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 <div class="doc_title"> LLVM Language Reference Manual </div>
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_internal">'<tt>internal</tt>' Linkage</a></li>
28 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
29 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
30 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
31 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
32 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
33 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
34 <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
35 <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
36 <li><a href="#linkage_external">'<tt>externally visible</tt>' Linkage</a></li>
37 <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
38 <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
41 <li><a href="#callingconv">Calling Conventions</a></li>
42 <li><a href="#namedtypes">Named Types</a></li>
43 <li><a href="#globalvars">Global Variables</a></li>
44 <li><a href="#functionstructure">Functions</a></li>
45 <li><a href="#aliasstructure">Aliases</a></li>
46 <li><a href="#namedmetadatastructure">Named Metadata</a></li>
47 <li><a href="#paramattrs">Parameter Attributes</a></li>
48 <li><a href="#fnattrs">Function Attributes</a></li>
49 <li><a href="#gc">Garbage Collector Names</a></li>
50 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
51 <li><a href="#datalayout">Data Layout</a></li>
52 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
55 <li><a href="#typesystem">Type System</a>
57 <li><a href="#t_classifications">Type Classifications</a></li>
58 <li><a href="#t_primitive">Primitive Types</a>
60 <li><a href="#t_integer">Integer Type</a></li>
61 <li><a href="#t_floating">Floating Point Types</a></li>
62 <li><a href="#t_void">Void Type</a></li>
63 <li><a href="#t_label">Label Type</a></li>
64 <li><a href="#t_metadata">Metadata Type</a></li>
67 <li><a href="#t_derived">Derived Types</a>
69 <li><a href="#t_aggregate">Aggregate Types</a>
71 <li><a href="#t_array">Array Type</a></li>
72 <li><a href="#t_struct">Structure Type</a></li>
73 <li><a href="#t_pstruct">Packed Structure Type</a></li>
74 <li><a href="#t_union">Union Type</a></li>
75 <li><a href="#t_vector">Vector Type</a></li>
78 <li><a href="#t_function">Function Type</a></li>
79 <li><a href="#t_pointer">Pointer Type</a></li>
80 <li><a href="#t_opaque">Opaque Type</a></li>
83 <li><a href="#t_uprefs">Type Up-references</a></li>
86 <li><a href="#constants">Constants</a>
88 <li><a href="#simpleconstants">Simple Constants</a></li>
89 <li><a href="#complexconstants">Complex Constants</a></li>
90 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
91 <li><a href="#undefvalues">Undefined Values</a></li>
92 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
93 <li><a href="#constantexprs">Constant Expressions</a></li>
96 <li><a href="#othervalues">Other Values</a>
98 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
99 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a></li>
102 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
104 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
105 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
106 Global Variable</a></li>
107 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
108 Global Variable</a></li>
109 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
110 Global Variable</a></li>
113 <li><a href="#instref">Instruction Reference</a>
115 <li><a href="#terminators">Terminator Instructions</a>
117 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
118 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
119 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
120 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
121 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
122 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
123 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
126 <li><a href="#binaryops">Binary Operations</a>
128 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
129 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
130 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
131 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
132 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
133 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
134 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
135 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
136 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
137 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
138 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
139 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
142 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
144 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
145 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
146 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
147 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
148 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
149 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
152 <li><a href="#vectorops">Vector Operations</a>
154 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
155 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
156 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
159 <li><a href="#aggregateops">Aggregate Operations</a>
161 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
162 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
165 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
167 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
168 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
169 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
170 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
173 <li><a href="#convertops">Conversion Operations</a>
175 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
176 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
177 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
178 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
179 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
180 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
181 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
182 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
183 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
184 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
185 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
186 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
189 <li><a href="#otherops">Other Operations</a>
191 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
192 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
193 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
194 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
195 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
196 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
201 <li><a href="#intrinsics">Intrinsic Functions</a>
203 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
205 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
206 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
207 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
210 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
212 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
213 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
214 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
217 <li><a href="#int_codegen">Code Generator Intrinsics</a>
219 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
220 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
221 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
222 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
223 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
224 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
225 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
228 <li><a href="#int_libc">Standard C Library Intrinsics</a>
230 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
231 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
232 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
233 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
234 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
235 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
236 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
237 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
240 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
242 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
243 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
244 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
245 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
248 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
250 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
251 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
252 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
253 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
254 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
255 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
258 <li><a href="#int_debugger">Debugger intrinsics</a></li>
259 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
260 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
262 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
265 <li><a href="#int_atomics">Atomic intrinsics</a>
267 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
268 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
269 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
270 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
271 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
272 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
273 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
274 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
275 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
276 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
277 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
278 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
279 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
282 <li><a href="#int_memorymarkers">Memory Use Markers</a>
284 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
285 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
286 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
287 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
290 <li><a href="#int_general">General intrinsics</a>
292 <li><a href="#int_var_annotation">
293 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
294 <li><a href="#int_annotation">
295 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
296 <li><a href="#int_trap">
297 '<tt>llvm.trap</tt>' Intrinsic</a></li>
298 <li><a href="#int_stackprotector">
299 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
300 <li><a href="#int_objectsize">
301 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
308 <div class="doc_author">
309 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
310 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
313 <!-- *********************************************************************** -->
314 <div class="doc_section"> <a name="abstract">Abstract </a></div>
315 <!-- *********************************************************************** -->
317 <div class="doc_text">
319 <p>This document is a reference manual for the LLVM assembly language. LLVM is
320 a Static Single Assignment (SSA) based representation that provides type
321 safety, low-level operations, flexibility, and the capability of representing
322 'all' high-level languages cleanly. It is the common code representation
323 used throughout all phases of the LLVM compilation strategy.</p>
327 <!-- *********************************************************************** -->
328 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
329 <!-- *********************************************************************** -->
331 <div class="doc_text">
333 <p>The LLVM code representation is designed to be used in three different forms:
334 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
335 for fast loading by a Just-In-Time compiler), and as a human readable
336 assembly language representation. This allows LLVM to provide a powerful
337 intermediate representation for efficient compiler transformations and
338 analysis, while providing a natural means to debug and visualize the
339 transformations. The three different forms of LLVM are all equivalent. This
340 document describes the human readable representation and notation.</p>
342 <p>The LLVM representation aims to be light-weight and low-level while being
343 expressive, typed, and extensible at the same time. It aims to be a
344 "universal IR" of sorts, by being at a low enough level that high-level ideas
345 may be cleanly mapped to it (similar to how microprocessors are "universal
346 IR's", allowing many source languages to be mapped to them). By providing
347 type information, LLVM can be used as the target of optimizations: for
348 example, through pointer analysis, it can be proven that a C automatic
349 variable is never accessed outside of the current function, allowing it to
350 be promoted to a simple SSA value instead of a memory location.</p>
354 <!-- _______________________________________________________________________ -->
355 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
357 <div class="doc_text">
359 <p>It is important to note that this document describes 'well formed' LLVM
360 assembly language. There is a difference between what the parser accepts and
361 what is considered 'well formed'. For example, the following instruction is
362 syntactically okay, but not well formed:</p>
364 <div class="doc_code">
366 %x = <a href="#i_add">add</a> i32 1, %x
370 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
371 LLVM infrastructure provides a verification pass that may be used to verify
372 that an LLVM module is well formed. This pass is automatically run by the
373 parser after parsing input assembly and by the optimizer before it outputs
374 bitcode. The violations pointed out by the verifier pass indicate bugs in
375 transformation passes or input to the parser.</p>
379 <!-- Describe the typesetting conventions here. -->
381 <!-- *********************************************************************** -->
382 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
383 <!-- *********************************************************************** -->
385 <div class="doc_text">
387 <p>LLVM identifiers come in two basic types: global and local. Global
388 identifiers (functions, global variables) begin with the <tt>'@'</tt>
389 character. Local identifiers (register names, types) begin with
390 the <tt>'%'</tt> character. Additionally, there are three different formats
391 for identifiers, for different purposes:</p>
394 <li>Named values are represented as a string of characters with their prefix.
395 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
396 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
397 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
398 other characters in their names can be surrounded with quotes. Special
399 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
400 ASCII code for the character in hexadecimal. In this way, any character
401 can be used in a name value, even quotes themselves.</li>
403 <li>Unnamed values are represented as an unsigned numeric value with their
404 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
406 <li>Constants, which are described in a <a href="#constants">section about
407 constants</a>, below.</li>
410 <p>LLVM requires that values start with a prefix for two reasons: Compilers
411 don't need to worry about name clashes with reserved words, and the set of
412 reserved words may be expanded in the future without penalty. Additionally,
413 unnamed identifiers allow a compiler to quickly come up with a temporary
414 variable without having to avoid symbol table conflicts.</p>
416 <p>Reserved words in LLVM are very similar to reserved words in other
417 languages. There are keywords for different opcodes
418 ('<tt><a href="#i_add">add</a></tt>',
419 '<tt><a href="#i_bitcast">bitcast</a></tt>',
420 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
421 ('<tt><a href="#t_void">void</a></tt>',
422 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
423 reserved words cannot conflict with variable names, because none of them
424 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
426 <p>Here is an example of LLVM code to multiply the integer variable
427 '<tt>%X</tt>' by 8:</p>
431 <div class="doc_code">
433 %result = <a href="#i_mul">mul</a> i32 %X, 8
437 <p>After strength reduction:</p>
439 <div class="doc_code">
441 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
445 <p>And the hard way:</p>
447 <div class="doc_code">
449 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
450 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
451 %result = <a href="#i_add">add</a> i32 %1, %1
455 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
456 lexical features of LLVM:</p>
459 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
462 <li>Unnamed temporaries are created when the result of a computation is not
463 assigned to a named value.</li>
465 <li>Unnamed temporaries are numbered sequentially</li>
468 <p>It also shows a convention that we follow in this document. When
469 demonstrating instructions, we will follow an instruction with a comment that
470 defines the type and name of value produced. Comments are shown in italic
475 <!-- *********************************************************************** -->
476 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
477 <!-- *********************************************************************** -->
479 <!-- ======================================================================= -->
480 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
483 <div class="doc_text">
485 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
486 of the input programs. Each module consists of functions, global variables,
487 and symbol table entries. Modules may be combined together with the LLVM
488 linker, which merges function (and global variable) definitions, resolves
489 forward declarations, and merges symbol table entries. Here is an example of
490 the "hello world" module:</p>
492 <div class="doc_code">
494 <i>; Declare the string constant as a global constant.</i>
495 <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>
497 <i>; External declaration of the puts function</i>
498 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
500 <i>; Definition of main function</i>
501 define i32 @main() { <i>; i32()* </i>
502 <i>; Convert [13 x i8]* to i8 *...</i>
503 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
505 <i>; Call puts function to write out the string to stdout.</i>
506 <a href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
507 <a href="#i_ret">ret</a> i32 0<br>}
509 <i>; Named metadata</i>
510 !1 = metadata !{i32 41}
515 <p>This example is made up of a <a href="#globalvars">global variable</a> named
516 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
517 a <a href="#functionstructure">function definition</a> for
518 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
521 <p>In general, a module is made up of a list of global values, where both
522 functions and global variables are global values. Global values are
523 represented by a pointer to a memory location (in this case, a pointer to an
524 array of char, and a pointer to a function), and have one of the
525 following <a href="#linkage">linkage types</a>.</p>
529 <!-- ======================================================================= -->
530 <div class="doc_subsection">
531 <a name="linkage">Linkage Types</a>
534 <div class="doc_text">
536 <p>All Global Variables and Functions have one of the following types of
540 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
541 <dd>Global values with private linkage are only directly accessible by objects
542 in the current module. In particular, linking code into a module with an
543 private global value may cause the private to be renamed as necessary to
544 avoid collisions. Because the symbol is private to the module, all
545 references can be updated. This doesn't show up in any symbol table in the
548 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
549 <dd>Similar to private, but the symbol is passed through the assembler and
550 removed by the linker after evaluation. Note that (unlike private
551 symbols) linker_private symbols are subject to coalescing by the linker:
552 weak symbols get merged and redefinitions are rejected. However, unlike
553 normal strong symbols, they are removed by the linker from the final
554 linked image (executable or dynamic library).</dd>
556 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
557 <dd>Similar to private, but the value shows as a local symbol
558 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
559 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
561 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
562 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
563 into the object file corresponding to the LLVM module. They exist to
564 allow inlining and other optimizations to take place given knowledge of
565 the definition of the global, which is known to be somewhere outside the
566 module. Globals with <tt>available_externally</tt> linkage are allowed to
567 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
568 This linkage type is only allowed on definitions, not declarations.</dd>
570 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
571 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
572 the same name when linkage occurs. This can be used to implement
573 some forms of inline functions, templates, or other code which must be
574 generated in each translation unit that uses it, but where the body may
575 be overridden with a more definitive definition later. Unreferenced
576 <tt>linkonce</tt> globals are allowed to be discarded. Note that
577 <tt>linkonce</tt> linkage does not actually allow the optimizer to
578 inline the body of this function into callers because it doesn't know if
579 this definition of the function is the definitive definition within the
580 program or whether it will be overridden by a stronger definition.
581 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
584 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
585 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
586 <tt>linkonce</tt> linkage, except that unreferenced globals with
587 <tt>weak</tt> linkage may not be discarded. This is used for globals that
588 are declared "weak" in C source code.</dd>
590 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
591 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
592 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
594 Symbols with "<tt>common</tt>" linkage are merged in the same way as
595 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
596 <tt>common</tt> symbols may not have an explicit section,
597 must have a zero initializer, and may not be marked '<a
598 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
599 have common linkage.</dd>
602 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
603 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
604 pointer to array type. When two global variables with appending linkage
605 are linked together, the two global arrays are appended together. This is
606 the LLVM, typesafe, equivalent of having the system linker append together
607 "sections" with identical names when .o files are linked.</dd>
609 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
610 <dd>The semantics of this linkage follow the ELF object file model: the symbol
611 is weak until linked, if not linked, the symbol becomes null instead of
612 being an undefined reference.</dd>
614 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
615 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
616 <dd>Some languages allow differing globals to be merged, such as two functions
617 with different semantics. Other languages, such as <tt>C++</tt>, ensure
618 that only equivalent globals are ever merged (the "one definition rule" -
619 "ODR"). Such languages can use the <tt>linkonce_odr</tt>
620 and <tt>weak_odr</tt> linkage types to indicate that the global will only
621 be merged with equivalent globals. These linkage types are otherwise the
622 same as their non-<tt>odr</tt> versions.</dd>
624 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
625 <dd>If none of the above identifiers are used, the global is externally
626 visible, meaning that it participates in linkage and can be used to
627 resolve external symbol references.</dd>
630 <p>The next two types of linkage are targeted for Microsoft Windows platform
631 only. They are designed to support importing (exporting) symbols from (to)
632 DLLs (Dynamic Link Libraries).</p>
635 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
636 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
637 or variable via a global pointer to a pointer that is set up by the DLL
638 exporting the symbol. On Microsoft Windows targets, the pointer name is
639 formed by combining <code>__imp_</code> and the function or variable
642 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
643 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
644 pointer to a pointer in a DLL, so that it can be referenced with the
645 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
646 name is formed by combining <code>__imp_</code> and the function or
650 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
651 another module defined a "<tt>.LC0</tt>" variable and was linked with this
652 one, one of the two would be renamed, preventing a collision. Since
653 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
654 declarations), they are accessible outside of the current module.</p>
656 <p>It is illegal for a function <i>declaration</i> to have any linkage type
657 other than "externally visible", <tt>dllimport</tt>
658 or <tt>extern_weak</tt>.</p>
660 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
661 or <tt>weak_odr</tt> linkages.</p>
665 <!-- ======================================================================= -->
666 <div class="doc_subsection">
667 <a name="callingconv">Calling Conventions</a>
670 <div class="doc_text">
672 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
673 and <a href="#i_invoke">invokes</a> can all have an optional calling
674 convention specified for the call. The calling convention of any pair of
675 dynamic caller/callee must match, or the behavior of the program is
676 undefined. The following calling conventions are supported by LLVM, and more
677 may be added in the future:</p>
680 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
681 <dd>This calling convention (the default if no other calling convention is
682 specified) matches the target C calling conventions. This calling
683 convention supports varargs function calls and tolerates some mismatch in
684 the declared prototype and implemented declaration of the function (as
687 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
688 <dd>This calling convention attempts to make calls as fast as possible
689 (e.g. by passing things in registers). This calling convention allows the
690 target to use whatever tricks it wants to produce fast code for the
691 target, without having to conform to an externally specified ABI
692 (Application Binary Interface).
693 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
694 when this convention is used.</a> This calling convention does not
695 support varargs and requires the prototype of all callees to exactly match
696 the prototype of the function definition.</dd>
698 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
699 <dd>This calling convention attempts to make code in the caller as efficient
700 as possible under the assumption that the call is not commonly executed.
701 As such, these calls often preserve all registers so that the call does
702 not break any live ranges in the caller side. This calling convention
703 does not support varargs and requires the prototype of all callees to
704 exactly match the prototype of the function definition.</dd>
706 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
707 <dd>Any calling convention may be specified by number, allowing
708 target-specific calling conventions to be used. Target specific calling
709 conventions start at 64.</dd>
712 <p>More calling conventions can be added/defined on an as-needed basis, to
713 support Pascal conventions or any other well-known target-independent
718 <!-- ======================================================================= -->
719 <div class="doc_subsection">
720 <a name="visibility">Visibility Styles</a>
723 <div class="doc_text">
725 <p>All Global Variables and Functions have one of the following visibility
729 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
730 <dd>On targets that use the ELF object file format, default visibility means
731 that the declaration is visible to other modules and, in shared libraries,
732 means that the declared entity may be overridden. On Darwin, default
733 visibility means that the declaration is visible to other modules. Default
734 visibility corresponds to "external linkage" in the language.</dd>
736 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
737 <dd>Two declarations of an object with hidden visibility refer to the same
738 object if they are in the same shared object. Usually, hidden visibility
739 indicates that the symbol will not be placed into the dynamic symbol
740 table, so no other module (executable or shared library) can reference it
743 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
744 <dd>On ELF, protected visibility indicates that the symbol will be placed in
745 the dynamic symbol table, but that references within the defining module
746 will bind to the local symbol. That is, the symbol cannot be overridden by
752 <!-- ======================================================================= -->
753 <div class="doc_subsection">
754 <a name="namedtypes">Named Types</a>
757 <div class="doc_text">
759 <p>LLVM IR allows you to specify name aliases for certain types. This can make
760 it easier to read the IR and make the IR more condensed (particularly when
761 recursive types are involved). An example of a name specification is:</p>
763 <div class="doc_code">
765 %mytype = type { %mytype*, i32 }
769 <p>You may give a name to any <a href="#typesystem">type</a> except
770 "<a href="t_void">void</a>". Type name aliases may be used anywhere a type
771 is expected with the syntax "%mytype".</p>
773 <p>Note that type names are aliases for the structural type that they indicate,
774 and that you can therefore specify multiple names for the same type. This
775 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
776 uses structural typing, the name is not part of the type. When printing out
777 LLVM IR, the printer will pick <em>one name</em> to render all types of a
778 particular shape. This means that if you have code where two different
779 source types end up having the same LLVM type, that the dumper will sometimes
780 print the "wrong" or unexpected type. This is an important design point and
781 isn't going to change.</p>
785 <!-- ======================================================================= -->
786 <div class="doc_subsection">
787 <a name="globalvars">Global Variables</a>
790 <div class="doc_text">
792 <p>Global variables define regions of memory allocated at compilation time
793 instead of run-time. Global variables may optionally be initialized, may
794 have an explicit section to be placed in, and may have an optional explicit
795 alignment specified. A variable may be defined as "thread_local", which
796 means that it will not be shared by threads (each thread will have a
797 separated copy of the variable). A variable may be defined as a global
798 "constant," which indicates that the contents of the variable
799 will <b>never</b> be modified (enabling better optimization, allowing the
800 global data to be placed in the read-only section of an executable, etc).
801 Note that variables that need runtime initialization cannot be marked
802 "constant" as there is a store to the variable.</p>
804 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
805 constant, even if the final definition of the global is not. This capability
806 can be used to enable slightly better optimization of the program, but
807 requires the language definition to guarantee that optimizations based on the
808 'constantness' are valid for the translation units that do not include the
811 <p>As SSA values, global variables define pointer values that are in scope
812 (i.e. they dominate) all basic blocks in the program. Global variables
813 always define a pointer to their "content" type because they describe a
814 region of memory, and all memory objects in LLVM are accessed through
817 <p>A global variable may be declared to reside in a target-specific numbered
818 address space. For targets that support them, address spaces may affect how
819 optimizations are performed and/or what target instructions are used to
820 access the variable. The default address space is zero. The address space
821 qualifier must precede any other attributes.</p>
823 <p>LLVM allows an explicit section to be specified for globals. If the target
824 supports it, it will emit globals to the section specified.</p>
826 <p>An explicit alignment may be specified for a global. If not present, or if
827 the alignment is set to zero, the alignment of the global is set by the
828 target to whatever it feels convenient. If an explicit alignment is
829 specified, the global is forced to have at least that much alignment. All
830 alignments must be a power of 2.</p>
832 <p>For example, the following defines a global in a numbered address space with
833 an initializer, section, and alignment:</p>
835 <div class="doc_code">
837 @G = addrspace(5) constant float 1.0, section "foo", align 4
844 <!-- ======================================================================= -->
845 <div class="doc_subsection">
846 <a name="functionstructure">Functions</a>
849 <div class="doc_text">
851 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord, an
852 optional <a href="#linkage">linkage type</a>, an optional
853 <a href="#visibility">visibility style</a>, an optional
854 <a href="#callingconv">calling convention</a>, a return type, an optional
855 <a href="#paramattrs">parameter attribute</a> for the return type, a function
856 name, a (possibly empty) argument list (each with optional
857 <a href="#paramattrs">parameter attributes</a>), optional
858 <a href="#fnattrs">function attributes</a>, an optional section, an optional
859 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
860 curly brace, a list of basic blocks, and a closing curly brace.</p>
862 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
863 optional <a href="#linkage">linkage type</a>, an optional
864 <a href="#visibility">visibility style</a>, an optional
865 <a href="#callingconv">calling convention</a>, a return type, an optional
866 <a href="#paramattrs">parameter attribute</a> for the return type, a function
867 name, a possibly empty list of arguments, an optional alignment, and an
868 optional <a href="#gc">garbage collector name</a>.</p>
870 <p>A function definition contains a list of basic blocks, forming the CFG
871 (Control Flow Graph) for the function. Each basic block may optionally start
872 with a label (giving the basic block a symbol table entry), contains a list
873 of instructions, and ends with a <a href="#terminators">terminator</a>
874 instruction (such as a branch or function return).</p>
876 <p>The first basic block in a function is special in two ways: it is immediately
877 executed on entrance to the function, and it is not allowed to have
878 predecessor basic blocks (i.e. there can not be any branches to the entry
879 block of a function). Because the block can have no predecessors, it also
880 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
882 <p>LLVM allows an explicit section to be specified for functions. If the target
883 supports it, it will emit functions to the section specified.</p>
885 <p>An explicit alignment may be specified for a function. If not present, or if
886 the alignment is set to zero, the alignment of the function is set by the
887 target to whatever it feels convenient. If an explicit alignment is
888 specified, the function is forced to have at least that much alignment. All
889 alignments must be a power of 2.</p>
892 <div class="doc_code">
894 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
895 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
896 <ResultType> @<FunctionName> ([argument list])
897 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
898 [<a href="#gc">gc</a>] { ... }
904 <!-- ======================================================================= -->
905 <div class="doc_subsection">
906 <a name="aliasstructure">Aliases</a>
909 <div class="doc_text">
911 <p>Aliases act as "second name" for the aliasee value (which can be either
912 function, global variable, another alias or bitcast of global value). Aliases
913 may have an optional <a href="#linkage">linkage type</a>, and an
914 optional <a href="#visibility">visibility style</a>.</p>
917 <div class="doc_code">
919 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
925 <!-- ======================================================================= -->
926 <div class="doc_subsection">
927 <a name="namedmetadatastructure">Named Metadata</a>
930 <div class="doc_text">
932 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
933 nodes</a> (but not metadata strings) and null are the only valid operands for
934 a named metadata.</p>
937 <div class="doc_code">
939 !1 = metadata !{metadata !"one"}
946 <!-- ======================================================================= -->
947 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
949 <div class="doc_text">
951 <p>The return type and each parameter of a function type may have a set of
952 <i>parameter attributes</i> associated with them. Parameter attributes are
953 used to communicate additional information about the result or parameters of
954 a function. Parameter attributes are considered to be part of the function,
955 not of the function type, so functions with different parameter attributes
956 can have the same function type.</p>
958 <p>Parameter attributes are simple keywords that follow the type specified. If
959 multiple parameter attributes are needed, they are space separated. For
962 <div class="doc_code">
964 declare i32 @printf(i8* noalias nocapture, ...)
965 declare i32 @atoi(i8 zeroext)
966 declare signext i8 @returns_signed_char()
970 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
971 <tt>readonly</tt>) come immediately after the argument list.</p>
973 <p>Currently, only the following parameter attributes are defined:</p>
976 <dt><tt><b>zeroext</b></tt></dt>
977 <dd>This indicates to the code generator that the parameter or return value
978 should be zero-extended to a 32-bit value by the caller (for a parameter)
979 or the callee (for a return value).</dd>
981 <dt><tt><b>signext</b></tt></dt>
982 <dd>This indicates to the code generator that the parameter or return value
983 should be sign-extended to a 32-bit value by the caller (for a parameter)
984 or the callee (for a return value).</dd>
986 <dt><tt><b>inreg</b></tt></dt>
987 <dd>This indicates that this parameter or return value should be treated in a
988 special target-dependent fashion during while emitting code for a function
989 call or return (usually, by putting it in a register as opposed to memory,
990 though some targets use it to distinguish between two different kinds of
991 registers). Use of this attribute is target-specific.</dd>
993 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
994 <dd>This indicates that the pointer parameter should really be passed by value
995 to the function. The attribute implies that a hidden copy of the pointee
996 is made between the caller and the callee, so the callee is unable to
997 modify the value in the callee. This attribute is only valid on LLVM
998 pointer arguments. It is generally used to pass structs and arrays by
999 value, but is also valid on pointers to scalars. The copy is considered
1000 to belong to the caller not the callee (for example,
1001 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1002 <tt>byval</tt> parameters). This is not a valid attribute for return
1003 values. The byval attribute also supports specifying an alignment with
1004 the align attribute. This has a target-specific effect on the code
1005 generator that usually indicates a desired alignment for the synthesized
1008 <dt><tt><b>sret</b></tt></dt>
1009 <dd>This indicates that the pointer parameter specifies the address of a
1010 structure that is the return value of the function in the source program.
1011 This pointer must be guaranteed by the caller to be valid: loads and
1012 stores to the structure may be assumed by the callee to not to trap. This
1013 may only be applied to the first parameter. This is not a valid attribute
1014 for return values. </dd>
1016 <dt><tt><b>noalias</b></tt></dt>
1017 <dd>This indicates that the pointer does not alias any global or any other
1018 parameter. The caller is responsible for ensuring that this is the
1019 case. On a function return value, <tt>noalias</tt> additionally indicates
1020 that the pointer does not alias any other pointers visible to the
1021 caller. For further details, please see the discussion of the NoAlias
1023 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
1026 <dt><tt><b>nocapture</b></tt></dt>
1027 <dd>This indicates that the callee does not make any copies of the pointer
1028 that outlive the callee itself. This is not a valid attribute for return
1031 <dt><tt><b>nest</b></tt></dt>
1032 <dd>This indicates that the pointer parameter can be excised using the
1033 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1034 attribute for return values.</dd>
1039 <!-- ======================================================================= -->
1040 <div class="doc_subsection">
1041 <a name="gc">Garbage Collector Names</a>
1044 <div class="doc_text">
1046 <p>Each function may specify a garbage collector name, which is simply a
1049 <div class="doc_code">
1051 define void @f() gc "name" { ... }
1055 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1056 collector which will cause the compiler to alter its output in order to
1057 support the named garbage collection algorithm.</p>
1061 <!-- ======================================================================= -->
1062 <div class="doc_subsection">
1063 <a name="fnattrs">Function Attributes</a>
1066 <div class="doc_text">
1068 <p>Function attributes are set to communicate additional information about a
1069 function. Function attributes are considered to be part of the function, not
1070 of the function type, so functions with different parameter attributes can
1071 have the same function type.</p>
1073 <p>Function attributes are simple keywords that follow the type specified. If
1074 multiple attributes are needed, they are space separated. For example:</p>
1076 <div class="doc_code">
1078 define void @f() noinline { ... }
1079 define void @f() alwaysinline { ... }
1080 define void @f() alwaysinline optsize { ... }
1081 define void @f() optsize { ... }
1086 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1087 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1088 the backend should forcibly align the stack pointer. Specify the
1089 desired alignment, which must be a power of two, in parentheses.
1091 <dt><tt><b>alwaysinline</b></tt></dt>
1092 <dd>This attribute indicates that the inliner should attempt to inline this
1093 function into callers whenever possible, ignoring any active inlining size
1094 threshold for this caller.</dd>
1096 <dt><tt><b>inlinehint</b></tt></dt>
1097 <dd>This attribute indicates that the source code contained a hint that inlining
1098 this function is desirable (such as the "inline" keyword in C/C++). It
1099 is just a hint; it imposes no requirements on the inliner.</dd>
1101 <dt><tt><b>noinline</b></tt></dt>
1102 <dd>This attribute indicates that the inliner should never inline this
1103 function in any situation. This attribute may not be used together with
1104 the <tt>alwaysinline</tt> attribute.</dd>
1106 <dt><tt><b>optsize</b></tt></dt>
1107 <dd>This attribute suggests that optimization passes and code generator passes
1108 make choices that keep the code size of this function low, and otherwise
1109 do optimizations specifically to reduce code size.</dd>
1111 <dt><tt><b>noreturn</b></tt></dt>
1112 <dd>This function attribute indicates that the function never returns
1113 normally. This produces undefined behavior at runtime if the function
1114 ever does dynamically return.</dd>
1116 <dt><tt><b>nounwind</b></tt></dt>
1117 <dd>This function attribute indicates that the function never returns with an
1118 unwind or exceptional control flow. If the function does unwind, its
1119 runtime behavior is undefined.</dd>
1121 <dt><tt><b>readnone</b></tt></dt>
1122 <dd>This attribute indicates that the function computes its result (or decides
1123 to unwind an exception) based strictly on its arguments, without
1124 dereferencing any pointer arguments or otherwise accessing any mutable
1125 state (e.g. memory, control registers, etc) visible to caller functions.
1126 It does not write through any pointer arguments
1127 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1128 changes any state visible to callers. This means that it cannot unwind
1129 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1130 could use the <tt>unwind</tt> instruction.</dd>
1132 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1133 <dd>This attribute indicates that the function does not write through any
1134 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1135 arguments) or otherwise modify any state (e.g. memory, control registers,
1136 etc) visible to caller functions. It may dereference pointer arguments
1137 and read state that may be set in the caller. A readonly function always
1138 returns the same value (or unwinds an exception identically) when called
1139 with the same set of arguments and global state. It cannot unwind an
1140 exception by calling the <tt>C++</tt> exception throwing methods, but may
1141 use the <tt>unwind</tt> instruction.</dd>
1143 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1144 <dd>This attribute indicates that the function should emit a stack smashing
1145 protector. It is in the form of a "canary"—a random value placed on
1146 the stack before the local variables that's checked upon return from the
1147 function to see if it has been overwritten. A heuristic is used to
1148 determine if a function needs stack protectors or not.<br>
1150 If a function that has an <tt>ssp</tt> attribute is inlined into a
1151 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1152 function will have an <tt>ssp</tt> attribute.</dd>
1154 <dt><tt><b>sspreq</b></tt></dt>
1155 <dd>This attribute indicates that the function should <em>always</em> emit a
1156 stack smashing protector. This overrides
1157 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1159 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1160 function that doesn't have an <tt>sspreq</tt> attribute or which has
1161 an <tt>ssp</tt> attribute, then the resulting function will have
1162 an <tt>sspreq</tt> attribute.</dd>
1164 <dt><tt><b>noredzone</b></tt></dt>
1165 <dd>This attribute indicates that the code generator should not use a red
1166 zone, even if the target-specific ABI normally permits it.</dd>
1168 <dt><tt><b>noimplicitfloat</b></tt></dt>
1169 <dd>This attributes disables implicit floating point instructions.</dd>
1171 <dt><tt><b>naked</b></tt></dt>
1172 <dd>This attribute disables prologue / epilogue emission for the function.
1173 This can have very system-specific consequences.</dd>
1178 <!-- ======================================================================= -->
1179 <div class="doc_subsection">
1180 <a name="moduleasm">Module-Level Inline Assembly</a>
1183 <div class="doc_text">
1185 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1186 the GCC "file scope inline asm" blocks. These blocks are internally
1187 concatenated by LLVM and treated as a single unit, but may be separated in
1188 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1190 <div class="doc_code">
1192 module asm "inline asm code goes here"
1193 module asm "more can go here"
1197 <p>The strings can contain any character by escaping non-printable characters.
1198 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1201 <p>The inline asm code is simply printed to the machine code .s file when
1202 assembly code is generated.</p>
1206 <!-- ======================================================================= -->
1207 <div class="doc_subsection">
1208 <a name="datalayout">Data Layout</a>
1211 <div class="doc_text">
1213 <p>A module may specify a target specific data layout string that specifies how
1214 data is to be laid out in memory. The syntax for the data layout is
1217 <div class="doc_code">
1219 target datalayout = "<i>layout specification</i>"
1223 <p>The <i>layout specification</i> consists of a list of specifications
1224 separated by the minus sign character ('-'). Each specification starts with
1225 a letter and may include other information after the letter to define some
1226 aspect of the data layout. The specifications accepted are as follows:</p>
1230 <dd>Specifies that the target lays out data in big-endian form. That is, the
1231 bits with the most significance have the lowest address location.</dd>
1234 <dd>Specifies that the target lays out data in little-endian form. That is,
1235 the bits with the least significance have the lowest address
1238 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1239 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1240 <i>preferred</i> alignments. All sizes are in bits. Specifying
1241 the <i>pref</i> alignment is optional. If omitted, the
1242 preceding <tt>:</tt> should be omitted too.</dd>
1244 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1245 <dd>This specifies the alignment for an integer type of a given bit
1246 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1248 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1249 <dd>This specifies the alignment for a vector type of a given bit
1252 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1253 <dd>This specifies the alignment for a floating point type of a given bit
1254 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1257 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1258 <dd>This specifies the alignment for an aggregate type of a given bit
1261 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1262 <dd>This specifies the alignment for a stack object of a given bit
1265 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1266 <dd>This specifies a set of native integer widths for the target CPU
1267 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1268 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1269 this set are considered to support most general arithmetic
1270 operations efficiently.</dd>
1273 <p>When constructing the data layout for a given target, LLVM starts with a
1274 default set of specifications which are then (possibly) overriden by the
1275 specifications in the <tt>datalayout</tt> keyword. The default specifications
1276 are given in this list:</p>
1279 <li><tt>E</tt> - big endian</li>
1280 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1281 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1282 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1283 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1284 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1285 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1286 alignment of 64-bits</li>
1287 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1288 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1289 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1290 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1291 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1292 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1295 <p>When LLVM is determining the alignment for a given type, it uses the
1296 following rules:</p>
1299 <li>If the type sought is an exact match for one of the specifications, that
1300 specification is used.</li>
1302 <li>If no match is found, and the type sought is an integer type, then the
1303 smallest integer type that is larger than the bitwidth of the sought type
1304 is used. If none of the specifications are larger than the bitwidth then
1305 the the largest integer type is used. For example, given the default
1306 specifications above, the i7 type will use the alignment of i8 (next
1307 largest) while both i65 and i256 will use the alignment of i64 (largest
1310 <li>If no match is found, and the type sought is a vector type, then the
1311 largest vector type that is smaller than the sought vector type will be
1312 used as a fall back. This happens because <128 x double> can be
1313 implemented in terms of 64 <2 x double>, for example.</li>
1318 <!-- ======================================================================= -->
1319 <div class="doc_subsection">
1320 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1323 <div class="doc_text">
1325 <p>Any memory access must be done through a pointer value associated
1326 with an address range of the memory access, otherwise the behavior
1327 is undefined. Pointer values are associated with address ranges
1328 according to the following rules:</p>
1331 <li>A pointer value formed from a
1332 <tt><a href="#i_getelementptr">getelementptr</a></tt> instruction
1333 is associated with the addresses associated with the first operand
1334 of the <tt>getelementptr</tt>.</li>
1335 <li>An address of a global variable is associated with the address
1336 range of the variable's storage.</li>
1337 <li>The result value of an allocation instruction is associated with
1338 the address range of the allocated storage.</li>
1339 <li>A null pointer in the default address-space is associated with
1341 <li>A pointer value formed by an
1342 <tt><a href="#i_inttoptr">inttoptr</a></tt> is associated with all
1343 address ranges of all pointer values that contribute (directly or
1344 indirectly) to the computation of the pointer's value.</li>
1345 <li>The result value of a
1346 <tt><a href="#i_bitcast">bitcast</a></tt> is associated with all
1347 addresses associated with the operand of the <tt>bitcast</tt>.</li>
1348 <li>An integer constant other than zero or a pointer value returned
1349 from a function not defined within LLVM may be associated with address
1350 ranges allocated through mechanisms other than those provided by
1351 LLVM. Such ranges shall not overlap with any ranges of addresses
1352 allocated by mechanisms provided by LLVM.</li>
1355 <p>LLVM IR does not associate types with memory. The result type of a
1356 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1357 alignment of the memory from which to load, as well as the
1358 interpretation of the value. The first operand of a
1359 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1360 and alignment of the store.</p>
1362 <p>Consequently, type-based alias analysis, aka TBAA, aka
1363 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1364 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1365 additional information which specialized optimization passes may use
1366 to implement type-based alias analysis.</p>
1370 <!-- *********************************************************************** -->
1371 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1372 <!-- *********************************************************************** -->
1374 <div class="doc_text">
1376 <p>The LLVM type system is one of the most important features of the
1377 intermediate representation. Being typed enables a number of optimizations
1378 to be performed on the intermediate representation directly, without having
1379 to do extra analyses on the side before the transformation. A strong type
1380 system makes it easier to read the generated code and enables novel analyses
1381 and transformations that are not feasible to perform on normal three address
1382 code representations.</p>
1386 <!-- ======================================================================= -->
1387 <div class="doc_subsection"> <a name="t_classifications">Type
1388 Classifications</a> </div>
1390 <div class="doc_text">
1392 <p>The types fall into a few useful classifications:</p>
1394 <table border="1" cellspacing="0" cellpadding="4">
1396 <tr><th>Classification</th><th>Types</th></tr>
1398 <td><a href="#t_integer">integer</a></td>
1399 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1402 <td><a href="#t_floating">floating point</a></td>
1403 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1406 <td><a name="t_firstclass">first class</a></td>
1407 <td><a href="#t_integer">integer</a>,
1408 <a href="#t_floating">floating point</a>,
1409 <a href="#t_pointer">pointer</a>,
1410 <a href="#t_vector">vector</a>,
1411 <a href="#t_struct">structure</a>,
1412 <a href="#t_union">union</a>,
1413 <a href="#t_array">array</a>,
1414 <a href="#t_label">label</a>,
1415 <a href="#t_metadata">metadata</a>.
1419 <td><a href="#t_primitive">primitive</a></td>
1420 <td><a href="#t_label">label</a>,
1421 <a href="#t_void">void</a>,
1422 <a href="#t_floating">floating point</a>,
1423 <a href="#t_metadata">metadata</a>.</td>
1426 <td><a href="#t_derived">derived</a></td>
1427 <td><a href="#t_array">array</a>,
1428 <a href="#t_function">function</a>,
1429 <a href="#t_pointer">pointer</a>,
1430 <a href="#t_struct">structure</a>,
1431 <a href="#t_pstruct">packed structure</a>,
1432 <a href="#t_union">union</a>,
1433 <a href="#t_vector">vector</a>,
1434 <a href="#t_opaque">opaque</a>.
1440 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1441 important. Values of these types are the only ones which can be produced by
1446 <!-- ======================================================================= -->
1447 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1449 <div class="doc_text">
1451 <p>The primitive types are the fundamental building blocks of the LLVM
1456 <!-- _______________________________________________________________________ -->
1457 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1459 <div class="doc_text">
1462 <p>The integer type is a very simple type that simply specifies an arbitrary
1463 bit width for the integer type desired. Any bit width from 1 bit to
1464 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1471 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1475 <table class="layout">
1477 <td class="left"><tt>i1</tt></td>
1478 <td class="left">a single-bit integer.</td>
1481 <td class="left"><tt>i32</tt></td>
1482 <td class="left">a 32-bit integer.</td>
1485 <td class="left"><tt>i1942652</tt></td>
1486 <td class="left">a really big integer of over 1 million bits.</td>
1492 <!-- _______________________________________________________________________ -->
1493 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1495 <div class="doc_text">
1499 <tr><th>Type</th><th>Description</th></tr>
1500 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1501 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1502 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1503 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1504 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1510 <!-- _______________________________________________________________________ -->
1511 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1513 <div class="doc_text">
1516 <p>The void type does not represent any value and has no size.</p>
1525 <!-- _______________________________________________________________________ -->
1526 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1528 <div class="doc_text">
1531 <p>The label type represents code labels.</p>
1540 <!-- _______________________________________________________________________ -->
1541 <div class="doc_subsubsection"> <a name="t_metadata">Metadata Type</a> </div>
1543 <div class="doc_text">
1546 <p>The metadata type represents embedded metadata. No derived types may be
1547 created from metadata except for <a href="#t_function">function</a>
1558 <!-- ======================================================================= -->
1559 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1561 <div class="doc_text">
1563 <p>The real power in LLVM comes from the derived types in the system. This is
1564 what allows a programmer to represent arrays, functions, pointers, and other
1565 useful types. Each of these types contain one or more element types which
1566 may be a primitive type, or another derived type. For example, it is
1567 possible to have a two dimensional array, using an array as the element type
1568 of another array.</p>
1573 <!-- _______________________________________________________________________ -->
1574 <div class="doc_subsubsection"> <a name="t_aggregate">Aggregate Types</a> </div>
1576 <div class="doc_text">
1578 <p>Aggregate Types are a subset of derived types that can contain multiple
1579 member types. <a href="#t_array">Arrays</a>,
1580 <a href="#t_struct">structs</a>, <a href="#t_vector">vectors</a> and
1581 <a href="#t_union">unions</a> are aggregate types.</p>
1587 <!-- _______________________________________________________________________ -->
1588 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1590 <div class="doc_text">
1593 <p>The array type is a very simple derived type that arranges elements
1594 sequentially in memory. The array type requires a size (number of elements)
1595 and an underlying data type.</p>
1599 [<# elements> x <elementtype>]
1602 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1603 be any type with a size.</p>
1606 <table class="layout">
1608 <td class="left"><tt>[40 x i32]</tt></td>
1609 <td class="left">Array of 40 32-bit integer values.</td>
1612 <td class="left"><tt>[41 x i32]</tt></td>
1613 <td class="left">Array of 41 32-bit integer values.</td>
1616 <td class="left"><tt>[4 x i8]</tt></td>
1617 <td class="left">Array of 4 8-bit integer values.</td>
1620 <p>Here are some examples of multidimensional arrays:</p>
1621 <table class="layout">
1623 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1624 <td class="left">3x4 array of 32-bit integer values.</td>
1627 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1628 <td class="left">12x10 array of single precision floating point values.</td>
1631 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1632 <td class="left">2x3x4 array of 16-bit integer values.</td>
1636 <p>There is no restriction on indexing beyond the end of the array implied by
1637 a static type (though there are restrictions on indexing beyond the bounds
1638 of an allocated object in some cases). This means that single-dimension
1639 'variable sized array' addressing can be implemented in LLVM with a zero
1640 length array type. An implementation of 'pascal style arrays' in LLVM could
1641 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1645 <!-- _______________________________________________________________________ -->
1646 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1648 <div class="doc_text">
1651 <p>The function type can be thought of as a function signature. It consists of
1652 a return type and a list of formal parameter types. The return type of a
1653 function type is a scalar type, a void type, a struct type, or a union
1654 type. If the return type is a struct type then all struct elements must be
1655 of first class types, and the struct must have at least one element.</p>
1659 <returntype> (<parameter list>)
1662 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1663 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1664 which indicates that the function takes a variable number of arguments.
1665 Variable argument functions can access their arguments with
1666 the <a href="#int_varargs">variable argument handling intrinsic</a>
1667 functions. '<tt><returntype></tt>' is a any type except
1668 <a href="#t_label">label</a>.</p>
1671 <table class="layout">
1673 <td class="left"><tt>i32 (i32)</tt></td>
1674 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1676 </tr><tr class="layout">
1677 <td class="left"><tt>float (i16 signext, i32 *) *
1679 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1680 an <tt>i16</tt> that should be sign extended and a
1681 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1684 </tr><tr class="layout">
1685 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1686 <td class="left">A vararg function that takes at least one
1687 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1688 which returns an integer. This is the signature for <tt>printf</tt> in
1691 </tr><tr class="layout">
1692 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1693 <td class="left">A function taking an <tt>i32</tt>, returning a
1694 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
1701 <!-- _______________________________________________________________________ -->
1702 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1704 <div class="doc_text">
1707 <p>The structure type is used to represent a collection of data members together
1708 in memory. The packing of the field types is defined to match the ABI of the
1709 underlying processor. The elements of a structure may be any type that has a
1712 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
1713 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
1714 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
1715 Structures in registers are accessed using the
1716 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
1717 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
1720 { <type list> }
1724 <table class="layout">
1726 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1727 <td class="left">A triple of three <tt>i32</tt> values</td>
1728 </tr><tr class="layout">
1729 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1730 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1731 second element is a <a href="#t_pointer">pointer</a> to a
1732 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1733 an <tt>i32</tt>.</td>
1739 <!-- _______________________________________________________________________ -->
1740 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1743 <div class="doc_text">
1746 <p>The packed structure type is used to represent a collection of data members
1747 together in memory. There is no padding between fields. Further, the
1748 alignment of a packed structure is 1 byte. The elements of a packed
1749 structure may be any type that has a size.</p>
1751 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1752 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1753 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1757 < { <type list> } >
1761 <table class="layout">
1763 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1764 <td class="left">A triple of three <tt>i32</tt> values</td>
1765 </tr><tr class="layout">
1767 <tt>< { float, i32 (i32)* } ></tt></td>
1768 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1769 second element is a <a href="#t_pointer">pointer</a> to a
1770 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1771 an <tt>i32</tt>.</td>
1777 <!-- _______________________________________________________________________ -->
1778 <div class="doc_subsubsection"> <a name="t_union">Union Type</a> </div>
1780 <div class="doc_text">
1783 <p>A union type describes an object with size and alignment suitable for
1784 an object of any one of a given set of types (also known as an "untagged"
1785 union). It is similar in concept and usage to a
1786 <a href="#t_struct">struct</a>, except that all members of the union
1787 have an offset of zero. The elements of a union may be any type that has a
1788 size. Unions must have at least one member - empty unions are not allowed.
1791 <p>The size of the union as a whole will be the size of its largest member,
1792 and the alignment requirements of the union as a whole will be the largest
1793 alignment requirement of any member.</p>
1795 <p>Union members are accessed using '<tt><a href="#i_load">load</a></tt> and
1796 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1797 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
1798 Since all members are at offset zero, the getelementptr instruction does
1799 not affect the address, only the type of the resulting pointer.</p>
1803 union { <type list> }
1807 <table class="layout">
1809 <td class="left"><tt>union { i32, i32*, float }</tt></td>
1810 <td class="left">A union of three types: an <tt>i32</tt>, a pointer to
1811 an <tt>i32</tt>, and a <tt>float</tt>.</td>
1812 </tr><tr class="layout">
1814 <tt>union { float, i32 (i32) * }</tt></td>
1815 <td class="left">A union, where the first element is a <tt>float</tt> and the
1816 second element is a <a href="#t_pointer">pointer</a> to a
1817 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1818 an <tt>i32</tt>.</td>
1824 <!-- _______________________________________________________________________ -->
1825 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1827 <div class="doc_text">
1830 <p>The pointer type is used to specify memory locations.
1831 Pointers are commonly used to reference objects in memory.</p>
1833 <p>Pointer types may have an optional address space attribute defining the
1834 numbered address space where the pointed-to object resides. The default
1835 address space is number zero. The semantics of non-zero address
1836 spaces are target-specific.</p>
1838 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
1839 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1847 <table class="layout">
1849 <td class="left"><tt>[4 x i32]*</tt></td>
1850 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1851 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1854 <td class="left"><tt>i32 (i32 *) *</tt></td>
1855 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1856 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1860 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1861 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1862 that resides in address space #5.</td>
1868 <!-- _______________________________________________________________________ -->
1869 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1871 <div class="doc_text">
1874 <p>A vector type is a simple derived type that represents a vector of elements.
1875 Vector types are used when multiple primitive data are operated in parallel
1876 using a single instruction (SIMD). A vector type requires a size (number of
1877 elements) and an underlying primitive data type. Vector types are considered
1878 <a href="#t_firstclass">first class</a>.</p>
1882 < <# elements> x <elementtype> >
1885 <p>The number of elements is a constant integer value; elementtype may be any
1886 integer or floating point type.</p>
1889 <table class="layout">
1891 <td class="left"><tt><4 x i32></tt></td>
1892 <td class="left">Vector of 4 32-bit integer values.</td>
1895 <td class="left"><tt><8 x float></tt></td>
1896 <td class="left">Vector of 8 32-bit floating-point values.</td>
1899 <td class="left"><tt><2 x i64></tt></td>
1900 <td class="left">Vector of 2 64-bit integer values.</td>
1906 <!-- _______________________________________________________________________ -->
1907 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1908 <div class="doc_text">
1911 <p>Opaque types are used to represent unknown types in the system. This
1912 corresponds (for example) to the C notion of a forward declared structure
1913 type. In LLVM, opaque types can eventually be resolved to any type (not just
1914 a structure type).</p>
1922 <table class="layout">
1924 <td class="left"><tt>opaque</tt></td>
1925 <td class="left">An opaque type.</td>
1931 <!-- ======================================================================= -->
1932 <div class="doc_subsection">
1933 <a name="t_uprefs">Type Up-references</a>
1936 <div class="doc_text">
1939 <p>An "up reference" allows you to refer to a lexically enclosing type without
1940 requiring it to have a name. For instance, a structure declaration may
1941 contain a pointer to any of the types it is lexically a member of. Example
1942 of up references (with their equivalent as named type declarations)
1946 { \2 * } %x = type { %x* }
1947 { \2 }* %y = type { %y }*
1951 <p>An up reference is needed by the asmprinter for printing out cyclic types
1952 when there is no declared name for a type in the cycle. Because the
1953 asmprinter does not want to print out an infinite type string, it needs a
1954 syntax to handle recursive types that have no names (all names are optional
1962 <p>The level is the count of the lexical type that is being referred to.</p>
1965 <table class="layout">
1967 <td class="left"><tt>\1*</tt></td>
1968 <td class="left">Self-referential pointer.</td>
1971 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1972 <td class="left">Recursive structure where the upref refers to the out-most
1979 <!-- *********************************************************************** -->
1980 <div class="doc_section"> <a name="constants">Constants</a> </div>
1981 <!-- *********************************************************************** -->
1983 <div class="doc_text">
1985 <p>LLVM has several different basic types of constants. This section describes
1986 them all and their syntax.</p>
1990 <!-- ======================================================================= -->
1991 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1993 <div class="doc_text">
1996 <dt><b>Boolean constants</b></dt>
1997 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1998 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2000 <dt><b>Integer constants</b></dt>
2001 <dd>Standard integers (such as '4') are constants of
2002 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2003 with integer types.</dd>
2005 <dt><b>Floating point constants</b></dt>
2006 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2007 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2008 notation (see below). The assembler requires the exact decimal value of a
2009 floating-point constant. For example, the assembler accepts 1.25 but
2010 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2011 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2013 <dt><b>Null pointer constants</b></dt>
2014 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2015 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2018 <p>The one non-intuitive notation for constants is the hexadecimal form of
2019 floating point constants. For example, the form '<tt>double
2020 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2021 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2022 constants are required (and the only time that they are generated by the
2023 disassembler) is when a floating point constant must be emitted but it cannot
2024 be represented as a decimal floating point number in a reasonable number of
2025 digits. For example, NaN's, infinities, and other special values are
2026 represented in their IEEE hexadecimal format so that assembly and disassembly
2027 do not cause any bits to change in the constants.</p>
2029 <p>When using the hexadecimal form, constants of types float and double are
2030 represented using the 16-digit form shown above (which matches the IEEE754
2031 representation for double); float values must, however, be exactly
2032 representable as IEE754 single precision. Hexadecimal format is always used
2033 for long double, and there are three forms of long double. The 80-bit format
2034 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2035 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2036 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2037 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2038 currently supported target uses this format. Long doubles will only work if
2039 they match the long double format on your target. All hexadecimal formats
2040 are big-endian (sign bit at the left).</p>
2044 <!-- ======================================================================= -->
2045 <div class="doc_subsection">
2046 <a name="aggregateconstants"></a> <!-- old anchor -->
2047 <a name="complexconstants">Complex Constants</a>
2050 <div class="doc_text">
2052 <p>Complex constants are a (potentially recursive) combination of simple
2053 constants and smaller complex constants.</p>
2056 <dt><b>Structure constants</b></dt>
2057 <dd>Structure constants are represented with notation similar to structure
2058 type definitions (a comma separated list of elements, surrounded by braces
2059 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2060 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2061 Structure constants must have <a href="#t_struct">structure type</a>, and
2062 the number and types of elements must match those specified by the
2065 <dt><b>Union constants</b></dt>
2066 <dd>Union constants are represented with notation similar to a structure with
2067 a single element - that is, a single typed element surrounded
2068 by braces (<tt>{}</tt>)). For example: "<tt>{ i32 4 }</tt>". The
2069 <a href="#t_union">union type</a> can be initialized with a single-element
2070 struct as long as the type of the struct element matches the type of
2071 one of the union members.</dd>
2073 <dt><b>Array constants</b></dt>
2074 <dd>Array constants are represented with notation similar to array type
2075 definitions (a comma separated list of elements, surrounded by square
2076 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2077 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2078 the number and types of elements must match those specified by the
2081 <dt><b>Vector constants</b></dt>
2082 <dd>Vector constants are represented with notation similar to vector type
2083 definitions (a comma separated list of elements, surrounded by
2084 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2085 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2086 have <a href="#t_vector">vector type</a>, and the number and types of
2087 elements must match those specified by the type.</dd>
2089 <dt><b>Zero initialization</b></dt>
2090 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2091 value to zero of <em>any</em> type, including scalar and
2092 <a href="#t_aggregate">aggregate</a> types.
2093 This is often used to avoid having to print large zero initializers
2094 (e.g. for large arrays) and is always exactly equivalent to using explicit
2095 zero initializers.</dd>
2097 <dt><b>Metadata node</b></dt>
2098 <dd>A metadata node is a structure-like constant with
2099 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2100 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2101 be interpreted as part of the instruction stream, metadata is a place to
2102 attach additional information such as debug info.</dd>
2107 <!-- ======================================================================= -->
2108 <div class="doc_subsection">
2109 <a name="globalconstants">Global Variable and Function Addresses</a>
2112 <div class="doc_text">
2114 <p>The addresses of <a href="#globalvars">global variables</a>
2115 and <a href="#functionstructure">functions</a> are always implicitly valid
2116 (link-time) constants. These constants are explicitly referenced when
2117 the <a href="#identifiers">identifier for the global</a> is used and always
2118 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2119 legal LLVM file:</p>
2121 <div class="doc_code">
2125 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2131 <!-- ======================================================================= -->
2132 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
2133 <div class="doc_text">
2135 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2136 indicates that the user of the value may receive an unspecified bit-pattern.
2137 Undefined values may be of any type (other than label or void) and be used
2138 anywhere a constant is permitted.</p>
2140 <p>Undefined values are useful because they indicate to the compiler that the
2141 program is well defined no matter what value is used. This gives the
2142 compiler more freedom to optimize. Here are some examples of (potentially
2143 surprising) transformations that are valid (in pseudo IR):</p>
2146 <div class="doc_code">
2158 <p>This is safe because all of the output bits are affected by the undef bits.
2159 Any output bit can have a zero or one depending on the input bits.</p>
2161 <div class="doc_code">
2174 <p>These logical operations have bits that are not always affected by the input.
2175 For example, if "%X" has a zero bit, then the output of the 'and' operation will
2176 always be a zero, no matter what the corresponding bit from the undef is. As
2177 such, it is unsafe to optimize or assume that the result of the and is undef.
2178 However, it is safe to assume that all bits of the undef could be 0, and
2179 optimize the and to 0. Likewise, it is safe to assume that all the bits of
2180 the undef operand to the or could be set, allowing the or to be folded to
2183 <div class="doc_code">
2185 %A = select undef, %X, %Y
2186 %B = select undef, 42, %Y
2187 %C = select %X, %Y, undef
2199 <p>This set of examples show that undefined select (and conditional branch)
2200 conditions can go "either way" but they have to come from one of the two
2201 operands. In the %A example, if %X and %Y were both known to have a clear low
2202 bit, then %A would have to have a cleared low bit. However, in the %C example,
2203 the optimizer is allowed to assume that the undef operand could be the same as
2204 %Y, allowing the whole select to be eliminated.</p>
2207 <div class="doc_code">
2209 %A = xor undef, undef
2228 <p>This example points out that two undef operands are not necessarily the same.
2229 This can be surprising to people (and also matches C semantics) where they
2230 assume that "X^X" is always zero, even if X is undef. This isn't true for a
2231 number of reasons, but the short answer is that an undef "variable" can
2232 arbitrarily change its value over its "live range". This is true because the
2233 "variable" doesn't actually <em>have a live range</em>. Instead, the value is
2234 logically read from arbitrary registers that happen to be around when needed,
2235 so the value is not necessarily consistent over time. In fact, %A and %C need
2236 to have the same semantics or the core LLVM "replace all uses with" concept
2239 <div class="doc_code">
2249 <p>These examples show the crucial difference between an <em>undefined
2250 value</em> and <em>undefined behavior</em>. An undefined value (like undef) is
2251 allowed to have an arbitrary bit-pattern. This means that the %A operation
2252 can be constant folded to undef because the undef could be an SNaN, and fdiv is
2253 not (currently) defined on SNaN's. However, in the second example, we can make
2254 a more aggressive assumption: because the undef is allowed to be an arbitrary
2255 value, we are allowed to assume that it could be zero. Since a divide by zero
2256 has <em>undefined behavior</em>, we are allowed to assume that the operation
2257 does not execute at all. This allows us to delete the divide and all code after
2258 it: since the undefined operation "can't happen", the optimizer can assume that
2259 it occurs in dead code.
2262 <div class="doc_code">
2264 a: store undef -> %X
2265 b: store %X -> undef
2272 <p>These examples reiterate the fdiv example: a store "of" an undefined value
2273 can be assumed to not have any effect: we can assume that the value is
2274 overwritten with bits that happen to match what was already there. However, a
2275 store "to" an undefined location could clobber arbitrary memory, therefore, it
2276 has undefined behavior.</p>
2280 <!-- ======================================================================= -->
2281 <div class="doc_subsection"><a name="blockaddress">Addresses of Basic
2283 <div class="doc_text">
2285 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2287 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2288 basic block in the specified function, and always has an i8* type. Taking
2289 the address of the entry block is illegal.</p>
2291 <p>This value only has defined behavior when used as an operand to the
2292 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction or for comparisons
2293 against null. Pointer equality tests between labels addresses is undefined
2294 behavior - though, again, comparison against null is ok, and no label is
2295 equal to the null pointer. This may also be passed around as an opaque
2296 pointer sized value as long as the bits are not inspected. This allows
2297 <tt>ptrtoint</tt> and arithmetic to be performed on these values so long as
2298 the original value is reconstituted before the <tt>indirectbr</tt>.</p>
2300 <p>Finally, some targets may provide defined semantics when
2301 using the value as the operand to an inline assembly, but that is target
2308 <!-- ======================================================================= -->
2309 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
2312 <div class="doc_text">
2314 <p>Constant expressions are used to allow expressions involving other constants
2315 to be used as constants. Constant expressions may be of
2316 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2317 operation that does not have side effects (e.g. load and call are not
2318 supported). The following is the syntax for constant expressions:</p>
2321 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
2322 <dd>Truncate a constant to another type. The bit size of CST must be larger
2323 than the bit size of TYPE. Both types must be integers.</dd>
2325 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
2326 <dd>Zero extend a constant to another type. The bit size of CST must be
2327 smaller or equal to the bit size of TYPE. Both types must be
2330 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
2331 <dd>Sign extend a constant to another type. The bit size of CST must be
2332 smaller or equal to the bit size of TYPE. Both types must be
2335 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
2336 <dd>Truncate a floating point constant to another floating point type. The
2337 size of CST must be larger than the size of TYPE. Both types must be
2338 floating point.</dd>
2340 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
2341 <dd>Floating point extend a constant to another type. The size of CST must be
2342 smaller or equal to the size of TYPE. Both types must be floating
2345 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
2346 <dd>Convert a floating point constant to the corresponding unsigned integer
2347 constant. TYPE must be a scalar or vector integer type. CST must be of
2348 scalar or vector floating point type. Both CST and TYPE must be scalars,
2349 or vectors of the same number of elements. If the value won't fit in the
2350 integer type, the results are undefined.</dd>
2352 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
2353 <dd>Convert a floating point constant to the corresponding signed integer
2354 constant. TYPE must be a scalar or vector integer type. CST must be of
2355 scalar or vector floating point type. Both CST and TYPE must be scalars,
2356 or vectors of the same number of elements. If the value won't fit in the
2357 integer type, the results are undefined.</dd>
2359 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
2360 <dd>Convert an unsigned integer constant to the corresponding floating point
2361 constant. TYPE must be a scalar or vector floating point type. CST must be
2362 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2363 vectors of the same number of elements. If the value won't fit in the
2364 floating point type, the results are undefined.</dd>
2366 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
2367 <dd>Convert a signed integer constant to the corresponding floating point
2368 constant. TYPE must be a scalar or vector floating point type. CST must be
2369 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2370 vectors of the same number of elements. If the value won't fit in the
2371 floating point type, the results are undefined.</dd>
2373 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
2374 <dd>Convert a pointer typed constant to the corresponding integer constant
2375 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2376 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2377 make it fit in <tt>TYPE</tt>.</dd>
2379 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
2380 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2381 type. CST must be of integer type. The CST value is zero extended,
2382 truncated, or unchanged to make it fit in a pointer size. This one is
2383 <i>really</i> dangerous!</dd>
2385 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
2386 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2387 are the same as those for the <a href="#i_bitcast">bitcast
2388 instruction</a>.</dd>
2390 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2391 <dt><b><tt>getelementptr inbounds ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2392 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2393 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2394 instruction, the index list may have zero or more indexes, which are
2395 required to make sense for the type of "CSTPTR".</dd>
2397 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
2398 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2400 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
2401 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2403 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
2404 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2406 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2407 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2410 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2411 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2414 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2415 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2418 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2419 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2420 be any of the <a href="#binaryops">binary</a>
2421 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2422 on operands are the same as those for the corresponding instruction
2423 (e.g. no bitwise operations on floating point values are allowed).</dd>
2428 <!-- *********************************************************************** -->
2429 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2430 <!-- *********************************************************************** -->
2432 <!-- ======================================================================= -->
2433 <div class="doc_subsection">
2434 <a name="inlineasm">Inline Assembler Expressions</a>
2437 <div class="doc_text">
2439 <p>LLVM supports inline assembler expressions (as opposed
2440 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2441 a special value. This value represents the inline assembler as a string
2442 (containing the instructions to emit), a list of operand constraints (stored
2443 as a string), a flag that indicates whether or not the inline asm
2444 expression has side effects, and a flag indicating whether the function
2445 containing the asm needs to align its stack conservatively. An example
2446 inline assembler expression is:</p>
2448 <div class="doc_code">
2450 i32 (i32) asm "bswap $0", "=r,r"
2454 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2455 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2458 <div class="doc_code">
2460 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2464 <p>Inline asms with side effects not visible in the constraint list must be
2465 marked as having side effects. This is done through the use of the
2466 '<tt>sideeffect</tt>' keyword, like so:</p>
2468 <div class="doc_code">
2470 call void asm sideeffect "eieio", ""()
2474 <p>In some cases inline asms will contain code that will not work unless the
2475 stack is aligned in some way, such as calls or SSE instructions on x86,
2476 yet will not contain code that does that alignment within the asm.
2477 The compiler should make conservative assumptions about what the asm might
2478 contain and should generate its usual stack alignment code in the prologue
2479 if the '<tt>alignstack</tt>' keyword is present:</p>
2481 <div class="doc_code">
2483 call void asm alignstack "eieio", ""()
2487 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2490 <p>TODO: The format of the asm and constraints string still need to be
2491 documented here. Constraints on what can be done (e.g. duplication, moving,
2492 etc need to be documented). This is probably best done by reference to
2493 another document that covers inline asm from a holistic perspective.</p>
2497 <!-- ======================================================================= -->
2498 <div class="doc_subsection"><a name="metadata">Metadata Nodes and Metadata
2502 <div class="doc_text">
2504 <p>LLVM IR allows metadata to be attached to instructions in the program that
2505 can convey extra information about the code to the optimizers and code
2506 generator. One example application of metadata is source-level debug
2507 information. There are two metadata primitives: strings and nodes. All
2508 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2509 preceding exclamation point ('<tt>!</tt>').</p>
2511 <p>A metadata string is a string surrounded by double quotes. It can contain
2512 any character by escaping non-printable characters with "\xx" where "xx" is
2513 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2515 <p>Metadata nodes are represented with notation similar to structure constants
2516 (a comma separated list of elements, surrounded by braces and preceded by an
2517 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2518 10}</tt>". Metadata nodes can have any values as their operand.</p>
2520 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2521 metadata nodes, which can be looked up in the module symbol table. For
2522 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2527 <!-- *********************************************************************** -->
2528 <div class="doc_section">
2529 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2531 <!-- *********************************************************************** -->
2533 <p>LLVM has a number of "magic" global variables that contain data that affect
2534 code generation or other IR semantics. These are documented here. All globals
2535 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2536 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2539 <!-- ======================================================================= -->
2540 <div class="doc_subsection">
2541 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2544 <div class="doc_text">
2546 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2547 href="#linkage_appending">appending linkage</a>. This array contains a list of
2548 pointers to global variables and functions which may optionally have a pointer
2549 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2555 @llvm.used = appending global [2 x i8*] [
2557 i8* bitcast (i32* @Y to i8*)
2558 ], section "llvm.metadata"
2561 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2562 compiler, assembler, and linker are required to treat the symbol as if there is
2563 a reference to the global that it cannot see. For example, if a variable has
2564 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2565 list, it cannot be deleted. This is commonly used to represent references from
2566 inline asms and other things the compiler cannot "see", and corresponds to
2567 "attribute((used))" in GNU C.</p>
2569 <p>On some targets, the code generator must emit a directive to the assembler or
2570 object file to prevent the assembler and linker from molesting the symbol.</p>
2574 <!-- ======================================================================= -->
2575 <div class="doc_subsection">
2576 <a name="intg_compiler_used">The '<tt>llvm.compiler.used</tt>' Global Variable</a>
2579 <div class="doc_text">
2581 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2582 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2583 touching the symbol. On targets that support it, this allows an intelligent
2584 linker to optimize references to the symbol without being impeded as it would be
2585 by <tt>@llvm.used</tt>.</p>
2587 <p>This is a rare construct that should only be used in rare circumstances, and
2588 should not be exposed to source languages.</p>
2592 <!-- ======================================================================= -->
2593 <div class="doc_subsection">
2594 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2597 <div class="doc_text">
2599 <p>TODO: Describe this.</p>
2603 <!-- ======================================================================= -->
2604 <div class="doc_subsection">
2605 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2608 <div class="doc_text">
2610 <p>TODO: Describe this.</p>
2615 <!-- *********************************************************************** -->
2616 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2617 <!-- *********************************************************************** -->
2619 <div class="doc_text">
2621 <p>The LLVM instruction set consists of several different classifications of
2622 instructions: <a href="#terminators">terminator
2623 instructions</a>, <a href="#binaryops">binary instructions</a>,
2624 <a href="#bitwiseops">bitwise binary instructions</a>,
2625 <a href="#memoryops">memory instructions</a>, and
2626 <a href="#otherops">other instructions</a>.</p>
2630 <!-- ======================================================================= -->
2631 <div class="doc_subsection"> <a name="terminators">Terminator
2632 Instructions</a> </div>
2634 <div class="doc_text">
2636 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
2637 in a program ends with a "Terminator" instruction, which indicates which
2638 block should be executed after the current block is finished. These
2639 terminator instructions typically yield a '<tt>void</tt>' value: they produce
2640 control flow, not values (the one exception being the
2641 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2643 <p>There are six different terminator instructions: the
2644 '<a href="#i_ret"><tt>ret</tt></a>' instruction, the
2645 '<a href="#i_br"><tt>br</tt></a>' instruction, the
2646 '<a href="#i_switch"><tt>switch</tt></a>' instruction, the
2647 '<a href="#i_indirectbr">'<tt>indirectbr</tt></a>' Instruction, the
2648 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the
2649 '<a href="#i_unwind"><tt>unwind</tt></a>' instruction, and the
2650 '<a href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2654 <!-- _______________________________________________________________________ -->
2655 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2656 Instruction</a> </div>
2658 <div class="doc_text">
2662 ret <type> <value> <i>; Return a value from a non-void function</i>
2663 ret void <i>; Return from void function</i>
2667 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
2668 a value) from a function back to the caller.</p>
2670 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
2671 value and then causes control flow, and one that just causes control flow to
2675 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
2676 return value. The type of the return value must be a
2677 '<a href="#t_firstclass">first class</a>' type.</p>
2679 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
2680 non-void return type and contains a '<tt>ret</tt>' instruction with no return
2681 value or a return value with a type that does not match its type, or if it
2682 has a void return type and contains a '<tt>ret</tt>' instruction with a
2686 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
2687 the calling function's context. If the caller is a
2688 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
2689 instruction after the call. If the caller was an
2690 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
2691 the beginning of the "normal" destination block. If the instruction returns
2692 a value, that value shall set the call or invoke instruction's return
2697 ret i32 5 <i>; Return an integer value of 5</i>
2698 ret void <i>; Return from a void function</i>
2699 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2703 <!-- _______________________________________________________________________ -->
2704 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2706 <div class="doc_text">
2710 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2714 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
2715 different basic block in the current function. There are two forms of this
2716 instruction, corresponding to a conditional branch and an unconditional
2720 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
2721 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
2722 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
2726 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2727 argument is evaluated. If the value is <tt>true</tt>, control flows to the
2728 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2729 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2734 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
2735 br i1 %cond, label %IfEqual, label %IfUnequal
2737 <a href="#i_ret">ret</a> i32 1
2739 <a href="#i_ret">ret</a> i32 0
2744 <!-- _______________________________________________________________________ -->
2745 <div class="doc_subsubsection">
2746 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2749 <div class="doc_text">
2753 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2757 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2758 several different places. It is a generalization of the '<tt>br</tt>'
2759 instruction, allowing a branch to occur to one of many possible
2763 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2764 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
2765 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
2766 The table is not allowed to contain duplicate constant entries.</p>
2769 <p>The <tt>switch</tt> instruction specifies a table of values and
2770 destinations. When the '<tt>switch</tt>' instruction is executed, this table
2771 is searched for the given value. If the value is found, control flow is
2772 transferred to the corresponding destination; otherwise, control flow is
2773 transferred to the default destination.</p>
2775 <h5>Implementation:</h5>
2776 <p>Depending on properties of the target machine and the particular
2777 <tt>switch</tt> instruction, this instruction may be code generated in
2778 different ways. For example, it could be generated as a series of chained
2779 conditional branches or with a lookup table.</p>
2783 <i>; Emulate a conditional br instruction</i>
2784 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2785 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2787 <i>; Emulate an unconditional br instruction</i>
2788 switch i32 0, label %dest [ ]
2790 <i>; Implement a jump table:</i>
2791 switch i32 %val, label %otherwise [ i32 0, label %onzero
2793 i32 2, label %ontwo ]
2799 <!-- _______________________________________________________________________ -->
2800 <div class="doc_subsubsection">
2801 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
2804 <div class="doc_text">
2808 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
2813 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
2814 within the current function, whose address is specified by
2815 "<tt>address</tt>". Address must be derived from a <a
2816 href="#blockaddress">blockaddress</a> constant.</p>
2820 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
2821 rest of the arguments indicate the full set of possible destinations that the
2822 address may point to. Blocks are allowed to occur multiple times in the
2823 destination list, though this isn't particularly useful.</p>
2825 <p>This destination list is required so that dataflow analysis has an accurate
2826 understanding of the CFG.</p>
2830 <p>Control transfers to the block specified in the address argument. All
2831 possible destination blocks must be listed in the label list, otherwise this
2832 instruction has undefined behavior. This implies that jumps to labels
2833 defined in other functions have undefined behavior as well.</p>
2835 <h5>Implementation:</h5>
2837 <p>This is typically implemented with a jump through a register.</p>
2841 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
2847 <!-- _______________________________________________________________________ -->
2848 <div class="doc_subsubsection">
2849 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2852 <div class="doc_text">
2856 <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>]
2857 to label <normal label> unwind label <exception label>
2861 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2862 function, with the possibility of control flow transfer to either the
2863 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
2864 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
2865 control flow will return to the "normal" label. If the callee (or any
2866 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
2867 instruction, control is interrupted and continued at the dynamically nearest
2868 "exception" label.</p>
2871 <p>This instruction requires several arguments:</p>
2874 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
2875 convention</a> the call should use. If none is specified, the call
2876 defaults to using C calling conventions.</li>
2878 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2879 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
2880 '<tt>inreg</tt>' attributes are valid here.</li>
2882 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2883 function value being invoked. In most cases, this is a direct function
2884 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
2885 off an arbitrary pointer to function value.</li>
2887 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2888 function to be invoked. </li>
2890 <li>'<tt>function args</tt>': argument list whose types match the function
2891 signature argument types. If the function signature indicates the
2892 function accepts a variable number of arguments, the extra arguments can
2895 <li>'<tt>normal label</tt>': the label reached when the called function
2896 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2898 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2899 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2901 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2902 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2903 '<tt>readnone</tt>' attributes are valid here.</li>
2907 <p>This instruction is designed to operate as a standard
2908 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
2909 primary difference is that it establishes an association with a label, which
2910 is used by the runtime library to unwind the stack.</p>
2912 <p>This instruction is used in languages with destructors to ensure that proper
2913 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2914 exception. Additionally, this is important for implementation of
2915 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2917 <p>For the purposes of the SSA form, the definition of the value returned by the
2918 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
2919 block to the "normal" label. If the callee unwinds then no return value is
2922 <p>Note that the code generator does not yet completely support unwind, and
2923 that the invoke/unwind semantics are likely to change in future versions.</p>
2927 %retval = invoke i32 @Test(i32 15) to label %Continue
2928 unwind label %TestCleanup <i>; {i32}:retval set</i>
2929 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2930 unwind label %TestCleanup <i>; {i32}:retval set</i>
2935 <!-- _______________________________________________________________________ -->
2937 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2938 Instruction</a> </div>
2940 <div class="doc_text">
2948 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2949 at the first callee in the dynamic call stack which used
2950 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
2951 This is primarily used to implement exception handling.</p>
2954 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2955 immediately halt. The dynamic call stack is then searched for the
2956 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
2957 Once found, execution continues at the "exceptional" destination block
2958 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
2959 instruction in the dynamic call chain, undefined behavior results.</p>
2961 <p>Note that the code generator does not yet completely support unwind, and
2962 that the invoke/unwind semantics are likely to change in future versions.</p>
2966 <!-- _______________________________________________________________________ -->
2968 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2969 Instruction</a> </div>
2971 <div class="doc_text">
2979 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2980 instruction is used to inform the optimizer that a particular portion of the
2981 code is not reachable. This can be used to indicate that the code after a
2982 no-return function cannot be reached, and other facts.</p>
2985 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2989 <!-- ======================================================================= -->
2990 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2992 <div class="doc_text">
2994 <p>Binary operators are used to do most of the computation in a program. They
2995 require two operands of the same type, execute an operation on them, and
2996 produce a single value. The operands might represent multiple data, as is
2997 the case with the <a href="#t_vector">vector</a> data type. The result value
2998 has the same type as its operands.</p>
3000 <p>There are several different binary operators:</p>
3004 <!-- _______________________________________________________________________ -->
3005 <div class="doc_subsubsection">
3006 <a name="i_add">'<tt>add</tt>' Instruction</a>
3009 <div class="doc_text">
3013 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3014 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3015 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3016 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3020 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3023 <p>The two arguments to the '<tt>add</tt>' instruction must
3024 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3025 integer values. Both arguments must have identical types.</p>
3028 <p>The value produced is the integer sum of the two operands.</p>
3030 <p>If the sum has unsigned overflow, the result returned is the mathematical
3031 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3033 <p>Because LLVM integers use a two's complement representation, this instruction
3034 is appropriate for both signed and unsigned integers.</p>
3036 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3037 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3038 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3039 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
3043 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3048 <!-- _______________________________________________________________________ -->
3049 <div class="doc_subsubsection">
3050 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3053 <div class="doc_text">
3057 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3061 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3064 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3065 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3066 floating point values. Both arguments must have identical types.</p>
3069 <p>The value produced is the floating point sum of the two operands.</p>
3073 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3078 <!-- _______________________________________________________________________ -->
3079 <div class="doc_subsubsection">
3080 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3083 <div class="doc_text">
3087 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3088 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3089 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3090 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3094 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3097 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3098 '<tt>neg</tt>' instruction present in most other intermediate
3099 representations.</p>
3102 <p>The two arguments to the '<tt>sub</tt>' instruction must
3103 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3104 integer values. Both arguments must have identical types.</p>
3107 <p>The value produced is the integer difference of the two operands.</p>
3109 <p>If the difference has unsigned overflow, the result returned is the
3110 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3113 <p>Because LLVM integers use a two's complement representation, this instruction
3114 is appropriate for both signed and unsigned integers.</p>
3116 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3117 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3118 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3119 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
3123 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3124 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3129 <!-- _______________________________________________________________________ -->
3130 <div class="doc_subsubsection">
3131 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3134 <div class="doc_text">
3138 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3142 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3145 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3146 '<tt>fneg</tt>' instruction present in most other intermediate
3147 representations.</p>
3150 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3151 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3152 floating point values. Both arguments must have identical types.</p>
3155 <p>The value produced is the floating point difference of the two operands.</p>
3159 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3160 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3165 <!-- _______________________________________________________________________ -->
3166 <div class="doc_subsubsection">
3167 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3170 <div class="doc_text">
3174 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3175 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3176 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3177 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3181 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3184 <p>The two arguments to the '<tt>mul</tt>' instruction must
3185 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3186 integer values. Both arguments must have identical types.</p>
3189 <p>The value produced is the integer product of the two operands.</p>
3191 <p>If the result of the multiplication has unsigned overflow, the result
3192 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3193 width of the result.</p>
3195 <p>Because LLVM integers use a two's complement representation, and the result
3196 is the same width as the operands, this instruction returns the correct
3197 result for both signed and unsigned integers. If a full product
3198 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3199 be sign-extended or zero-extended as appropriate to the width of the full
3202 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3203 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3204 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3205 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
3209 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3214 <!-- _______________________________________________________________________ -->
3215 <div class="doc_subsubsection">
3216 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3219 <div class="doc_text">
3223 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3227 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3230 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3231 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3232 floating point values. Both arguments must have identical types.</p>
3235 <p>The value produced is the floating point product of the two operands.</p>
3239 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3244 <!-- _______________________________________________________________________ -->
3245 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
3248 <div class="doc_text">
3252 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3256 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3259 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3260 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3261 values. Both arguments must have identical types.</p>
3264 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3266 <p>Note that unsigned integer division and signed integer division are distinct
3267 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3269 <p>Division by zero leads to undefined behavior.</p>
3273 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3278 <!-- _______________________________________________________________________ -->
3279 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
3282 <div class="doc_text">
3286 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3287 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3291 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3294 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3295 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3296 values. Both arguments must have identical types.</p>
3299 <p>The value produced is the signed integer quotient of the two operands rounded
3302 <p>Note that signed integer division and unsigned integer division are distinct
3303 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3305 <p>Division by zero leads to undefined behavior. Overflow also leads to
3306 undefined behavior; this is a rare case, but can occur, for example, by doing
3307 a 32-bit division of -2147483648 by -1.</p>
3309 <p>If the <tt>exact</tt> keyword is present, the result value of the
3310 <tt>sdiv</tt> is undefined if the result would be rounded or if overflow
3315 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3320 <!-- _______________________________________________________________________ -->
3321 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
3322 Instruction</a> </div>
3324 <div class="doc_text">
3328 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3332 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3335 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3336 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3337 floating point values. Both arguments must have identical types.</p>
3340 <p>The value produced is the floating point quotient of the two operands.</p>
3344 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3349 <!-- _______________________________________________________________________ -->
3350 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3353 <div class="doc_text">
3357 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3361 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3362 division of its two arguments.</p>
3365 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3366 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3367 values. Both arguments must have identical types.</p>
3370 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3371 This instruction always performs an unsigned division to get the
3374 <p>Note that unsigned integer remainder and signed integer remainder are
3375 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3377 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3381 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3386 <!-- _______________________________________________________________________ -->
3387 <div class="doc_subsubsection">
3388 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3391 <div class="doc_text">
3395 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3399 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3400 division of its two operands. This instruction can also take
3401 <a href="#t_vector">vector</a> versions of the values in which case the
3402 elements must be integers.</p>
3405 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3406 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3407 values. Both arguments must have identical types.</p>
3410 <p>This instruction returns the <i>remainder</i> of a division (where the result
3411 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
3412 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
3413 a value. For more information about the difference,
3414 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3415 Math Forum</a>. For a table of how this is implemented in various languages,
3416 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3417 Wikipedia: modulo operation</a>.</p>
3419 <p>Note that signed integer remainder and unsigned integer remainder are
3420 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3422 <p>Taking the remainder of a division by zero leads to undefined behavior.
3423 Overflow also leads to undefined behavior; this is a rare case, but can
3424 occur, for example, by taking the remainder of a 32-bit division of
3425 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3426 lets srem be implemented using instructions that return both the result of
3427 the division and the remainder.)</p>
3431 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3436 <!-- _______________________________________________________________________ -->
3437 <div class="doc_subsubsection">
3438 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
3440 <div class="doc_text">
3444 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3448 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3449 its two operands.</p>
3452 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3453 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3454 floating point values. Both arguments must have identical types.</p>
3457 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3458 has the same sign as the dividend.</p>
3462 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3467 <!-- ======================================================================= -->
3468 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
3469 Operations</a> </div>
3471 <div class="doc_text">
3473 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3474 program. They are generally very efficient instructions and can commonly be
3475 strength reduced from other instructions. They require two operands of the
3476 same type, execute an operation on them, and produce a single value. The
3477 resulting value is the same type as its operands.</p>
3481 <!-- _______________________________________________________________________ -->
3482 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
3483 Instruction</a> </div>
3485 <div class="doc_text">
3489 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3493 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3494 a specified number of bits.</p>
3497 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3498 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3499 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3502 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3503 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3504 is (statically or dynamically) negative or equal to or larger than the number
3505 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3506 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3507 shift amount in <tt>op2</tt>.</p>
3511 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3512 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3513 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3514 <result> = shl i32 1, 32 <i>; undefined</i>
3515 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3520 <!-- _______________________________________________________________________ -->
3521 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
3522 Instruction</a> </div>
3524 <div class="doc_text">
3528 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3532 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
3533 operand shifted to the right a specified number of bits with zero fill.</p>
3536 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
3537 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3538 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3541 <p>This instruction always performs a logical shift right operation. The most
3542 significant bits of the result will be filled with zero bits after the shift.
3543 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
3544 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3545 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3546 shift amount in <tt>op2</tt>.</p>
3550 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
3551 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
3552 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
3553 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
3554 <result> = lshr i32 1, 32 <i>; undefined</i>
3555 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3560 <!-- _______________________________________________________________________ -->
3561 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
3562 Instruction</a> </div>
3563 <div class="doc_text">
3567 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3571 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
3572 operand shifted to the right a specified number of bits with sign
3576 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
3577 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3578 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3581 <p>This instruction always performs an arithmetic shift right operation, The
3582 most significant bits of the result will be filled with the sign bit
3583 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
3584 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
3585 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
3586 the corresponding shift amount in <tt>op2</tt>.</p>
3590 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
3591 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
3592 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
3593 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
3594 <result> = ashr i32 1, 32 <i>; undefined</i>
3595 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3600 <!-- _______________________________________________________________________ -->
3601 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
3602 Instruction</a> </div>
3604 <div class="doc_text">
3608 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3612 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
3616 <p>The two arguments to the '<tt>and</tt>' instruction must be
3617 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3618 values. Both arguments must have identical types.</p>
3621 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
3623 <table border="1" cellspacing="0" cellpadding="4">
3655 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
3656 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
3657 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
3660 <!-- _______________________________________________________________________ -->
3661 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
3663 <div class="doc_text">
3667 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3671 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
3675 <p>The two arguments to the '<tt>or</tt>' instruction must be
3676 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3677 values. Both arguments must have identical types.</p>
3680 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
3682 <table border="1" cellspacing="0" cellpadding="4">
3714 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3715 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3716 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3721 <!-- _______________________________________________________________________ -->
3722 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3723 Instruction</a> </div>
3725 <div class="doc_text">
3729 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3733 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
3734 its two operands. The <tt>xor</tt> is used to implement the "one's
3735 complement" operation, which is the "~" operator in C.</p>
3738 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3739 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3740 values. Both arguments must have identical types.</p>
3743 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3745 <table border="1" cellspacing="0" cellpadding="4">
3777 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3778 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3779 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3780 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3785 <!-- ======================================================================= -->
3786 <div class="doc_subsection">
3787 <a name="vectorops">Vector Operations</a>
3790 <div class="doc_text">
3792 <p>LLVM supports several instructions to represent vector operations in a
3793 target-independent manner. These instructions cover the element-access and
3794 vector-specific operations needed to process vectors effectively. While LLVM
3795 does directly support these vector operations, many sophisticated algorithms
3796 will want to use target-specific intrinsics to take full advantage of a
3797 specific target.</p>
3801 <!-- _______________________________________________________________________ -->
3802 <div class="doc_subsubsection">
3803 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3806 <div class="doc_text">
3810 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3814 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
3815 from a vector at a specified index.</p>
3819 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
3820 of <a href="#t_vector">vector</a> type. The second operand is an index
3821 indicating the position from which to extract the element. The index may be
3825 <p>The result is a scalar of the same type as the element type of
3826 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3827 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3828 results are undefined.</p>
3832 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3837 <!-- _______________________________________________________________________ -->
3838 <div class="doc_subsubsection">
3839 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3842 <div class="doc_text">
3846 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3850 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
3851 vector at a specified index.</p>
3854 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
3855 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
3856 whose type must equal the element type of the first operand. The third
3857 operand is an index indicating the position at which to insert the value.
3858 The index may be a variable.</p>
3861 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
3862 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
3863 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3864 results are undefined.</p>
3868 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3873 <!-- _______________________________________________________________________ -->
3874 <div class="doc_subsubsection">
3875 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3878 <div class="doc_text">
3882 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3886 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3887 from two input vectors, returning a vector with the same element type as the
3888 input and length that is the same as the shuffle mask.</p>
3891 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3892 with types that match each other. The third argument is a shuffle mask whose
3893 element type is always 'i32'. The result of the instruction is a vector
3894 whose length is the same as the shuffle mask and whose element type is the
3895 same as the element type of the first two operands.</p>
3897 <p>The shuffle mask operand is required to be a constant vector with either
3898 constant integer or undef values.</p>
3901 <p>The elements of the two input vectors are numbered from left to right across
3902 both of the vectors. The shuffle mask operand specifies, for each element of
3903 the result vector, which element of the two input vectors the result element
3904 gets. The element selector may be undef (meaning "don't care") and the
3905 second operand may be undef if performing a shuffle from only one vector.</p>
3909 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3910 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3911 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
3912 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3913 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
3914 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3915 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3916 <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>
3921 <!-- ======================================================================= -->
3922 <div class="doc_subsection">
3923 <a name="aggregateops">Aggregate Operations</a>
3926 <div class="doc_text">
3928 <p>LLVM supports several instructions for working with
3929 <a href="#t_aggregate">aggregate</a> values.</p>
3933 <!-- _______________________________________________________________________ -->
3934 <div class="doc_subsubsection">
3935 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3938 <div class="doc_text">
3942 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3946 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
3947 from an <a href="#t_aggregate">aggregate</a> value.</p>
3950 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
3951 of <a href="#t_struct">struct</a>, <a href="#t_union">union</a> or
3952 <a href="#t_array">array</a> type. The operands are constant indices to
3953 specify which value to extract in a similar manner as indices in a
3954 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
3957 <p>The result is the value at the position in the aggregate specified by the
3962 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3967 <!-- _______________________________________________________________________ -->
3968 <div class="doc_subsubsection">
3969 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3972 <div class="doc_text">
3976 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx> <i>; yields <aggregate type></i>
3980 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
3981 in an <a href="#t_aggregate">aggregate</a> value.</p>
3984 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
3985 of <a href="#t_struct">struct</a>, <a href="#t_union">union</a> or
3986 <a href="#t_array">array</a> type. The second operand is a first-class
3987 value to insert. The following operands are constant indices indicating
3988 the position at which to insert the value in a similar manner as indices in a
3989 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction. The
3990 value to insert must have the same type as the value identified by the
3994 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
3995 that of <tt>val</tt> except that the value at the position specified by the
3996 indices is that of <tt>elt</tt>.</p>
4000 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4001 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4007 <!-- ======================================================================= -->
4008 <div class="doc_subsection">
4009 <a name="memoryops">Memory Access and Addressing Operations</a>
4012 <div class="doc_text">
4014 <p>A key design point of an SSA-based representation is how it represents
4015 memory. In LLVM, no memory locations are in SSA form, which makes things
4016 very simple. This section describes how to read, write, and allocate
4021 <!-- _______________________________________________________________________ -->
4022 <div class="doc_subsubsection">
4023 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4026 <div class="doc_text">
4030 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4034 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4035 currently executing function, to be automatically released when this function
4036 returns to its caller. The object is always allocated in the generic address
4037 space (address space zero).</p>
4040 <p>The '<tt>alloca</tt>' instruction
4041 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4042 runtime stack, returning a pointer of the appropriate type to the program.
4043 If "NumElements" is specified, it is the number of elements allocated,
4044 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4045 specified, the value result of the allocation is guaranteed to be aligned to
4046 at least that boundary. If not specified, or if zero, the target can choose
4047 to align the allocation on any convenient boundary compatible with the
4050 <p>'<tt>type</tt>' may be any sized type.</p>
4053 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4054 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4055 memory is automatically released when the function returns. The
4056 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4057 variables that must have an address available. When the function returns
4058 (either with the <tt><a href="#i_ret">ret</a></tt>
4059 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
4060 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4064 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4065 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4066 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4067 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4072 <!-- _______________________________________________________________________ -->
4073 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
4074 Instruction</a> </div>
4076 <div class="doc_text">
4080 <result> = load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4081 <result> = volatile load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4082 !<index> = !{ i32 1 }
4086 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4089 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4090 from which to load. The pointer must point to
4091 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4092 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4093 number or order of execution of this <tt>load</tt> with other
4094 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
4097 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4098 operation (that is, the alignment of the memory address). A value of 0 or an
4099 omitted <tt>align</tt> argument means that the operation has the preferential
4100 alignment for the target. It is the responsibility of the code emitter to
4101 ensure that the alignment information is correct. Overestimating the
4102 alignment results in undefined behavior. Underestimating the alignment may
4103 produce less efficient code. An alignment of 1 is always safe.</p>
4105 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4106 metatadata name <index> corresponding to a metadata node with
4107 one <tt>i32</tt> entry of value 1. The existance of
4108 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4109 and code generator that this load is not expected to be reused in the cache.
4110 The code generator may select special instructions to save cache bandwidth,
4111 such as the <tt>MOVNT</tt> intruction on x86.</p>
4114 <p>The location of memory pointed to is loaded. If the value being loaded is of
4115 scalar type then the number of bytes read does not exceed the minimum number
4116 of bytes needed to hold all bits of the type. For example, loading an
4117 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4118 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4119 is undefined if the value was not originally written using a store of the
4124 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4125 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4126 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4131 <!-- _______________________________________________________________________ -->
4132 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
4133 Instruction</a> </div>
4135 <div class="doc_text">
4139 store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4140 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4144 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4147 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4148 and an address at which to store it. The type of the
4149 '<tt><pointer></tt>' operand must be a pointer to
4150 the <a href="#t_firstclass">first class</a> type of the
4151 '<tt><value></tt>' operand. If the <tt>store</tt> is marked
4152 as <tt>volatile</tt>, then the optimizer is not allowed to modify the number
4153 or order of execution of this <tt>store</tt> with other
4154 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
4157 <p>The optional constant "align" argument specifies the alignment of the
4158 operation (that is, the alignment of the memory address). A value of 0 or an
4159 omitted "align" argument means that the operation has the preferential
4160 alignment for the target. It is the responsibility of the code emitter to
4161 ensure that the alignment information is correct. Overestimating the
4162 alignment results in an undefined behavior. Underestimating the alignment may
4163 produce less efficient code. An alignment of 1 is always safe.</p>
4165 <p>The optional !nontemporal metadata must reference a single metatadata
4166 name <index> corresponding to a metadata node with one i32 entry of
4167 value 1. The existance of the !nontemporal metatadata on the
4168 instruction tells the optimizer and code generator that this load is
4169 not expected to be reused in the cache. The code generator may
4170 select special instructions to save cache bandwidth, such as the
4171 MOVNT intruction on x86.</p>
4175 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4176 location specified by the '<tt><pointer></tt>' operand. If
4177 '<tt><value></tt>' is of scalar type then the number of bytes written
4178 does not exceed the minimum number of bytes needed to hold all bits of the
4179 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4180 writing a value of a type like <tt>i20</tt> with a size that is not an
4181 integral number of bytes, it is unspecified what happens to the extra bits
4182 that do not belong to the type, but they will typically be overwritten.</p>
4186 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4187 store i32 3, i32* %ptr <i>; yields {void}</i>
4188 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4193 <!-- _______________________________________________________________________ -->
4194 <div class="doc_subsubsection">
4195 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
4198 <div class="doc_text">
4202 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4203 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4207 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4208 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
4209 It performs address calculation only and does not access memory.</p>
4212 <p>The first argument is always a pointer, and forms the basis of the
4213 calculation. The remaining arguments are indices that indicate which of the
4214 elements of the aggregate object are indexed. The interpretation of each
4215 index is dependent on the type being indexed into. The first index always
4216 indexes the pointer value given as the first argument, the second index
4217 indexes a value of the type pointed to (not necessarily the value directly
4218 pointed to, since the first index can be non-zero), etc. The first type
4219 indexed into must be a pointer value, subsequent types can be arrays,
4220 vectors, structs and unions. Note that subsequent types being indexed into
4221 can never be pointers, since that would require loading the pointer before
4222 continuing calculation.</p>
4224 <p>The type of each index argument depends on the type it is indexing into.
4225 When indexing into a (optionally packed) structure or union, only <tt>i32</tt>
4226 integer <b>constants</b> are allowed. When indexing into an array, pointer
4227 or vector, integers of any width are allowed, and they are not required to be
4230 <p>For example, let's consider a C code fragment and how it gets compiled to
4233 <div class="doc_code">
4246 int *foo(struct ST *s) {
4247 return &s[1].Z.B[5][13];
4252 <p>The LLVM code generated by the GCC frontend is:</p>
4254 <div class="doc_code">
4256 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4257 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
4259 define i32* @foo(%ST* %s) {
4261 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
4268 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
4269 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
4270 }</tt>' type, a structure. The second index indexes into the third element
4271 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
4272 i8 }</tt>' type, another structure. The third index indexes into the second
4273 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
4274 array. The two dimensions of the array are subscripted into, yielding an
4275 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
4276 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
4278 <p>Note that it is perfectly legal to index partially through a structure,
4279 returning a pointer to an inner element. Because of this, the LLVM code for
4280 the given testcase is equivalent to:</p>
4283 define i32* @foo(%ST* %s) {
4284 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
4285 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
4286 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
4287 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
4288 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
4293 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
4294 <tt>getelementptr</tt> is undefined if the base pointer is not an
4295 <i>in bounds</i> address of an allocated object, or if any of the addresses
4296 that would be formed by successive addition of the offsets implied by the
4297 indices to the base address with infinitely precise arithmetic are not an
4298 <i>in bounds</i> address of that allocated object.
4299 The <i>in bounds</i> addresses for an allocated object are all the addresses
4300 that point into the object, plus the address one byte past the end.</p>
4302 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
4303 the base address with silently-wrapping two's complement arithmetic, and
4304 the result value of the <tt>getelementptr</tt> may be outside the object
4305 pointed to by the base pointer. The result value may not necessarily be
4306 used to access memory though, even if it happens to point into allocated
4307 storage. See the <a href="#pointeraliasing">Pointer Aliasing Rules</a>
4308 section for more information.</p>
4310 <p>The getelementptr instruction is often confusing. For some more insight into
4311 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
4315 <i>; yields [12 x i8]*:aptr</i>
4316 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4317 <i>; yields i8*:vptr</i>
4318 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4319 <i>; yields i8*:eptr</i>
4320 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4321 <i>; yields i32*:iptr</i>
4322 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4327 <!-- ======================================================================= -->
4328 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
4331 <div class="doc_text">
4333 <p>The instructions in this category are the conversion instructions (casting)
4334 which all take a single operand and a type. They perform various bit
4335 conversions on the operand.</p>
4339 <!-- _______________________________________________________________________ -->
4340 <div class="doc_subsubsection">
4341 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
4343 <div class="doc_text">
4347 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
4351 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
4352 type <tt>ty2</tt>.</p>
4355 <p>The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
4356 be an <a href="#t_integer">integer</a> type, and a type that specifies the
4357 size and type of the result, which must be
4358 an <a href="#t_integer">integer</a> type. The bit size of <tt>value</tt> must
4359 be larger than the bit size of <tt>ty2</tt>. Equal sized types are not
4363 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
4364 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
4365 source size must be larger than the destination size, <tt>trunc</tt> cannot
4366 be a <i>no-op cast</i>. It will always truncate bits.</p>
4370 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
4371 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
4372 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
4377 <!-- _______________________________________________________________________ -->
4378 <div class="doc_subsubsection">
4379 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
4381 <div class="doc_text">
4385 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
4389 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
4394 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
4395 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4396 also be of <a href="#t_integer">integer</a> type. The bit size of the
4397 <tt>value</tt> must be smaller than the bit size of the destination type,
4401 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
4402 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
4404 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
4408 %X = zext i32 257 to i64 <i>; yields i64:257</i>
4409 %Y = zext i1 true to i32 <i>; yields i32:1</i>
4414 <!-- _______________________________________________________________________ -->
4415 <div class="doc_subsubsection">
4416 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
4418 <div class="doc_text">
4422 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
4426 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
4429 <p>The '<tt>sext</tt>' instruction takes a value to cast, which must be of
4430 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4431 also be of <a href="#t_integer">integer</a> type. The bit size of the
4432 <tt>value</tt> must be smaller than the bit size of the destination type,
4436 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
4437 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
4438 of the type <tt>ty2</tt>.</p>
4440 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
4444 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
4445 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
4450 <!-- _______________________________________________________________________ -->
4451 <div class="doc_subsubsection">
4452 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
4455 <div class="doc_text">
4459 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
4463 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
4467 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
4468 point</a> value to cast and a <a href="#t_floating">floating point</a> type
4469 to cast it to. The size of <tt>value</tt> must be larger than the size of
4470 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
4471 <i>no-op cast</i>.</p>
4474 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
4475 <a href="#t_floating">floating point</a> type to a smaller
4476 <a href="#t_floating">floating point</a> type. If the value cannot fit
4477 within the destination type, <tt>ty2</tt>, then the results are
4482 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
4483 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
4488 <!-- _______________________________________________________________________ -->
4489 <div class="doc_subsubsection">
4490 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
4492 <div class="doc_text">
4496 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
4500 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
4501 floating point value.</p>
4504 <p>The '<tt>fpext</tt>' instruction takes a
4505 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
4506 a <a href="#t_floating">floating point</a> type to cast it to. The source
4507 type must be smaller than the destination type.</p>
4510 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
4511 <a href="#t_floating">floating point</a> type to a larger
4512 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
4513 used to make a <i>no-op cast</i> because it always changes bits. Use
4514 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
4518 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
4519 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
4524 <!-- _______________________________________________________________________ -->
4525 <div class="doc_subsubsection">
4526 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
4528 <div class="doc_text">
4532 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
4536 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
4537 unsigned integer equivalent of type <tt>ty2</tt>.</p>
4540 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
4541 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4542 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4543 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4544 vector integer type with the same number of elements as <tt>ty</tt></p>
4547 <p>The '<tt>fptoui</tt>' instruction converts its
4548 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4549 towards zero) unsigned integer value. If the value cannot fit
4550 in <tt>ty2</tt>, the results are undefined.</p>
4554 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
4555 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
4556 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
4561 <!-- _______________________________________________________________________ -->
4562 <div class="doc_subsubsection">
4563 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
4565 <div class="doc_text">
4569 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
4573 <p>The '<tt>fptosi</tt>' instruction converts
4574 <a href="#t_floating">floating point</a> <tt>value</tt> to
4575 type <tt>ty2</tt>.</p>
4578 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
4579 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4580 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4581 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4582 vector integer type with the same number of elements as <tt>ty</tt></p>
4585 <p>The '<tt>fptosi</tt>' instruction converts its
4586 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4587 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
4588 the results are undefined.</p>
4592 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
4593 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
4594 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
4599 <!-- _______________________________________________________________________ -->
4600 <div class="doc_subsubsection">
4601 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
4603 <div class="doc_text">
4607 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4611 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4612 integer and converts that value to the <tt>ty2</tt> type.</p>
4615 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4616 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4617 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4618 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4619 floating point type with the same number of elements as <tt>ty</tt></p>
4622 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4623 integer quantity and converts it to the corresponding floating point
4624 value. If the value cannot fit in the floating point value, the results are
4629 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4630 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4635 <!-- _______________________________________________________________________ -->
4636 <div class="doc_subsubsection">
4637 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4639 <div class="doc_text">
4643 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4647 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
4648 and converts that value to the <tt>ty2</tt> type.</p>
4651 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4652 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4653 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4654 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4655 floating point type with the same number of elements as <tt>ty</tt></p>
4658 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
4659 quantity and converts it to the corresponding floating point value. If the
4660 value cannot fit in the floating point value, the results are undefined.</p>
4664 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4665 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4670 <!-- _______________________________________________________________________ -->
4671 <div class="doc_subsubsection">
4672 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4674 <div class="doc_text">
4678 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4682 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4683 the integer type <tt>ty2</tt>.</p>
4686 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4687 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4688 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4691 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4692 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4693 truncating or zero extending that value to the size of the integer type. If
4694 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4695 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4696 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4701 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4702 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4707 <!-- _______________________________________________________________________ -->
4708 <div class="doc_subsubsection">
4709 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4711 <div class="doc_text">
4715 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4719 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
4720 pointer type, <tt>ty2</tt>.</p>
4723 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4724 value to cast, and a type to cast it to, which must be a
4725 <a href="#t_pointer">pointer</a> type.</p>
4728 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4729 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4730 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4731 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
4732 than the size of a pointer then a zero extension is done. If they are the
4733 same size, nothing is done (<i>no-op cast</i>).</p>
4737 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4738 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4739 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4744 <!-- _______________________________________________________________________ -->
4745 <div class="doc_subsubsection">
4746 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4748 <div class="doc_text">
4752 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4756 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4757 <tt>ty2</tt> without changing any bits.</p>
4760 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
4761 non-aggregate first class value, and a type to cast it to, which must also be
4762 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
4763 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
4764 identical. If the source type is a pointer, the destination type must also be
4765 a pointer. This instruction supports bitwise conversion of vectors to
4766 integers and to vectors of other types (as long as they have the same
4770 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4771 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4772 this conversion. The conversion is done as if the <tt>value</tt> had been
4773 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
4774 be converted to other pointer types with this instruction. To convert
4775 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
4776 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4780 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4781 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4782 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4787 <!-- ======================================================================= -->
4788 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4790 <div class="doc_text">
4792 <p>The instructions in this category are the "miscellaneous" instructions, which
4793 defy better classification.</p>
4797 <!-- _______________________________________________________________________ -->
4798 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4801 <div class="doc_text">
4805 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4809 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
4810 boolean values based on comparison of its two integer, integer vector, or
4811 pointer operands.</p>
4814 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4815 the condition code indicating the kind of comparison to perform. It is not a
4816 value, just a keyword. The possible condition code are:</p>
4819 <li><tt>eq</tt>: equal</li>
4820 <li><tt>ne</tt>: not equal </li>
4821 <li><tt>ugt</tt>: unsigned greater than</li>
4822 <li><tt>uge</tt>: unsigned greater or equal</li>
4823 <li><tt>ult</tt>: unsigned less than</li>
4824 <li><tt>ule</tt>: unsigned less or equal</li>
4825 <li><tt>sgt</tt>: signed greater than</li>
4826 <li><tt>sge</tt>: signed greater or equal</li>
4827 <li><tt>slt</tt>: signed less than</li>
4828 <li><tt>sle</tt>: signed less or equal</li>
4831 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4832 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
4833 typed. They must also be identical types.</p>
4836 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
4837 condition code given as <tt>cond</tt>. The comparison performed always yields
4838 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
4839 result, as follows:</p>
4842 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4843 <tt>false</tt> otherwise. No sign interpretation is necessary or
4846 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4847 <tt>false</tt> otherwise. No sign interpretation is necessary or
4850 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4851 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4853 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4854 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4855 to <tt>op2</tt>.</li>
4857 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4858 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4860 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4861 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4863 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4864 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4866 <li><tt>sge</tt>: interprets the operands as signed values and yields
4867 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4868 to <tt>op2</tt>.</li>
4870 <li><tt>slt</tt>: interprets the operands as signed values and yields
4871 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4873 <li><tt>sle</tt>: interprets the operands as signed values and yields
4874 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4877 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4878 values are compared as if they were integers.</p>
4880 <p>If the operands are integer vectors, then they are compared element by
4881 element. The result is an <tt>i1</tt> vector with the same number of elements
4882 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
4886 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4887 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4888 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4889 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4890 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4891 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4894 <p>Note that the code generator does not yet support vector types with
4895 the <tt>icmp</tt> instruction.</p>
4899 <!-- _______________________________________________________________________ -->
4900 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4903 <div class="doc_text">
4907 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4911 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
4912 values based on comparison of its operands.</p>
4914 <p>If the operands are floating point scalars, then the result type is a boolean
4915 (<a href="#t_integer"><tt>i1</tt></a>).</p>
4917 <p>If the operands are floating point vectors, then the result type is a vector
4918 of boolean with the same number of elements as the operands being
4922 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4923 the condition code indicating the kind of comparison to perform. It is not a
4924 value, just a keyword. The possible condition code are:</p>
4927 <li><tt>false</tt>: no comparison, always returns false</li>
4928 <li><tt>oeq</tt>: ordered and equal</li>
4929 <li><tt>ogt</tt>: ordered and greater than </li>
4930 <li><tt>oge</tt>: ordered and greater than or equal</li>
4931 <li><tt>olt</tt>: ordered and less than </li>
4932 <li><tt>ole</tt>: ordered and less than or equal</li>
4933 <li><tt>one</tt>: ordered and not equal</li>
4934 <li><tt>ord</tt>: ordered (no nans)</li>
4935 <li><tt>ueq</tt>: unordered or equal</li>
4936 <li><tt>ugt</tt>: unordered or greater than </li>
4937 <li><tt>uge</tt>: unordered or greater than or equal</li>
4938 <li><tt>ult</tt>: unordered or less than </li>
4939 <li><tt>ule</tt>: unordered or less than or equal</li>
4940 <li><tt>une</tt>: unordered or not equal</li>
4941 <li><tt>uno</tt>: unordered (either nans)</li>
4942 <li><tt>true</tt>: no comparison, always returns true</li>
4945 <p><i>Ordered</i> means that neither operand is a QNAN while
4946 <i>unordered</i> means that either operand may be a QNAN.</p>
4948 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
4949 a <a href="#t_floating">floating point</a> type or
4950 a <a href="#t_vector">vector</a> of floating point type. They must have
4951 identical types.</p>
4954 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4955 according to the condition code given as <tt>cond</tt>. If the operands are
4956 vectors, then the vectors are compared element by element. Each comparison
4957 performed always yields an <a href="#t_integer">i1</a> result, as
4961 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4963 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4964 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4966 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4967 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4969 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4970 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4972 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4973 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4975 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4976 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4978 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4979 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4981 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4983 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4984 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4986 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4987 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4989 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4990 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4992 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4993 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4995 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4996 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4998 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4999 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5001 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
5003 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5008 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
5009 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5010 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5011 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
5014 <p>Note that the code generator does not yet support vector types with
5015 the <tt>fcmp</tt> instruction.</p>
5019 <!-- _______________________________________________________________________ -->
5020 <div class="doc_subsubsection">
5021 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
5024 <div class="doc_text">
5028 <result> = phi <ty> [ <val0>, <label0>], ...
5032 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
5033 SSA graph representing the function.</p>
5036 <p>The type of the incoming values is specified with the first type field. After
5037 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5038 one pair for each predecessor basic block of the current block. Only values
5039 of <a href="#t_firstclass">first class</a> type may be used as the value
5040 arguments to the PHI node. Only labels may be used as the label
5043 <p>There must be no non-phi instructions between the start of a basic block and
5044 the PHI instructions: i.e. PHI instructions must be first in a basic
5047 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5048 occur on the edge from the corresponding predecessor block to the current
5049 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5050 value on the same edge).</p>
5053 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
5054 specified by the pair corresponding to the predecessor basic block that
5055 executed just prior to the current block.</p>
5059 Loop: ; Infinite loop that counts from 0 on up...
5060 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5061 %nextindvar = add i32 %indvar, 1
5067 <!-- _______________________________________________________________________ -->
5068 <div class="doc_subsubsection">
5069 <a name="i_select">'<tt>select</tt>' Instruction</a>
5072 <div class="doc_text">
5076 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5078 <i>selty</i> is either i1 or {<N x i1>}
5082 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
5083 condition, without branching.</p>
5087 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5088 values indicating the condition, and two values of the
5089 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
5090 vectors and the condition is a scalar, then entire vectors are selected, not
5091 individual elements.</p>
5094 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
5095 first value argument; otherwise, it returns the second value argument.</p>
5097 <p>If the condition is a vector of i1, then the value arguments must be vectors
5098 of the same size, and the selection is done element by element.</p>
5102 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
5105 <p>Note that the code generator does not yet support conditions
5106 with vector type.</p>
5110 <!-- _______________________________________________________________________ -->
5111 <div class="doc_subsubsection">
5112 <a name="i_call">'<tt>call</tt>' Instruction</a>
5115 <div class="doc_text">
5119 <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>]
5123 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
5126 <p>This instruction requires several arguments:</p>
5129 <li>The optional "tail" marker indicates that the callee function does not
5130 access any allocas or varargs in the caller. Note that calls may be
5131 marked "tail" even if they do not occur before
5132 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
5133 present, the function call is eligible for tail call optimization,
5134 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
5135 optimized into a jump</a>. As of this writing, the extra requirements for
5136 a call to actually be optimized are:
5138 <li>Caller and callee both have the calling
5139 convention <tt>fastcc</tt>.</li>
5140 <li>The call is in tail position (ret immediately follows call and ret
5141 uses value of call or is void).</li>
5142 <li>Option <tt>-tailcallopt</tt> is enabled,
5143 or <code>llvm::PerformTailCallOpt</code> is <code>true</code>.</li>
5144 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
5145 constraints are met.</a></li>
5149 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5150 convention</a> the call should use. If none is specified, the call
5151 defaults to using C calling conventions. The calling convention of the
5152 call must match the calling convention of the target function, or else the
5153 behavior is undefined.</li>
5155 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5156 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5157 '<tt>inreg</tt>' attributes are valid here.</li>
5159 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5160 type of the return value. Functions that return no value are marked
5161 <tt><a href="#t_void">void</a></tt>.</li>
5163 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5164 being invoked. The argument types must match the types implied by this
5165 signature. This type can be omitted if the function is not varargs and if
5166 the function type does not return a pointer to a function.</li>
5168 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5169 be invoked. In most cases, this is a direct function invocation, but
5170 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5171 to function value.</li>
5173 <li>'<tt>function args</tt>': argument list whose types match the function
5174 signature argument types. All arguments must be of
5175 <a href="#t_firstclass">first class</a> type. If the function signature
5176 indicates the function accepts a variable number of arguments, the extra
5177 arguments can be specified.</li>
5179 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5180 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5181 '<tt>readnone</tt>' attributes are valid here.</li>
5185 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5186 a specified function, with its incoming arguments bound to the specified
5187 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5188 function, control flow continues with the instruction after the function
5189 call, and the return value of the function is bound to the result
5194 %retval = call i32 @test(i32 %argc)
5195 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
5196 %X = tail call i32 @foo() <i>; yields i32</i>
5197 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5198 call void %foo(i8 97 signext)
5200 %struct.A = type { i32, i8 }
5201 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
5202 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5203 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
5204 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
5205 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
5208 <p>llvm treats calls to some functions with names and arguments that match the
5209 standard C99 library as being the C99 library functions, and may perform
5210 optimizations or generate code for them under that assumption. This is
5211 something we'd like to change in the future to provide better support for
5212 freestanding environments and non-C-based langauges.</p>
5216 <!-- _______________________________________________________________________ -->
5217 <div class="doc_subsubsection">
5218 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5221 <div class="doc_text">
5225 <resultval> = va_arg <va_list*> <arglist>, <argty>
5229 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
5230 the "variable argument" area of a function call. It is used to implement the
5231 <tt>va_arg</tt> macro in C.</p>
5234 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
5235 argument. It returns a value of the specified argument type and increments
5236 the <tt>va_list</tt> to point to the next argument. The actual type
5237 of <tt>va_list</tt> is target specific.</p>
5240 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
5241 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
5242 to the next argument. For more information, see the variable argument
5243 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
5245 <p>It is legal for this instruction to be called in a function which does not
5246 take a variable number of arguments, for example, the <tt>vfprintf</tt>
5249 <p><tt>va_arg</tt> is an LLVM instruction instead of
5250 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
5254 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
5256 <p>Note that the code generator does not yet fully support va_arg on many
5257 targets. Also, it does not currently support va_arg with aggregate types on
5262 <!-- *********************************************************************** -->
5263 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
5264 <!-- *********************************************************************** -->
5266 <div class="doc_text">
5268 <p>LLVM supports the notion of an "intrinsic function". These functions have
5269 well known names and semantics and are required to follow certain
5270 restrictions. Overall, these intrinsics represent an extension mechanism for
5271 the LLVM language that does not require changing all of the transformations
5272 in LLVM when adding to the language (or the bitcode reader/writer, the
5273 parser, etc...).</p>
5275 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
5276 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5277 begin with this prefix. Intrinsic functions must always be external
5278 functions: you cannot define the body of intrinsic functions. Intrinsic
5279 functions may only be used in call or invoke instructions: it is illegal to
5280 take the address of an intrinsic function. Additionally, because intrinsic
5281 functions are part of the LLVM language, it is required if any are added that
5282 they be documented here.</p>
5284 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5285 family of functions that perform the same operation but on different data
5286 types. Because LLVM can represent over 8 million different integer types,
5287 overloading is used commonly to allow an intrinsic function to operate on any
5288 integer type. One or more of the argument types or the result type can be
5289 overloaded to accept any integer type. Argument types may also be defined as
5290 exactly matching a previous argument's type or the result type. This allows
5291 an intrinsic function which accepts multiple arguments, but needs all of them
5292 to be of the same type, to only be overloaded with respect to a single
5293 argument or the result.</p>
5295 <p>Overloaded intrinsics will have the names of its overloaded argument types
5296 encoded into its function name, each preceded by a period. Only those types
5297 which are overloaded result in a name suffix. Arguments whose type is matched
5298 against another type do not. For example, the <tt>llvm.ctpop</tt> function
5299 can take an integer of any width and returns an integer of exactly the same
5300 integer width. This leads to a family of functions such as
5301 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
5302 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5303 suffix is required. Because the argument's type is matched against the return
5304 type, it does not require its own name suffix.</p>
5306 <p>To learn how to add an intrinsic function, please see the
5307 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
5311 <!-- ======================================================================= -->
5312 <div class="doc_subsection">
5313 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
5316 <div class="doc_text">
5318 <p>Variable argument support is defined in LLVM with
5319 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
5320 intrinsic functions. These functions are related to the similarly named
5321 macros defined in the <tt><stdarg.h></tt> header file.</p>
5323 <p>All of these functions operate on arguments that use a target-specific value
5324 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5325 not define what this type is, so all transformations should be prepared to
5326 handle these functions regardless of the type used.</p>
5328 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
5329 instruction and the variable argument handling intrinsic functions are
5332 <div class="doc_code">
5334 define i32 @test(i32 %X, ...) {
5335 ; Initialize variable argument processing
5337 %ap2 = bitcast i8** %ap to i8*
5338 call void @llvm.va_start(i8* %ap2)
5340 ; Read a single integer argument
5341 %tmp = va_arg i8** %ap, i32
5343 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5345 %aq2 = bitcast i8** %aq to i8*
5346 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5347 call void @llvm.va_end(i8* %aq2)
5349 ; Stop processing of arguments.
5350 call void @llvm.va_end(i8* %ap2)
5354 declare void @llvm.va_start(i8*)
5355 declare void @llvm.va_copy(i8*, i8*)
5356 declare void @llvm.va_end(i8*)
5362 <!-- _______________________________________________________________________ -->
5363 <div class="doc_subsubsection">
5364 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
5368 <div class="doc_text">
5372 declare void %llvm.va_start(i8* <arglist>)
5376 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
5377 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
5380 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
5383 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
5384 macro available in C. In a target-dependent way, it initializes
5385 the <tt>va_list</tt> element to which the argument points, so that the next
5386 call to <tt>va_arg</tt> will produce the first variable argument passed to
5387 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
5388 need to know the last argument of the function as the compiler can figure
5393 <!-- _______________________________________________________________________ -->
5394 <div class="doc_subsubsection">
5395 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
5398 <div class="doc_text">
5402 declare void @llvm.va_end(i8* <arglist>)
5406 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
5407 which has been initialized previously
5408 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
5409 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
5412 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
5415 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
5416 macro available in C. In a target-dependent way, it destroys
5417 the <tt>va_list</tt> element to which the argument points. Calls
5418 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
5419 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
5420 with calls to <tt>llvm.va_end</tt>.</p>
5424 <!-- _______________________________________________________________________ -->
5425 <div class="doc_subsubsection">
5426 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
5429 <div class="doc_text">
5433 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
5437 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
5438 from the source argument list to the destination argument list.</p>
5441 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
5442 The second argument is a pointer to a <tt>va_list</tt> element to copy
5446 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
5447 macro available in C. In a target-dependent way, it copies the
5448 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
5449 element. This intrinsic is necessary because
5450 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
5451 arbitrarily complex and require, for example, memory allocation.</p>
5455 <!-- ======================================================================= -->
5456 <div class="doc_subsection">
5457 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
5460 <div class="doc_text">
5462 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
5463 Collection</a> (GC) requires the implementation and generation of these
5464 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
5465 roots on the stack</a>, as well as garbage collector implementations that
5466 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
5467 barriers. Front-ends for type-safe garbage collected languages should generate
5468 these intrinsics to make use of the LLVM garbage collectors. For more details,
5469 see <a href="GarbageCollection.html">Accurate Garbage Collection with
5472 <p>The garbage collection intrinsics only operate on objects in the generic
5473 address space (address space zero).</p>
5477 <!-- _______________________________________________________________________ -->
5478 <div class="doc_subsubsection">
5479 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
5482 <div class="doc_text">
5486 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5490 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
5491 the code generator, and allows some metadata to be associated with it.</p>
5494 <p>The first argument specifies the address of a stack object that contains the
5495 root pointer. The second pointer (which must be either a constant or a
5496 global value address) contains the meta-data to be associated with the
5500 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
5501 location. At compile-time, the code generator generates information to allow
5502 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
5503 intrinsic may only be used in a function which <a href="#gc">specifies a GC
5508 <!-- _______________________________________________________________________ -->
5509 <div class="doc_subsubsection">
5510 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
5513 <div class="doc_text">
5517 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5521 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
5522 locations, allowing garbage collector implementations that require read
5526 <p>The second argument is the address to read from, which should be an address
5527 allocated from the garbage collector. The first object is a pointer to the
5528 start of the referenced object, if needed by the language runtime (otherwise
5532 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
5533 instruction, but may be replaced with substantially more complex code by the
5534 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
5535 may only be used in a function which <a href="#gc">specifies a GC
5540 <!-- _______________________________________________________________________ -->
5541 <div class="doc_subsubsection">
5542 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5545 <div class="doc_text">
5549 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5553 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5554 locations, allowing garbage collector implementations that require write
5555 barriers (such as generational or reference counting collectors).</p>
5558 <p>The first argument is the reference to store, the second is the start of the
5559 object to store it to, and the third is the address of the field of Obj to
5560 store to. If the runtime does not require a pointer to the object, Obj may
5564 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5565 instruction, but may be replaced with substantially more complex code by the
5566 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5567 may only be used in a function which <a href="#gc">specifies a GC
5572 <!-- ======================================================================= -->
5573 <div class="doc_subsection">
5574 <a name="int_codegen">Code Generator Intrinsics</a>
5577 <div class="doc_text">
5579 <p>These intrinsics are provided by LLVM to expose special features that may
5580 only be implemented with code generator support.</p>
5584 <!-- _______________________________________________________________________ -->
5585 <div class="doc_subsubsection">
5586 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5589 <div class="doc_text">
5593 declare i8 *@llvm.returnaddress(i32 <level>)
5597 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5598 target-specific value indicating the return address of the current function
5599 or one of its callers.</p>
5602 <p>The argument to this intrinsic indicates which function to return the address
5603 for. Zero indicates the calling function, one indicates its caller, etc.
5604 The argument is <b>required</b> to be a constant integer value.</p>
5607 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
5608 indicating the return address of the specified call frame, or zero if it
5609 cannot be identified. The value returned by this intrinsic is likely to be
5610 incorrect or 0 for arguments other than zero, so it should only be used for
5611 debugging purposes.</p>
5613 <p>Note that calling this intrinsic does not prevent function inlining or other
5614 aggressive transformations, so the value returned may not be that of the
5615 obvious source-language caller.</p>
5619 <!-- _______________________________________________________________________ -->
5620 <div class="doc_subsubsection">
5621 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5624 <div class="doc_text">
5628 declare i8 *@llvm.frameaddress(i32 <level>)
5632 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5633 target-specific frame pointer value for the specified stack frame.</p>
5636 <p>The argument to this intrinsic indicates which function to return the frame
5637 pointer for. Zero indicates the calling function, one indicates its caller,
5638 etc. The argument is <b>required</b> to be a constant integer value.</p>
5641 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
5642 indicating the frame address of the specified call frame, or zero if it
5643 cannot be identified. The value returned by this intrinsic is likely to be
5644 incorrect or 0 for arguments other than zero, so it should only be used for
5645 debugging purposes.</p>
5647 <p>Note that calling this intrinsic does not prevent function inlining or other
5648 aggressive transformations, so the value returned may not be that of the
5649 obvious source-language caller.</p>
5653 <!-- _______________________________________________________________________ -->
5654 <div class="doc_subsubsection">
5655 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5658 <div class="doc_text">
5662 declare i8 *@llvm.stacksave()
5666 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
5667 of the function stack, for use
5668 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
5669 useful for implementing language features like scoped automatic variable
5670 sized arrays in C99.</p>
5673 <p>This intrinsic returns a opaque pointer value that can be passed
5674 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
5675 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
5676 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
5677 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
5678 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
5679 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
5683 <!-- _______________________________________________________________________ -->
5684 <div class="doc_subsubsection">
5685 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5688 <div class="doc_text">
5692 declare void @llvm.stackrestore(i8 * %ptr)
5696 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5697 the function stack to the state it was in when the
5698 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
5699 executed. This is useful for implementing language features like scoped
5700 automatic variable sized arrays in C99.</p>
5703 <p>See the description
5704 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
5708 <!-- _______________________________________________________________________ -->
5709 <div class="doc_subsubsection">
5710 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5713 <div class="doc_text">
5717 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5721 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
5722 insert a prefetch instruction if supported; otherwise, it is a noop.
5723 Prefetches have no effect on the behavior of the program but can change its
5724 performance characteristics.</p>
5727 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
5728 specifier determining if the fetch should be for a read (0) or write (1),
5729 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5730 locality, to (3) - extremely local keep in cache. The <tt>rw</tt>
5731 and <tt>locality</tt> arguments must be constant integers.</p>
5734 <p>This intrinsic does not modify the behavior of the program. In particular,
5735 prefetches cannot trap and do not produce a value. On targets that support
5736 this intrinsic, the prefetch can provide hints to the processor cache for
5737 better performance.</p>
5741 <!-- _______________________________________________________________________ -->
5742 <div class="doc_subsubsection">
5743 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5746 <div class="doc_text">
5750 declare void @llvm.pcmarker(i32 <id>)
5754 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
5755 Counter (PC) in a region of code to simulators and other tools. The method
5756 is target specific, but it is expected that the marker will use exported
5757 symbols to transmit the PC of the marker. The marker makes no guarantees
5758 that it will remain with any specific instruction after optimizations. It is
5759 possible that the presence of a marker will inhibit optimizations. The
5760 intended use is to be inserted after optimizations to allow correlations of
5761 simulation runs.</p>
5764 <p><tt>id</tt> is a numerical id identifying the marker.</p>
5767 <p>This intrinsic does not modify the behavior of the program. Backends that do
5768 not support this intrinisic may ignore it.</p>
5772 <!-- _______________________________________________________________________ -->
5773 <div class="doc_subsubsection">
5774 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5777 <div class="doc_text">
5781 declare i64 @llvm.readcyclecounter( )
5785 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5786 counter register (or similar low latency, high accuracy clocks) on those
5787 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5788 should map to RPCC. As the backing counters overflow quickly (on the order
5789 of 9 seconds on alpha), this should only be used for small timings.</p>
5792 <p>When directly supported, reading the cycle counter should not modify any
5793 memory. Implementations are allowed to either return a application specific
5794 value or a system wide value. On backends without support, this is lowered
5795 to a constant 0.</p>
5799 <!-- ======================================================================= -->
5800 <div class="doc_subsection">
5801 <a name="int_libc">Standard C Library Intrinsics</a>
5804 <div class="doc_text">
5806 <p>LLVM provides intrinsics for a few important standard C library functions.
5807 These intrinsics allow source-language front-ends to pass information about
5808 the alignment of the pointer arguments to the code generator, providing
5809 opportunity for more efficient code generation.</p>
5813 <!-- _______________________________________________________________________ -->
5814 <div class="doc_subsubsection">
5815 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5818 <div class="doc_text">
5821 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
5822 integer bit width. Not all targets support all bit widths however.</p>
5825 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5826 i8 <len>, i32 <align>)
5827 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5828 i16 <len>, i32 <align>)
5829 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5830 i32 <len>, i32 <align>)
5831 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5832 i64 <len>, i32 <align>)
5836 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5837 source location to the destination location.</p>
5839 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5840 intrinsics do not return a value, and takes an extra alignment argument.</p>
5843 <p>The first argument is a pointer to the destination, the second is a pointer
5844 to the source. The third argument is an integer argument specifying the
5845 number of bytes to copy, and the fourth argument is the alignment of the
5846 source and destination locations.</p>
5848 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5849 then the caller guarantees that both the source and destination pointers are
5850 aligned to that boundary.</p>
5853 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5854 source location to the destination location, which are not allowed to
5855 overlap. It copies "len" bytes of memory over. If the argument is known to
5856 be aligned to some boundary, this can be specified as the fourth argument,
5857 otherwise it should be set to 0 or 1.</p>
5861 <!-- _______________________________________________________________________ -->
5862 <div class="doc_subsubsection">
5863 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5866 <div class="doc_text">
5869 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5870 width. Not all targets support all bit widths however.</p>
5873 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5874 i8 <len>, i32 <align>)
5875 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5876 i16 <len>, i32 <align>)
5877 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5878 i32 <len>, i32 <align>)
5879 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5880 i64 <len>, i32 <align>)
5884 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
5885 source location to the destination location. It is similar to the
5886 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
5889 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5890 intrinsics do not return a value, and takes an extra alignment argument.</p>
5893 <p>The first argument is a pointer to the destination, the second is a pointer
5894 to the source. The third argument is an integer argument specifying the
5895 number of bytes to copy, and the fourth argument is the alignment of the
5896 source and destination locations.</p>
5898 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5899 then the caller guarantees that the source and destination pointers are
5900 aligned to that boundary.</p>
5903 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
5904 source location to the destination location, which may overlap. It copies
5905 "len" bytes of memory over. If the argument is known to be aligned to some
5906 boundary, this can be specified as the fourth argument, otherwise it should
5907 be set to 0 or 1.</p>
5911 <!-- _______________________________________________________________________ -->
5912 <div class="doc_subsubsection">
5913 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5916 <div class="doc_text">
5919 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5920 width. Not all targets support all bit widths however.</p>
5923 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5924 i8 <len>, i32 <align>)
5925 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5926 i16 <len>, i32 <align>)
5927 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5928 i32 <len>, i32 <align>)
5929 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5930 i64 <len>, i32 <align>)
5934 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
5935 particular byte value.</p>
5937 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
5938 intrinsic does not return a value, and takes an extra alignment argument.</p>
5941 <p>The first argument is a pointer to the destination to fill, the second is the
5942 byte value to fill it with, the third argument is an integer argument
5943 specifying the number of bytes to fill, and the fourth argument is the known
5944 alignment of destination location.</p>
5946 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5947 then the caller guarantees that the destination pointer is aligned to that
5951 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
5952 at the destination location. If the argument is known to be aligned to some
5953 boundary, this can be specified as the fourth argument, otherwise it should
5954 be set to 0 or 1.</p>
5958 <!-- _______________________________________________________________________ -->
5959 <div class="doc_subsubsection">
5960 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5963 <div class="doc_text">
5966 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5967 floating point or vector of floating point type. Not all targets support all
5971 declare float @llvm.sqrt.f32(float %Val)
5972 declare double @llvm.sqrt.f64(double %Val)
5973 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5974 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5975 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5979 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5980 returning the same value as the libm '<tt>sqrt</tt>' functions would.
5981 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
5982 behavior for negative numbers other than -0.0 (which allows for better
5983 optimization, because there is no need to worry about errno being
5984 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
5987 <p>The argument and return value are floating point numbers of the same
5991 <p>This function returns the sqrt of the specified operand if it is a
5992 nonnegative floating point number.</p>
5996 <!-- _______________________________________________________________________ -->
5997 <div class="doc_subsubsection">
5998 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
6001 <div class="doc_text">
6004 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
6005 floating point or vector of floating point type. Not all targets support all
6009 declare float @llvm.powi.f32(float %Val, i32 %power)
6010 declare double @llvm.powi.f64(double %Val, i32 %power)
6011 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6012 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6013 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6017 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
6018 specified (positive or negative) power. The order of evaluation of
6019 multiplications is not defined. When a vector of floating point type is
6020 used, the second argument remains a scalar integer value.</p>
6023 <p>The second argument is an integer power, and the first is a value to raise to
6027 <p>This function returns the first value raised to the second power with an
6028 unspecified sequence of rounding operations.</p>
6032 <!-- _______________________________________________________________________ -->
6033 <div class="doc_subsubsection">
6034 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
6037 <div class="doc_text">
6040 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
6041 floating point or vector of floating point type. Not all targets support all
6045 declare float @llvm.sin.f32(float %Val)
6046 declare double @llvm.sin.f64(double %Val)
6047 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6048 declare fp128 @llvm.sin.f128(fp128 %Val)
6049 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6053 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
6056 <p>The argument and return value are floating point numbers of the same
6060 <p>This function returns the sine of the specified operand, returning the same
6061 values as the libm <tt>sin</tt> functions would, and handles error conditions
6062 in the same way.</p>
6066 <!-- _______________________________________________________________________ -->
6067 <div class="doc_subsubsection">
6068 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
6071 <div class="doc_text">
6074 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
6075 floating point or vector of floating point type. Not all targets support all
6079 declare float @llvm.cos.f32(float %Val)
6080 declare double @llvm.cos.f64(double %Val)
6081 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6082 declare fp128 @llvm.cos.f128(fp128 %Val)
6083 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6087 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
6090 <p>The argument and return value are floating point numbers of the same
6094 <p>This function returns the cosine of the specified operand, returning the same
6095 values as the libm <tt>cos</tt> functions would, and handles error conditions
6096 in the same way.</p>
6100 <!-- _______________________________________________________________________ -->
6101 <div class="doc_subsubsection">
6102 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
6105 <div class="doc_text">
6108 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
6109 floating point or vector of floating point type. Not all targets support all
6113 declare float @llvm.pow.f32(float %Val, float %Power)
6114 declare double @llvm.pow.f64(double %Val, double %Power)
6115 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6116 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6117 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6121 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
6122 specified (positive or negative) power.</p>
6125 <p>The second argument is a floating point power, and the first is a value to
6126 raise to that power.</p>
6129 <p>This function returns the first value raised to the second power, returning
6130 the same values as the libm <tt>pow</tt> functions would, and handles error
6131 conditions in the same way.</p>
6135 <!-- ======================================================================= -->
6136 <div class="doc_subsection">
6137 <a name="int_manip">Bit Manipulation Intrinsics</a>
6140 <div class="doc_text">
6142 <p>LLVM provides intrinsics for a few important bit manipulation operations.
6143 These allow efficient code generation for some algorithms.</p>
6147 <!-- _______________________________________________________________________ -->
6148 <div class="doc_subsubsection">
6149 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
6152 <div class="doc_text">
6155 <p>This is an overloaded intrinsic function. You can use bswap on any integer
6156 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
6159 declare i16 @llvm.bswap.i16(i16 <id>)
6160 declare i32 @llvm.bswap.i32(i32 <id>)
6161 declare i64 @llvm.bswap.i64(i64 <id>)
6165 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
6166 values with an even number of bytes (positive multiple of 16 bits). These
6167 are useful for performing operations on data that is not in the target's
6168 native byte order.</p>
6171 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
6172 and low byte of the input i16 swapped. Similarly,
6173 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
6174 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
6175 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
6176 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
6177 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
6178 more, respectively).</p>
6182 <!-- _______________________________________________________________________ -->
6183 <div class="doc_subsubsection">
6184 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
6187 <div class="doc_text">
6190 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6191 width. Not all targets support all bit widths however.</p>
6194 declare i8 @llvm.ctpop.i8(i8 <src>)
6195 declare i16 @llvm.ctpop.i16(i16 <src>)
6196 declare i32 @llvm.ctpop.i32(i32 <src>)
6197 declare i64 @llvm.ctpop.i64(i64 <src>)
6198 declare i256 @llvm.ctpop.i256(i256 <src>)
6202 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
6206 <p>The only argument is the value to be counted. The argument may be of any
6207 integer type. The return type must match the argument type.</p>
6210 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.</p>
6214 <!-- _______________________________________________________________________ -->
6215 <div class="doc_subsubsection">
6216 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
6219 <div class="doc_text">
6222 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
6223 integer bit width. Not all targets support all bit widths however.</p>
6226 declare i8 @llvm.ctlz.i8 (i8 <src>)
6227 declare i16 @llvm.ctlz.i16(i16 <src>)
6228 declare i32 @llvm.ctlz.i32(i32 <src>)
6229 declare i64 @llvm.ctlz.i64(i64 <src>)
6230 declare i256 @llvm.ctlz.i256(i256 <src>)
6234 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
6235 leading zeros in a variable.</p>
6238 <p>The only argument is the value to be counted. The argument may be of any
6239 integer type. The return type must match the argument type.</p>
6242 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
6243 zeros in a variable. If the src == 0 then the result is the size in bits of
6244 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
6248 <!-- _______________________________________________________________________ -->
6249 <div class="doc_subsubsection">
6250 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
6253 <div class="doc_text">
6256 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
6257 integer bit width. Not all targets support all bit widths however.</p>
6260 declare i8 @llvm.cttz.i8 (i8 <src>)
6261 declare i16 @llvm.cttz.i16(i16 <src>)
6262 declare i32 @llvm.cttz.i32(i32 <src>)
6263 declare i64 @llvm.cttz.i64(i64 <src>)
6264 declare i256 @llvm.cttz.i256(i256 <src>)
6268 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
6272 <p>The only argument is the value to be counted. The argument may be of any
6273 integer type. The return type must match the argument type.</p>
6276 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
6277 zeros in a variable. If the src == 0 then the result is the size in bits of
6278 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
6282 <!-- ======================================================================= -->
6283 <div class="doc_subsection">
6284 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
6287 <div class="doc_text">
6289 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
6293 <!-- _______________________________________________________________________ -->
6294 <div class="doc_subsubsection">
6295 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
6298 <div class="doc_text">
6301 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
6302 on any integer bit width.</p>
6305 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6306 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6307 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6311 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6312 a signed addition of the two arguments, and indicate whether an overflow
6313 occurred during the signed summation.</p>
6316 <p>The arguments (%a and %b) and the first element of the result structure may
6317 be of integer types of any bit width, but they must have the same bit
6318 width. The second element of the result structure must be of
6319 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6320 undergo signed addition.</p>
6323 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6324 a signed addition of the two variables. They return a structure — the
6325 first element of which is the signed summation, and the second element of
6326 which is a bit specifying if the signed summation resulted in an
6331 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6332 %sum = extractvalue {i32, i1} %res, 0
6333 %obit = extractvalue {i32, i1} %res, 1
6334 br i1 %obit, label %overflow, label %normal
6339 <!-- _______________________________________________________________________ -->
6340 <div class="doc_subsubsection">
6341 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
6344 <div class="doc_text">
6347 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
6348 on any integer bit width.</p>
6351 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6352 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6353 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6357 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6358 an unsigned addition of the two arguments, and indicate whether a carry
6359 occurred during the unsigned summation.</p>
6362 <p>The arguments (%a and %b) and the first element of the result structure may
6363 be of integer types of any bit width, but they must have the same bit
6364 width. The second element of the result structure must be of
6365 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6366 undergo unsigned addition.</p>
6369 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6370 an unsigned addition of the two arguments. They return a structure —
6371 the first element of which is the sum, and the second element of which is a
6372 bit specifying if the unsigned summation resulted in a carry.</p>
6376 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6377 %sum = extractvalue {i32, i1} %res, 0
6378 %obit = extractvalue {i32, i1} %res, 1
6379 br i1 %obit, label %carry, label %normal
6384 <!-- _______________________________________________________________________ -->
6385 <div class="doc_subsubsection">
6386 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6389 <div class="doc_text">
6392 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6393 on any integer bit width.</p>
6396 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6397 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6398 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6402 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6403 a signed subtraction of the two arguments, and indicate whether an overflow
6404 occurred during the signed subtraction.</p>
6407 <p>The arguments (%a and %b) and the first element of the result structure may
6408 be of integer types of any bit width, but they must have the same bit
6409 width. The second element of the result structure must be of
6410 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6411 undergo signed subtraction.</p>
6414 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6415 a signed subtraction of the two arguments. They return a structure —
6416 the first element of which is the subtraction, and the second element of
6417 which is a bit specifying if the signed subtraction resulted in an
6422 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6423 %sum = extractvalue {i32, i1} %res, 0
6424 %obit = extractvalue {i32, i1} %res, 1
6425 br i1 %obit, label %overflow, label %normal
6430 <!-- _______________________________________________________________________ -->
6431 <div class="doc_subsubsection">
6432 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6435 <div class="doc_text">
6438 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6439 on any integer bit width.</p>
6442 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6443 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6444 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6448 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6449 an unsigned subtraction of the two arguments, and indicate whether an
6450 overflow occurred during the unsigned subtraction.</p>
6453 <p>The arguments (%a and %b) and the first element of the result structure may
6454 be of integer types of any bit width, but they must have the same bit
6455 width. The second element of the result structure must be of
6456 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6457 undergo unsigned subtraction.</p>
6460 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6461 an unsigned subtraction of the two arguments. They return a structure —
6462 the first element of which is the subtraction, and the second element of
6463 which is a bit specifying if the unsigned subtraction resulted in an
6468 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6469 %sum = extractvalue {i32, i1} %res, 0
6470 %obit = extractvalue {i32, i1} %res, 1
6471 br i1 %obit, label %overflow, label %normal
6476 <!-- _______________________________________________________________________ -->
6477 <div class="doc_subsubsection">
6478 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6481 <div class="doc_text">
6484 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6485 on any integer bit width.</p>
6488 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6489 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6490 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6495 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6496 a signed multiplication of the two arguments, and indicate whether an
6497 overflow occurred during the signed multiplication.</p>
6500 <p>The arguments (%a and %b) and the first element of the result structure may
6501 be of integer types of any bit width, but they must have the same bit
6502 width. The second element of the result structure must be of
6503 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6504 undergo signed multiplication.</p>
6507 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6508 a signed multiplication of the two arguments. They return a structure —
6509 the first element of which is the multiplication, and the second element of
6510 which is a bit specifying if the signed multiplication resulted in an
6515 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6516 %sum = extractvalue {i32, i1} %res, 0
6517 %obit = extractvalue {i32, i1} %res, 1
6518 br i1 %obit, label %overflow, label %normal
6523 <!-- _______________________________________________________________________ -->
6524 <div class="doc_subsubsection">
6525 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6528 <div class="doc_text">
6531 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6532 on any integer bit width.</p>
6535 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6536 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6537 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6541 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6542 a unsigned multiplication of the two arguments, and indicate whether an
6543 overflow occurred during the unsigned multiplication.</p>
6546 <p>The arguments (%a and %b) and the first element of the result structure may
6547 be of integer types of any bit width, but they must have the same bit
6548 width. The second element of the result structure must be of
6549 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6550 undergo unsigned multiplication.</p>
6553 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6554 an unsigned multiplication of the two arguments. They return a structure
6555 — the first element of which is the multiplication, and the second
6556 element of which is a bit specifying if the unsigned multiplication resulted
6561 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6562 %sum = extractvalue {i32, i1} %res, 0
6563 %obit = extractvalue {i32, i1} %res, 1
6564 br i1 %obit, label %overflow, label %normal
6569 <!-- ======================================================================= -->
6570 <div class="doc_subsection">
6571 <a name="int_debugger">Debugger Intrinsics</a>
6574 <div class="doc_text">
6576 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
6577 prefix), are described in
6578 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
6579 Level Debugging</a> document.</p>
6583 <!-- ======================================================================= -->
6584 <div class="doc_subsection">
6585 <a name="int_eh">Exception Handling Intrinsics</a>
6588 <div class="doc_text">
6590 <p>The LLVM exception handling intrinsics (which all start with
6591 <tt>llvm.eh.</tt> prefix), are described in
6592 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6593 Handling</a> document.</p>
6597 <!-- ======================================================================= -->
6598 <div class="doc_subsection">
6599 <a name="int_trampoline">Trampoline Intrinsic</a>
6602 <div class="doc_text">
6604 <p>This intrinsic makes it possible to excise one parameter, marked with
6605 the <tt>nest</tt> attribute, from a function. The result is a callable
6606 function pointer lacking the nest parameter - the caller does not need to
6607 provide a value for it. Instead, the value to use is stored in advance in a
6608 "trampoline", a block of memory usually allocated on the stack, which also
6609 contains code to splice the nest value into the argument list. This is used
6610 to implement the GCC nested function address extension.</p>
6612 <p>For example, if the function is
6613 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6614 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
6617 <div class="doc_code">
6619 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6620 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6621 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6622 %fp = bitcast i8* %p to i32 (i32, i32)*
6626 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6627 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6631 <!-- _______________________________________________________________________ -->
6632 <div class="doc_subsubsection">
6633 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6636 <div class="doc_text">
6640 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6644 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
6645 function pointer suitable for executing it.</p>
6648 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6649 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
6650 sufficiently aligned block of memory; this memory is written to by the
6651 intrinsic. Note that the size and the alignment are target-specific - LLVM
6652 currently provides no portable way of determining them, so a front-end that
6653 generates this intrinsic needs to have some target-specific knowledge.
6654 The <tt>func</tt> argument must hold a function bitcast to
6655 an <tt>i8*</tt>.</p>
6658 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
6659 dependent code, turning it into a function. A pointer to this function is
6660 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
6661 function pointer type</a> before being called. The new function's signature
6662 is the same as that of <tt>func</tt> with any arguments marked with
6663 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
6664 is allowed, and it must be of pointer type. Calling the new function is
6665 equivalent to calling <tt>func</tt> with the same argument list, but
6666 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
6667 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
6668 by <tt>tramp</tt> is modified, then the effect of any later call to the
6669 returned function pointer is undefined.</p>
6673 <!-- ======================================================================= -->
6674 <div class="doc_subsection">
6675 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6678 <div class="doc_text">
6680 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
6681 hardware constructs for atomic operations and memory synchronization. This
6682 provides an interface to the hardware, not an interface to the programmer. It
6683 is aimed at a low enough level to allow any programming models or APIs
6684 (Application Programming Interfaces) which need atomic behaviors to map
6685 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6686 hardware provides a "universal IR" for source languages, it also provides a
6687 starting point for developing a "universal" atomic operation and
6688 synchronization IR.</p>
6690 <p>These do <em>not</em> form an API such as high-level threading libraries,
6691 software transaction memory systems, atomic primitives, and intrinsic
6692 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6693 application libraries. The hardware interface provided by LLVM should allow
6694 a clean implementation of all of these APIs and parallel programming models.
6695 No one model or paradigm should be selected above others unless the hardware
6696 itself ubiquitously does so.</p>
6700 <!-- _______________________________________________________________________ -->
6701 <div class="doc_subsubsection">
6702 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6704 <div class="doc_text">
6707 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device> )
6711 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6712 specific pairs of memory access types.</p>
6715 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6716 The first four arguments enables a specific barrier as listed below. The
6717 fith argument specifies that the barrier applies to io or device or uncached
6721 <li><tt>ll</tt>: load-load barrier</li>
6722 <li><tt>ls</tt>: load-store barrier</li>
6723 <li><tt>sl</tt>: store-load barrier</li>
6724 <li><tt>ss</tt>: store-store barrier</li>
6725 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6729 <p>This intrinsic causes the system to enforce some ordering constraints upon
6730 the loads and stores of the program. This barrier does not
6731 indicate <em>when</em> any events will occur, it only enforces
6732 an <em>order</em> in which they occur. For any of the specified pairs of load
6733 and store operations (f.ex. load-load, or store-load), all of the first
6734 operations preceding the barrier will complete before any of the second
6735 operations succeeding the barrier begin. Specifically the semantics for each
6736 pairing is as follows:</p>
6739 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6740 after the barrier begins.</li>
6741 <li><tt>ls</tt>: All loads before the barrier must complete before any
6742 store after the barrier begins.</li>
6743 <li><tt>ss</tt>: All stores before the barrier must complete before any
6744 store after the barrier begins.</li>
6745 <li><tt>sl</tt>: All stores before the barrier must complete before any
6746 load after the barrier begins.</li>
6749 <p>These semantics are applied with a logical "and" behavior when more than one
6750 is enabled in a single memory barrier intrinsic.</p>
6752 <p>Backends may implement stronger barriers than those requested when they do
6753 not support as fine grained a barrier as requested. Some architectures do
6754 not need all types of barriers and on such architectures, these become
6759 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6760 %ptr = bitcast i8* %mallocP to i32*
6763 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6764 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6765 <i>; guarantee the above finishes</i>
6766 store i32 8, %ptr <i>; before this begins</i>
6771 <!-- _______________________________________________________________________ -->
6772 <div class="doc_subsubsection">
6773 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6776 <div class="doc_text">
6779 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6780 any integer bit width and for different address spaces. Not all targets
6781 support all bit widths however.</p>
6784 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6785 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6786 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6787 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6791 <p>This loads a value in memory and compares it to a given value. If they are
6792 equal, it stores a new value into the memory.</p>
6795 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
6796 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6797 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6798 this integer type. While any bit width integer may be used, targets may only
6799 lower representations they support in hardware.</p>
6802 <p>This entire intrinsic must be executed atomically. It first loads the value
6803 in memory pointed to by <tt>ptr</tt> and compares it with the
6804 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
6805 memory. The loaded value is yielded in all cases. This provides the
6806 equivalent of an atomic compare-and-swap operation within the SSA
6811 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6812 %ptr = bitcast i8* %mallocP to i32*
6815 %val1 = add i32 4, 4
6816 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6817 <i>; yields {i32}:result1 = 4</i>
6818 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6819 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6821 %val2 = add i32 1, 1
6822 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6823 <i>; yields {i32}:result2 = 8</i>
6824 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6826 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6831 <!-- _______________________________________________________________________ -->
6832 <div class="doc_subsubsection">
6833 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6835 <div class="doc_text">
6838 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6839 integer bit width. Not all targets support all bit widths however.</p>
6842 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6843 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6844 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6845 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6849 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6850 the value from memory. It then stores the value in <tt>val</tt> in the memory
6851 at <tt>ptr</tt>.</p>
6854 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
6855 the <tt>val</tt> argument and the result must be integers of the same bit
6856 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6857 integer type. The targets may only lower integer representations they
6861 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6862 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6863 equivalent of an atomic swap operation within the SSA framework.</p>
6867 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6868 %ptr = bitcast i8* %mallocP to i32*
6871 %val1 = add i32 4, 4
6872 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6873 <i>; yields {i32}:result1 = 4</i>
6874 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6875 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6877 %val2 = add i32 1, 1
6878 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6879 <i>; yields {i32}:result2 = 8</i>
6881 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6882 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6887 <!-- _______________________________________________________________________ -->
6888 <div class="doc_subsubsection">
6889 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6893 <div class="doc_text">
6896 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
6897 any integer bit width. Not all targets support all bit widths however.</p>
6900 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6901 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6902 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6903 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6907 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
6908 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6911 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6912 and the second an integer value. The result is also an integer value. These
6913 integer types can have any bit width, but they must all have the same bit
6914 width. The targets may only lower integer representations they support.</p>
6917 <p>This intrinsic does a series of operations atomically. It first loads the
6918 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6919 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
6923 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6924 %ptr = bitcast i8* %mallocP to i32*
6926 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6927 <i>; yields {i32}:result1 = 4</i>
6928 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6929 <i>; yields {i32}:result2 = 8</i>
6930 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6931 <i>; yields {i32}:result3 = 10</i>
6932 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6937 <!-- _______________________________________________________________________ -->
6938 <div class="doc_subsubsection">
6939 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6943 <div class="doc_text">
6946 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6947 any integer bit width and for different address spaces. Not all targets
6948 support all bit widths however.</p>
6951 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6952 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6953 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6954 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6958 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6959 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6962 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6963 and the second an integer value. The result is also an integer value. These
6964 integer types can have any bit width, but they must all have the same bit
6965 width. The targets may only lower integer representations they support.</p>
6968 <p>This intrinsic does a series of operations atomically. It first loads the
6969 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6970 result to <tt>ptr</tt>. It yields the original value stored
6971 at <tt>ptr</tt>.</p>
6975 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6976 %ptr = bitcast i8* %mallocP to i32*
6978 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6979 <i>; yields {i32}:result1 = 8</i>
6980 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6981 <i>; yields {i32}:result2 = 4</i>
6982 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6983 <i>; yields {i32}:result3 = 2</i>
6984 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6989 <!-- _______________________________________________________________________ -->
6990 <div class="doc_subsubsection">
6991 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6992 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6993 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6994 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6997 <div class="doc_text">
7000 <p>These are overloaded intrinsics. You can
7001 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
7002 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
7003 bit width and for different address spaces. Not all targets support all bit
7007 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
7008 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
7009 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
7010 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
7014 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
7015 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
7016 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
7017 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
7021 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
7022 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
7023 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
7024 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
7028 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
7029 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
7030 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
7031 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
7035 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
7036 the value stored in memory at <tt>ptr</tt>. It yields the original value
7037 at <tt>ptr</tt>.</p>
7040 <p>These intrinsics take two arguments, the first a pointer to an integer value
7041 and the second an integer value. The result is also an integer value. These
7042 integer types can have any bit width, but they must all have the same bit
7043 width. The targets may only lower integer representations they support.</p>
7046 <p>These intrinsics does a series of operations atomically. They first load the
7047 value stored at <tt>ptr</tt>. They then do the bitwise
7048 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
7049 original value stored at <tt>ptr</tt>.</p>
7053 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7054 %ptr = bitcast i8* %mallocP to i32*
7055 store i32 0x0F0F, %ptr
7056 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
7057 <i>; yields {i32}:result0 = 0x0F0F</i>
7058 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
7059 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
7060 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
7061 <i>; yields {i32}:result2 = 0xF0</i>
7062 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
7063 <i>; yields {i32}:result3 = FF</i>
7064 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
7069 <!-- _______________________________________________________________________ -->
7070 <div class="doc_subsubsection">
7071 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
7072 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
7073 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
7074 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
7077 <div class="doc_text">
7080 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
7081 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
7082 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
7083 address spaces. Not all targets support all bit widths however.</p>
7086 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
7087 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
7088 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
7089 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
7093 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
7094 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
7095 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
7096 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
7100 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
7101 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
7102 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
7103 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
7107 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
7108 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
7109 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
7110 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
7114 <p>These intrinsics takes the signed or unsigned minimum or maximum of
7115 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
7116 original value at <tt>ptr</tt>.</p>
7119 <p>These intrinsics take two arguments, the first a pointer to an integer value
7120 and the second an integer value. The result is also an integer value. These
7121 integer types can have any bit width, but they must all have the same bit
7122 width. The targets may only lower integer representations they support.</p>
7125 <p>These intrinsics does a series of operations atomically. They first load the
7126 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
7127 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
7128 yield the original value stored at <tt>ptr</tt>.</p>
7132 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7133 %ptr = bitcast i8* %mallocP to i32*
7135 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
7136 <i>; yields {i32}:result0 = 7</i>
7137 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
7138 <i>; yields {i32}:result1 = -2</i>
7139 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
7140 <i>; yields {i32}:result2 = 8</i>
7141 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
7142 <i>; yields {i32}:result3 = 8</i>
7143 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
7149 <!-- ======================================================================= -->
7150 <div class="doc_subsection">
7151 <a name="int_memorymarkers">Memory Use Markers</a>
7154 <div class="doc_text">
7156 <p>This class of intrinsics exists to information about the lifetime of memory
7157 objects and ranges where variables are immutable.</p>
7161 <!-- _______________________________________________________________________ -->
7162 <div class="doc_subsubsection">
7163 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
7166 <div class="doc_text">
7170 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
7174 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
7175 object's lifetime.</p>
7178 <p>The first argument is a constant integer representing the size of the
7179 object, or -1 if it is variable sized. The second argument is a pointer to
7183 <p>This intrinsic indicates that before this point in the code, the value of the
7184 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7185 never be used and has an undefined value. A load from the pointer that
7186 precedes this intrinsic can be replaced with
7187 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
7191 <!-- _______________________________________________________________________ -->
7192 <div class="doc_subsubsection">
7193 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
7196 <div class="doc_text">
7200 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
7204 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
7205 object's lifetime.</p>
7208 <p>The first argument is a constant integer representing the size of the
7209 object, or -1 if it is variable sized. The second argument is a pointer to
7213 <p>This intrinsic indicates that after this point in the code, the value of the
7214 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7215 never be used and has an undefined value. Any stores into the memory object
7216 following this intrinsic may be removed as dead.
7220 <!-- _______________________________________________________________________ -->
7221 <div class="doc_subsubsection">
7222 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
7225 <div class="doc_text">
7229 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>) readonly
7233 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
7234 a memory object will not change.</p>
7237 <p>The first argument is a constant integer representing the size of the
7238 object, or -1 if it is variable sized. The second argument is a pointer to
7242 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
7243 the return value, the referenced memory location is constant and
7248 <!-- _______________________________________________________________________ -->
7249 <div class="doc_subsubsection">
7250 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
7253 <div class="doc_text">
7257 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7261 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
7262 a memory object are mutable.</p>
7265 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
7266 The second argument is a constant integer representing the size of the
7267 object, or -1 if it is variable sized and the third argument is a pointer
7271 <p>This intrinsic indicates that the memory is mutable again.</p>
7275 <!-- ======================================================================= -->
7276 <div class="doc_subsection">
7277 <a name="int_general">General Intrinsics</a>
7280 <div class="doc_text">
7282 <p>This class of intrinsics is designed to be generic and has no specific
7287 <!-- _______________________________________________________________________ -->
7288 <div class="doc_subsubsection">
7289 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
7292 <div class="doc_text">
7296 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
7300 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
7303 <p>The first argument is a pointer to a value, the second is a pointer to a
7304 global string, the third is a pointer to a global string which is the source
7305 file name, and the last argument is the line number.</p>
7308 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7309 This can be useful for special purpose optimizations that want to look for
7310 these annotations. These have no other defined use, they are ignored by code
7311 generation and optimization.</p>
7315 <!-- _______________________________________________________________________ -->
7316 <div class="doc_subsubsection">
7317 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
7320 <div class="doc_text">
7323 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
7324 any integer bit width.</p>
7327 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
7328 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
7329 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
7330 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
7331 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
7335 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
7338 <p>The first argument is an integer value (result of some expression), the
7339 second is a pointer to a global string, the third is a pointer to a global
7340 string which is the source file name, and the last argument is the line
7341 number. It returns the value of the first argument.</p>
7344 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7345 arbitrary strings. This can be useful for special purpose optimizations that
7346 want to look for these annotations. These have no other defined use, they
7347 are ignored by code generation and optimization.</p>
7351 <!-- _______________________________________________________________________ -->
7352 <div class="doc_subsubsection">
7353 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
7356 <div class="doc_text">
7360 declare void @llvm.trap()
7364 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
7370 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7371 target does not have a trap instruction, this intrinsic will be lowered to
7372 the call of the <tt>abort()</tt> function.</p>
7376 <!-- _______________________________________________________________________ -->
7377 <div class="doc_subsubsection">
7378 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
7381 <div class="doc_text">
7385 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
7389 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
7390 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
7391 ensure that it is placed on the stack before local variables.</p>
7394 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
7395 arguments. The first argument is the value loaded from the stack
7396 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
7397 that has enough space to hold the value of the guard.</p>
7400 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7401 the <tt>AllocaInst</tt> stack slot to be before local variables on the
7402 stack. This is to ensure that if a local variable on the stack is
7403 overwritten, it will destroy the value of the guard. When the function exits,
7404 the guard on the stack is checked against the original guard. If they're
7405 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
7410 <!-- _______________________________________________________________________ -->
7411 <div class="doc_subsubsection">
7412 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
7415 <div class="doc_text">
7419 declare i32 @llvm.objectsize.i32( i8* <object>, i1 <type> )
7420 declare i64 @llvm.objectsize.i64( i8* <object>, i1 <type> )
7424 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information
7425 to the optimizers to discover at compile time either a) when an
7426 operation like memcpy will either overflow a buffer that corresponds to
7427 an object, or b) to determine that a runtime check for overflow isn't
7428 necessary. An object in this context means an allocation of a
7429 specific class, structure, array, or other object.</p>
7432 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
7433 argument is a pointer to or into the <tt>object</tt>. The second argument
7434 is a boolean 0 or 1. This argument determines whether you want the
7435 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
7436 1, variables are not allowed.</p>
7439 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
7440 representing the size of the object concerned or <tt>i32/i64 -1 or 0</tt>
7441 (depending on the <tt>type</tt> argument if the size cannot be determined
7442 at compile time.</p>
7446 <!-- *********************************************************************** -->
7449 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
7450 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
7451 <a href="http://validator.w3.org/check/referer"><img
7452 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
7454 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
7455 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
7456 Last modified: $Date$