %Val, i32 %power)
6876 declare double @llvm.powi.f64(double %Val, i32 %power)
6877 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6878 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6879 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6883 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
6884 specified (positive or negative) power. The order of evaluation of
6885 multiplications is not defined. When a vector of floating point type is
6886 used, the second argument remains a scalar integer value.</p>
6889 <p>The second argument is an integer power, and the first is a value to raise to
6893 <p>This function returns the first value raised to the second power with an
6894 unspecified sequence of rounding operations.</p>
6898 <!-- _______________________________________________________________________ -->
6900 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
6906 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
6907 floating point or vector of floating point type. Not all targets support all
6911 declare float @llvm.sin.f32(float %Val)
6912 declare double @llvm.sin.f64(double %Val)
6913 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6914 declare fp128 @llvm.sin.f128(fp128 %Val)
6915 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6919 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
6922 <p>The argument and return value are floating point numbers of the same
6926 <p>This function returns the sine of the specified operand, returning the same
6927 values as the libm <tt>sin</tt> functions would, and handles error conditions
6928 in the same way.</p>
6932 <!-- _______________________________________________________________________ -->
6934 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
6940 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
6941 floating point or vector of floating point type. Not all targets support all
6945 declare float @llvm.cos.f32(float %Val)
6946 declare double @llvm.cos.f64(double %Val)
6947 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6948 declare fp128 @llvm.cos.f128(fp128 %Val)
6949 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6953 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
6956 <p>The argument and return value are floating point numbers of the same
6960 <p>This function returns the cosine of the specified operand, returning the same
6961 values as the libm <tt>cos</tt> functions would, and handles error conditions
6962 in the same way.</p>
6966 <!-- _______________________________________________________________________ -->
6968 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
6974 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
6975 floating point or vector of floating point type. Not all targets support all
6979 declare float @llvm.pow.f32(float %Val, float %Power)
6980 declare double @llvm.pow.f64(double %Val, double %Power)
6981 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6982 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6983 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6987 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
6988 specified (positive or negative) power.</p>
6991 <p>The second argument is a floating point power, and the first is a value to
6992 raise to that power.</p>
6995 <p>This function returns the first value raised to the second power, returning
6996 the same values as the libm <tt>pow</tt> functions would, and handles error
6997 conditions in the same way.</p>
7003 <!-- _______________________________________________________________________ -->
7005 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7011 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7012 floating point or vector of floating point type. Not all targets support all
7016 declare float @llvm.exp.f32(float %Val)
7017 declare double @llvm.exp.f64(double %Val)
7018 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7019 declare fp128 @llvm.exp.f128(fp128 %Val)
7020 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7024 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7027 <p>The argument and return value are floating point numbers of the same
7031 <p>This function returns the same values as the libm <tt>exp</tt> functions
7032 would, and handles error conditions in the same way.</p>
7036 <!-- _______________________________________________________________________ -->
7038 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7044 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7045 floating point or vector of floating point type. Not all targets support all
7049 declare float @llvm.log.f32(float %Val)
7050 declare double @llvm.log.f64(double %Val)
7051 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7052 declare fp128 @llvm.log.f128(fp128 %Val)
7053 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7057 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7060 <p>The argument and return value are floating point numbers of the same
7064 <p>This function returns the same values as the libm <tt>log</tt> functions
7065 would, and handles error conditions in the same way.</p>
7068 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7074 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7075 floating point or vector of floating point type. Not all targets support all
7079 declare float @llvm.fma.f32(float %a, float %b, float %c)
7080 declare double @llvm.fma.f64(double %a, double %b, double %c)
7081 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7082 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7083 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7087 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7091 <p>The argument and return value are floating point numbers of the same
7095 <p>This function returns the same values as the libm <tt>fma</tt> functions
7100 <!-- ======================================================================= -->
7102 <a name="int_manip">Bit Manipulation Intrinsics</a>
7107 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7108 These allow efficient code generation for some algorithms.</p>
7110 <!-- _______________________________________________________________________ -->
7112 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7118 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7119 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7122 declare i16 @llvm.bswap.i16(i16 <id>)
7123 declare i32 @llvm.bswap.i32(i32 <id>)
7124 declare i64 @llvm.bswap.i64(i64 <id>)
7128 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7129 values with an even number of bytes (positive multiple of 16 bits). These
7130 are useful for performing operations on data that is not in the target's
7131 native byte order.</p>
7134 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7135 and low byte of the input i16 swapped. Similarly,
7136 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7137 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7138 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7139 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7140 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7141 more, respectively).</p>
7145 <!-- _______________________________________________________________________ -->
7147 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7153 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7154 width, or on any vector with integer elements. Not all targets support all
7155 bit widths or vector types, however.</p>
7158 declare i8 @llvm.ctpop.i8(i8 <src>)
7159 declare i16 @llvm.ctpop.i16(i16 <src>)
7160 declare i32 @llvm.ctpop.i32(i32 <src>)
7161 declare i64 @llvm.ctpop.i64(i64 <src>)
7162 declare i256 @llvm.ctpop.i256(i256 <src>)
7163 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7167 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7171 <p>The only argument is the value to be counted. The argument may be of any
7172 integer type, or a vector with integer elements.
7173 The return type must match the argument type.</p>
7176 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7177 element of a vector.</p>
7181 <!-- _______________________________________________________________________ -->
7183 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7189 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7190 integer bit width, or any vector whose elements are integers. Not all
7191 targets support all bit widths or vector types, however.</p>
7194 declare i8 @llvm.ctlz.i8 (i8 <src>)
7195 declare i16 @llvm.ctlz.i16(i16 <src>)
7196 declare i32 @llvm.ctlz.i32(i32 <src>)
7197 declare i64 @llvm.ctlz.i64(i64 <src>)
7198 declare i256 @llvm.ctlz.i256(i256 <src>)
7199 declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src;gt)
7203 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7204 leading zeros in a variable.</p>
7207 <p>The only argument is the value to be counted. The argument may be of any
7208 integer type, or any vector type with integer element type.
7209 The return type must match the argument type.</p>
7212 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7213 zeros in a variable, or within each element of the vector if the operation
7214 is of vector type. If the src == 0 then the result is the size in bits of
7215 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7219 <!-- _______________________________________________________________________ -->
7221 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7227 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7228 integer bit width, or any vector of integer elements. Not all targets
7229 support all bit widths or vector types, however.</p>
7232 declare i8 @llvm.cttz.i8 (i8 <src>)
7233 declare i16 @llvm.cttz.i16(i16 <src>)
7234 declare i32 @llvm.cttz.i32(i32 <src>)
7235 declare i64 @llvm.cttz.i64(i64 <src>)
7236 declare i256 @llvm.cttz.i256(i256 <src>)
7237 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>)
7241 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7245 <p>The only argument is the value to be counted. The argument may be of any
7246 integer type, or a vectory with integer element type.. The return type
7247 must match the argument type.</p>
7250 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7251 zeros in a variable, or within each element of a vector.
7252 If the src == 0 then the result is the size in bits of
7253 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7259 <!-- ======================================================================= -->
7261 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7266 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7268 <!-- _______________________________________________________________________ -->
7270 <a name="int_sadd_overflow">
7271 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7278 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7279 on any integer bit width.</p>
7282 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7283 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7284 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7288 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7289 a signed addition of the two arguments, and indicate whether an overflow
7290 occurred during the signed summation.</p>
7293 <p>The arguments (%a and %b) and the first element of the result structure may
7294 be of integer types of any bit width, but they must have the same bit
7295 width. The second element of the result structure must be of
7296 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7297 undergo signed addition.</p>
7300 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7301 a signed addition of the two variables. They return a structure — the
7302 first element of which is the signed summation, and the second element of
7303 which is a bit specifying if the signed summation resulted in an
7308 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7309 %sum = extractvalue {i32, i1} %res, 0
7310 %obit = extractvalue {i32, i1} %res, 1
7311 br i1 %obit, label %overflow, label %normal
7316 <!-- _______________________________________________________________________ -->
7318 <a name="int_uadd_overflow">
7319 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7326 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7327 on any integer bit width.</p>
7330 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7331 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7332 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7336 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7337 an unsigned addition of the two arguments, and indicate whether a carry
7338 occurred during the unsigned summation.</p>
7341 <p>The arguments (%a and %b) and the first element of the result structure may
7342 be of integer types of any bit width, but they must have the same bit
7343 width. The second element of the result structure must be of
7344 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7345 undergo unsigned addition.</p>
7348 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7349 an unsigned addition of the two arguments. They return a structure —
7350 the first element of which is the sum, and the second element of which is a
7351 bit specifying if the unsigned summation resulted in a carry.</p>
7355 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7356 %sum = extractvalue {i32, i1} %res, 0
7357 %obit = extractvalue {i32, i1} %res, 1
7358 br i1 %obit, label %carry, label %normal
7363 <!-- _______________________________________________________________________ -->
7365 <a name="int_ssub_overflow">
7366 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7373 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7374 on any integer bit width.</p>
7377 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7378 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7379 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7383 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7384 a signed subtraction of the two arguments, and indicate whether an overflow
7385 occurred during the signed subtraction.</p>
7388 <p>The arguments (%a and %b) and the first element of the result structure may
7389 be of integer types of any bit width, but they must have the same bit
7390 width. The second element of the result structure must be of
7391 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7392 undergo signed subtraction.</p>
7395 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7396 a signed subtraction of the two arguments. They return a structure —
7397 the first element of which is the subtraction, and the second element of
7398 which is a bit specifying if the signed subtraction resulted in an
7403 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7404 %sum = extractvalue {i32, i1} %res, 0
7405 %obit = extractvalue {i32, i1} %res, 1
7406 br i1 %obit, label %overflow, label %normal
7411 <!-- _______________________________________________________________________ -->
7413 <a name="int_usub_overflow">
7414 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7421 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7422 on any integer bit width.</p>
7425 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7426 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7427 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7431 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7432 an unsigned subtraction of the two arguments, and indicate whether an
7433 overflow occurred during the unsigned subtraction.</p>
7436 <p>The arguments (%a and %b) and the first element of the result structure may
7437 be of integer types of any bit width, but they must have the same bit
7438 width. The second element of the result structure must be of
7439 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7440 undergo unsigned subtraction.</p>
7443 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7444 an unsigned subtraction of the two arguments. They return a structure —
7445 the first element of which is the subtraction, and the second element of
7446 which is a bit specifying if the unsigned subtraction resulted in an
7451 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7452 %sum = extractvalue {i32, i1} %res, 0
7453 %obit = extractvalue {i32, i1} %res, 1
7454 br i1 %obit, label %overflow, label %normal
7459 <!-- _______________________________________________________________________ -->
7461 <a name="int_smul_overflow">
7462 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7469 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7470 on any integer bit width.</p>
7473 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7474 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7475 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7480 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7481 a signed multiplication of the two arguments, and indicate whether an
7482 overflow occurred during the signed multiplication.</p>
7485 <p>The arguments (%a and %b) and the first element of the result structure may
7486 be of integer types of any bit width, but they must have the same bit
7487 width. The second element of the result structure must be of
7488 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7489 undergo signed multiplication.</p>
7492 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7493 a signed multiplication of the two arguments. They return a structure —
7494 the first element of which is the multiplication, and the second element of
7495 which is a bit specifying if the signed multiplication resulted in an
7500 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7501 %sum = extractvalue {i32, i1} %res, 0
7502 %obit = extractvalue {i32, i1} %res, 1
7503 br i1 %obit, label %overflow, label %normal
7508 <!-- _______________________________________________________________________ -->
7510 <a name="int_umul_overflow">
7511 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7518 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
7519 on any integer bit width.</p>
7522 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7523 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7524 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7528 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7529 a unsigned multiplication of the two arguments, and indicate whether an
7530 overflow occurred during the unsigned multiplication.</p>
7533 <p>The arguments (%a and %b) and the first element of the result structure may
7534 be of integer types of any bit width, but they must have the same bit
7535 width. The second element of the result structure must be of
7536 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7537 undergo unsigned multiplication.</p>
7540 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7541 an unsigned multiplication of the two arguments. They return a structure
7542 — the first element of which is the multiplication, and the second
7543 element of which is a bit specifying if the unsigned multiplication resulted
7548 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7549 %sum = extractvalue {i32, i1} %res, 0
7550 %obit = extractvalue {i32, i1} %res, 1
7551 br i1 %obit, label %overflow, label %normal
7558 <!-- ======================================================================= -->
7560 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
7565 <p>Half precision floating point is a storage-only format. This means that it is
7566 a dense encoding (in memory) but does not support computation in the
7569 <p>This means that code must first load the half-precision floating point
7570 value as an i16, then convert it to float with <a
7571 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
7572 Computation can then be performed on the float value (including extending to
7573 double etc). To store the value back to memory, it is first converted to
7574 float if needed, then converted to i16 with
7575 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
7576 storing as an i16 value.</p>
7578 <!-- _______________________________________________________________________ -->
7580 <a name="int_convert_to_fp16">
7581 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
7589 declare i16 @llvm.convert.to.fp16(f32 %a)
7593 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7594 a conversion from single precision floating point format to half precision
7595 floating point format.</p>
7598 <p>The intrinsic function contains single argument - the value to be
7602 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7603 a conversion from single precision floating point format to half precision
7604 floating point format. The return value is an <tt>i16</tt> which
7605 contains the converted number.</p>
7609 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7610 store i16 %res, i16* @x, align 2
7615 <!-- _______________________________________________________________________ -->
7617 <a name="int_convert_from_fp16">
7618 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
7626 declare f32 @llvm.convert.from.fp16(i16 %a)
7630 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
7631 a conversion from half precision floating point format to single precision
7632 floating point format.</p>
7635 <p>The intrinsic function contains single argument - the value to be
7639 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
7640 conversion from half single precision floating point format to single
7641 precision floating point format. The input half-float value is represented by
7642 an <tt>i16</tt> value.</p>
7646 %a = load i16* @x, align 2
7647 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7654 <!-- ======================================================================= -->
7656 <a name="int_debugger">Debugger Intrinsics</a>
7661 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
7662 prefix), are described in
7663 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
7664 Level Debugging</a> document.</p>
7668 <!-- ======================================================================= -->
7670 <a name="int_eh">Exception Handling Intrinsics</a>
7675 <p>The LLVM exception handling intrinsics (which all start with
7676 <tt>llvm.eh.</tt> prefix), are described in
7677 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
7678 Handling</a> document.</p>
7682 <!-- ======================================================================= -->
7684 <a name="int_trampoline">Trampoline Intrinsics</a>
7689 <p>These intrinsics make it possible to excise one parameter, marked with
7690 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
7691 The result is a callable
7692 function pointer lacking the nest parameter - the caller does not need to
7693 provide a value for it. Instead, the value to use is stored in advance in a
7694 "trampoline", a block of memory usually allocated on the stack, which also
7695 contains code to splice the nest value into the argument list. This is used
7696 to implement the GCC nested function address extension.</p>
7698 <p>For example, if the function is
7699 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
7700 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
7703 <pre class="doc_code">
7704 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7705 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
7706 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
7707 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
7708 %fp = bitcast i8* %p to i32 (i32, i32)*
7711 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
7712 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
7714 <!-- _______________________________________________________________________ -->
7717 '<tt>llvm.init.trampoline</tt>' Intrinsic
7725 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
7729 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
7730 turning it into a trampoline.</p>
7733 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
7734 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
7735 sufficiently aligned block of memory; this memory is written to by the
7736 intrinsic. Note that the size and the alignment are target-specific - LLVM
7737 currently provides no portable way of determining them, so a front-end that
7738 generates this intrinsic needs to have some target-specific knowledge.
7739 The <tt>func</tt> argument must hold a function bitcast to
7740 an <tt>i8*</tt>.</p>
7743 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
7744 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
7745 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
7746 which can be <a href="#int_trampoline">bitcast (to a new function) and
7747 called</a>. The new function's signature is the same as that of
7748 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
7749 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
7750 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
7751 with the same argument list, but with <tt>nval</tt> used for the missing
7752 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
7753 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
7754 to the returned function pointer is undefined.</p>
7757 <!-- _______________________________________________________________________ -->
7760 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
7768 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
7772 <p>This performs any required machine-specific adjustment to the address of a
7773 trampoline (passed as <tt>tramp</tt>).</p>
7776 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
7777 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
7781 <p>On some architectures the address of the code to be executed needs to be
7782 different to the address where the trampoline is actually stored. This
7783 intrinsic returns the executable address corresponding to <tt>tramp</tt>
7784 after performing the required machine specific adjustments.
7785 The pointer returned can then be <a href="#int_trampoline"> bitcast and
7793 <!-- ======================================================================= -->
7795 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
7800 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
7801 hardware constructs for atomic operations and memory synchronization. This
7802 provides an interface to the hardware, not an interface to the programmer. It
7803 is aimed at a low enough level to allow any programming models or APIs
7804 (Application Programming Interfaces) which need atomic behaviors to map
7805 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
7806 hardware provides a "universal IR" for source languages, it also provides a
7807 starting point for developing a "universal" atomic operation and
7808 synchronization IR.</p>
7810 <p>These do <em>not</em> form an API such as high-level threading libraries,
7811 software transaction memory systems, atomic primitives, and intrinsic
7812 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
7813 application libraries. The hardware interface provided by LLVM should allow
7814 a clean implementation of all of these APIs and parallel programming models.
7815 No one model or paradigm should be selected above others unless the hardware
7816 itself ubiquitously does so.</p>
7818 <!-- _______________________________________________________________________ -->
7820 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
7826 declare void @llvm.memory.barrier(i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device>)
7830 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
7831 specific pairs of memory access types.</p>
7834 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
7835 The first four arguments enables a specific barrier as listed below. The
7836 fifth argument specifies that the barrier applies to io or device or uncached
7840 <li><tt>ll</tt>: load-load barrier</li>
7841 <li><tt>ls</tt>: load-store barrier</li>
7842 <li><tt>sl</tt>: store-load barrier</li>
7843 <li><tt>ss</tt>: store-store barrier</li>
7844 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
7848 <p>This intrinsic causes the system to enforce some ordering constraints upon
7849 the loads and stores of the program. This barrier does not
7850 indicate <em>when</em> any events will occur, it only enforces
7851 an <em>order</em> in which they occur. For any of the specified pairs of load
7852 and store operations (f.ex. load-load, or store-load), all of the first
7853 operations preceding the barrier will complete before any of the second
7854 operations succeeding the barrier begin. Specifically the semantics for each
7855 pairing is as follows:</p>
7858 <li><tt>ll</tt>: All loads before the barrier must complete before any load
7859 after the barrier begins.</li>
7860 <li><tt>ls</tt>: All loads before the barrier must complete before any
7861 store after the barrier begins.</li>
7862 <li><tt>ss</tt>: All stores before the barrier must complete before any
7863 store after the barrier begins.</li>
7864 <li><tt>sl</tt>: All stores before the barrier must complete before any
7865 load after the barrier begins.</li>
7868 <p>These semantics are applied with a logical "and" behavior when more than one
7869 is enabled in a single memory barrier intrinsic.</p>
7871 <p>Backends may implement stronger barriers than those requested when they do
7872 not support as fine grained a barrier as requested. Some architectures do
7873 not need all types of barriers and on such architectures, these become
7878 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7879 %ptr = bitcast i8* %mallocP to i32*
7882 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
7883 call void @llvm.memory.barrier(i1 false, i1 true, i1 false, i1 false, i1 true)
7884 <i>; guarantee the above finishes</i>
7885 store i32 8, %ptr <i>; before this begins</i>
7890 <!-- _______________________________________________________________________ -->
7892 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
7898 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
7899 any integer bit width and for different address spaces. Not all targets
7900 support all bit widths however.</p>
7903 declare i8 @llvm.atomic.cmp.swap.i8.p0i8(i8* <ptr>, i8 <cmp>, i8 <val>)
7904 declare i16 @llvm.atomic.cmp.swap.i16.p0i16(i16* <ptr>, i16 <cmp>, i16 <val>)
7905 declare i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* <ptr>, i32 <cmp>, i32 <val>)
7906 declare i64 @llvm.atomic.cmp.swap.i64.p0i64(i64* <ptr>, i64 <cmp>, i64 <val>)
7910 <p>This loads a value in memory and compares it to a given value. If they are
7911 equal, it stores a new value into the memory.</p>
7914 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
7915 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
7916 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
7917 this integer type. While any bit width integer may be used, targets may only
7918 lower representations they support in hardware.</p>
7921 <p>This entire intrinsic must be executed atomically. It first loads the value
7922 in memory pointed to by <tt>ptr</tt> and compares it with the
7923 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
7924 memory. The loaded value is yielded in all cases. This provides the
7925 equivalent of an atomic compare-and-swap operation within the SSA
7930 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7931 %ptr = bitcast i8* %mallocP to i32*
7934 %val1 = add i32 4, 4
7935 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 4, %val1)
7936 <i>; yields {i32}:result1 = 4</i>
7937 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7938 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7940 %val2 = add i32 1, 1
7941 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 5, %val2)
7942 <i>; yields {i32}:result2 = 8</i>
7943 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
7945 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
7950 <!-- _______________________________________________________________________ -->
7952 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
7958 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
7959 integer bit width. Not all targets support all bit widths however.</p>
7962 declare i8 @llvm.atomic.swap.i8.p0i8(i8* <ptr>, i8 <val>)
7963 declare i16 @llvm.atomic.swap.i16.p0i16(i16* <ptr>, i16 <val>)
7964 declare i32 @llvm.atomic.swap.i32.p0i32(i32* <ptr>, i32 <val>)
7965 declare i64 @llvm.atomic.swap.i64.p0i64(i64* <ptr>, i64 <val>)
7969 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
7970 the value from memory. It then stores the value in <tt>val</tt> in the memory
7971 at <tt>ptr</tt>.</p>
7974 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
7975 the <tt>val</tt> argument and the result must be integers of the same bit
7976 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
7977 integer type. The targets may only lower integer representations they
7981 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
7982 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
7983 equivalent of an atomic swap operation within the SSA framework.</p>
7987 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7988 %ptr = bitcast i8* %mallocP to i32*
7991 %val1 = add i32 4, 4
7992 %result1 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val1)
7993 <i>; yields {i32}:result1 = 4</i>
7994 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7995 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7997 %val2 = add i32 1, 1
7998 %result2 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val2)
7999 <i>; yields {i32}:result2 = 8</i>
8001 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
8002 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
8007 <!-- _______________________________________________________________________ -->
8009 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
8015 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
8016 any integer bit width. Not all targets support all bit widths however.</p>
8019 declare i8 @llvm.atomic.load.add.i8.p0i8(i8* <ptr>, i8 <delta>)
8020 declare i16 @llvm.atomic.load.add.i16.p0i16(i16* <ptr>, i16 <delta>)
8021 declare i32 @llvm.atomic.load.add.i32.p0i32(i32* <ptr>, i32 <delta>)
8022 declare i64 @llvm.atomic.load.add.i64.p0i64(i64* <ptr>, i64 <delta>)
8026 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
8027 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
8030 <p>The intrinsic takes two arguments, the first a pointer to an integer value
8031 and the second an integer value. The result is also an integer value. These
8032 integer types can have any bit width, but they must all have the same bit
8033 width. The targets may only lower integer representations they support.</p>
8036 <p>This intrinsic does a series of operations atomically. It first loads the
8037 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
8038 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
8042 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8043 %ptr = bitcast i8* %mallocP to i32*
8045 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 4)
8046 <i>; yields {i32}:result1 = 4</i>
8047 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 2)
8048 <i>; yields {i32}:result2 = 8</i>
8049 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 5)
8050 <i>; yields {i32}:result3 = 10</i>
8051 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
8056 <!-- _______________________________________________________________________ -->
8058 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
8064 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
8065 any integer bit width and for different address spaces. Not all targets
8066 support all bit widths however.</p>
8069 declare i8 @llvm.atomic.load.sub.i8.p0i32(i8* <ptr>, i8 <delta>)
8070 declare i16 @llvm.atomic.load.sub.i16.p0i32(i16* <ptr>, i16 <delta>)
8071 declare i32 @llvm.atomic.load.sub.i32.p0i32(i32* <ptr>, i32 <delta>)
8072 declare i64 @llvm.atomic.load.sub.i64.p0i32(i64* <ptr>, i64 <delta>)
8076 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
8077 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
8080 <p>The intrinsic takes two arguments, the first a pointer to an integer value
8081 and the second an integer value. The result is also an integer value. These
8082 integer types can have any bit width, but they must all have the same bit
8083 width. The targets may only lower integer representations they support.</p>
8086 <p>This intrinsic does a series of operations atomically. It first loads the
8087 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
8088 result to <tt>ptr</tt>. It yields the original value stored
8089 at <tt>ptr</tt>.</p>
8093 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8094 %ptr = bitcast i8* %mallocP to i32*
8096 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 4)
8097 <i>; yields {i32}:result1 = 8</i>
8098 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 2)
8099 <i>; yields {i32}:result2 = 4</i>
8100 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 5)
8101 <i>; yields {i32}:result3 = 2</i>
8102 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
8107 <!-- _______________________________________________________________________ -->
8109 <a name="int_atomic_load_and">
8110 '<tt>llvm.atomic.load.and.*</tt>' Intrinsic
8113 <a name="int_atomic_load_nand">
8114 '<tt>llvm.atomic.load.nand.*</tt>' Intrinsic
8117 <a name="int_atomic_load_or">
8118 '<tt>llvm.atomic.load.or.*</tt>' Intrinsic
8121 <a name="int_atomic_load_xor">
8122 '<tt>llvm.atomic.load.xor.*</tt>' Intrinsic
8129 <p>These are overloaded intrinsics. You can
8130 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
8131 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
8132 bit width and for different address spaces. Not all targets support all bit
8136 declare i8 @llvm.atomic.load.and.i8.p0i8(i8* <ptr>, i8 <delta>)
8137 declare i16 @llvm.atomic.load.and.i16.p0i16(i16* <ptr>, i16 <delta>)
8138 declare i32 @llvm.atomic.load.and.i32.p0i32(i32* <ptr>, i32 <delta>)
8139 declare i64 @llvm.atomic.load.and.i64.p0i64(i64* <ptr>, i64 <delta>)
8143 declare i8 @llvm.atomic.load.or.i8.p0i8(i8* <ptr>, i8 <delta>)
8144 declare i16 @llvm.atomic.load.or.i16.p0i16(i16* <ptr>, i16 <delta>)
8145 declare i32 @llvm.atomic.load.or.i32.p0i32(i32* <ptr>, i32 <delta>)
8146 declare i64 @llvm.atomic.load.or.i64.p0i64(i64* <ptr>, i64 <delta>)
8150 declare i8 @llvm.atomic.load.nand.i8.p0i32(i8* <ptr>, i8 <delta>)
8151 declare i16 @llvm.atomic.load.nand.i16.p0i32(i16* <ptr>, i16 <delta>)
8152 declare i32 @llvm.atomic.load.nand.i32.p0i32(i32* <ptr>, i32 <delta>)
8153 declare i64 @llvm.atomic.load.nand.i64.p0i32(i64* <ptr>, i64 <delta>)
8157 declare i8 @llvm.atomic.load.xor.i8.p0i32(i8* <ptr>, i8 <delta>)
8158 declare i16 @llvm.atomic.load.xor.i16.p0i32(i16* <ptr>, i16 <delta>)
8159 declare i32 @llvm.atomic.load.xor.i32.p0i32(i32* <ptr>, i32 <delta>)
8160 declare i64 @llvm.atomic.load.xor.i64.p0i32(i64* <ptr>, i64 <delta>)
8164 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
8165 the value stored in memory at <tt>ptr</tt>. It yields the original value
8166 at <tt>ptr</tt>.</p>
8169 <p>These intrinsics take two arguments, the first a pointer to an integer value
8170 and the second an integer value. The result is also an integer value. These
8171 integer types can have any bit width, but they must all have the same bit
8172 width. The targets may only lower integer representations they support.</p>
8175 <p>These intrinsics does a series of operations atomically. They first load the
8176 value stored at <tt>ptr</tt>. They then do the bitwise
8177 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
8178 original value stored at <tt>ptr</tt>.</p>
8182 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8183 %ptr = bitcast i8* %mallocP to i32*
8184 store i32 0x0F0F, %ptr
8185 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32(i32* %ptr, i32 0xFF)
8186 <i>; yields {i32}:result0 = 0x0F0F</i>
8187 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32(i32* %ptr, i32 0xFF)
8188 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
8189 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32(i32* %ptr, i32 0F)
8190 <i>; yields {i32}:result2 = 0xF0</i>
8191 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32(i32* %ptr, i32 0F)
8192 <i>; yields {i32}:result3 = FF</i>
8193 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
8198 <!-- _______________________________________________________________________ -->
8200 <a name="int_atomic_load_max">
8201 '<tt>llvm.atomic.load.max.*</tt>' Intrinsic
8204 <a name="int_atomic_load_min">
8205 '<tt>llvm.atomic.load.min.*</tt>' Intrinsic
8208 <a name="int_atomic_load_umax">
8209 '<tt>llvm.atomic.load.umax.*</tt>' Intrinsic
8212 <a name="int_atomic_load_umin">
8213 '<tt>llvm.atomic.load.umin.*</tt>' Intrinsic
8220 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
8221 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
8222 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
8223 address spaces. Not all targets support all bit widths however.</p>
8226 declare i8 @llvm.atomic.load.max.i8.p0i8(i8* <ptr>, i8 <delta>)
8227 declare i16 @llvm.atomic.load.max.i16.p0i16(i16* <ptr>, i16 <delta>)
8228 declare i32 @llvm.atomic.load.max.i32.p0i32(i32* <ptr>, i32 <delta>)
8229 declare i64 @llvm.atomic.load.max.i64.p0i64(i64* <ptr>, i64 <delta>)
8233 declare i8 @llvm.atomic.load.min.i8.p0i8(i8* <ptr>, i8 <delta>)
8234 declare i16 @llvm.atomic.load.min.i16.p0i16(i16* <ptr>, i16 <delta>)
8235 declare i32 @llvm.atomic.load.min.i32.p0i32(i32* <ptr>, i32 <delta>)
8236 declare i64 @llvm.atomic.load.min.i64.p0i64(i64* <ptr>, i64 <delta>)
8240 declare i8 @llvm.atomic.load.umax.i8.p0i8(i8* <ptr>, i8 <delta>)
8241 declare i16 @llvm.atomic.load.umax.i16.p0i16(i16* <ptr>, i16 <delta>)
8242 declare i32 @llvm.atomic.load.umax.i32.p0i32(i32* <ptr>, i32 <delta>)
8243 declare i64 @llvm.atomic.load.umax.i64.p0i64(i64* <ptr>, i64 <delta>)
8247 declare i8 @llvm.atomic.load.umin.i8.p0i8(i8* <ptr>, i8 <delta>)
8248 declare i16 @llvm.atomic.load.umin.i16.p0i16(i16* <ptr>, i16 <delta>)
8249 declare i32 @llvm.atomic.load.umin.i32.p0i32(i32* <ptr>, i32 <delta>)
8250 declare i64 @llvm.atomic.load.umin.i64.p0i64(i64* <ptr>, i64 <delta>)
8254 <p>These intrinsics takes the signed or unsigned minimum or maximum of
8255 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
8256 original value at <tt>ptr</tt>.</p>
8259 <p>These intrinsics take two arguments, the first a pointer to an integer value
8260 and the second an integer value. The result is also an integer value. These
8261 integer types can have any bit width, but they must all have the same bit
8262 width. The targets may only lower integer representations they support.</p>
8265 <p>These intrinsics does a series of operations atomically. They first load the
8266 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
8267 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
8268 yield the original value stored at <tt>ptr</tt>.</p>
8272 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8273 %ptr = bitcast i8* %mallocP to i32*
8275 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32(i32* %ptr, i32 -2)
8276 <i>; yields {i32}:result0 = 7</i>
8277 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32(i32* %ptr, i32 8)
8278 <i>; yields {i32}:result1 = -2</i>
8279 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32(i32* %ptr, i32 10)
8280 <i>; yields {i32}:result2 = 8</i>
8281 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32(i32* %ptr, i32 30)
8282 <i>; yields {i32}:result3 = 8</i>
8283 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
8290 <!-- ======================================================================= -->
8292 <a name="int_memorymarkers">Memory Use Markers</a>
8297 <p>This class of intrinsics exists to information about the lifetime of memory
8298 objects and ranges where variables are immutable.</p>
8300 <!-- _______________________________________________________________________ -->
8302 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8309 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8313 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8314 object's lifetime.</p>
8317 <p>The first argument is a constant integer representing the size of the
8318 object, or -1 if it is variable sized. The second argument is a pointer to
8322 <p>This intrinsic indicates that before this point in the code, the value of the
8323 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8324 never be used and has an undefined value. A load from the pointer that
8325 precedes this intrinsic can be replaced with
8326 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8330 <!-- _______________________________________________________________________ -->
8332 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8339 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8343 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8344 object's lifetime.</p>
8347 <p>The first argument is a constant integer representing the size of the
8348 object, or -1 if it is variable sized. The second argument is a pointer to
8352 <p>This intrinsic indicates that after this point in the code, the value of the
8353 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8354 never be used and has an undefined value. Any stores into the memory object
8355 following this intrinsic may be removed as dead.
8359 <!-- _______________________________________________________________________ -->
8361 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8368 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8372 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8373 a memory object will not change.</p>
8376 <p>The first argument is a constant integer representing the size of the
8377 object, or -1 if it is variable sized. The second argument is a pointer to
8381 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8382 the return value, the referenced memory location is constant and
8387 <!-- _______________________________________________________________________ -->
8389 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8396 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8400 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8401 a memory object are mutable.</p>
8404 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8405 The second argument is a constant integer representing the size of the
8406 object, or -1 if it is variable sized and the third argument is a pointer
8410 <p>This intrinsic indicates that the memory is mutable again.</p>
8416 <!-- ======================================================================= -->
8418 <a name="int_general">General Intrinsics</a>
8423 <p>This class of intrinsics is designed to be generic and has no specific
8426 <!-- _______________________________________________________________________ -->
8428 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8435 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8439 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8442 <p>The first argument is a pointer to a value, the second is a pointer to a
8443 global string, the third is a pointer to a global string which is the source
8444 file name, and the last argument is the line number.</p>
8447 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8448 This can be useful for special purpose optimizations that want to look for
8449 these annotations. These have no other defined use; they are ignored by code
8450 generation and optimization.</p>
8454 <!-- _______________________________________________________________________ -->
8456 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8462 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8463 any integer bit width.</p>
8466 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8467 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8468 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8469 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8470 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8474 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8477 <p>The first argument is an integer value (result of some expression), the
8478 second is a pointer to a global string, the third is a pointer to a global
8479 string which is the source file name, and the last argument is the line
8480 number. It returns the value of the first argument.</p>
8483 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8484 arbitrary strings. This can be useful for special purpose optimizations that
8485 want to look for these annotations. These have no other defined use; they
8486 are ignored by code generation and optimization.</p>
8490 <!-- _______________________________________________________________________ -->
8492 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8499 declare void @llvm.trap()
8503 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8509 <p>This intrinsics is lowered to the target dependent trap instruction. If the
8510 target does not have a trap instruction, this intrinsic will be lowered to
8511 the call of the <tt>abort()</tt> function.</p>
8515 <!-- _______________________________________________________________________ -->
8517 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8524 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8528 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8529 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8530 ensure that it is placed on the stack before local variables.</p>
8533 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8534 arguments. The first argument is the value loaded from the stack
8535 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8536 that has enough space to hold the value of the guard.</p>
8539 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8540 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8541 stack. This is to ensure that if a local variable on the stack is
8542 overwritten, it will destroy the value of the guard. When the function exits,
8543 the guard on the stack is checked against the original guard. If they are
8544 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8549 <!-- _______________________________________________________________________ -->
8551 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8558 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <type>)
8559 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <type>)
8563 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8564 the optimizers to determine at compile time whether a) an operation (like
8565 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8566 runtime check for overflow isn't necessary. An object in this context means
8567 an allocation of a specific class, structure, array, or other object.</p>
8570 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8571 argument is a pointer to or into the <tt>object</tt>. The second argument
8572 is a boolean 0 or 1. This argument determines whether you want the
8573 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
8574 1, variables are not allowed.</p>
8577 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
8578 representing the size of the object concerned, or <tt>i32/i64 -1 or 0</tt>,
8579 depending on the <tt>type</tt> argument, if the size cannot be determined at
8588 <!-- *********************************************************************** -->
8591 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8592 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8593 <a href="http://validator.w3.org/check/referer"><img
8594 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8596 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8597 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8598 Last modified: $Date$