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_fp16">Half Precision Floating Point Intrinsics</a>
260 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
261 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
264 <li><a href="#int_debugger">Debugger intrinsics</a></li>
265 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
266 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
268 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
271 <li><a href="#int_atomics">Atomic intrinsics</a>
273 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
274 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
275 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
276 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
277 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
278 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
279 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
280 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
281 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
282 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
283 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
284 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
285 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
288 <li><a href="#int_memorymarkers">Memory Use Markers</a>
290 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
291 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
292 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
293 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
296 <li><a href="#int_general">General intrinsics</a>
298 <li><a href="#int_var_annotation">
299 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
300 <li><a href="#int_annotation">
301 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
302 <li><a href="#int_trap">
303 '<tt>llvm.trap</tt>' Intrinsic</a></li>
304 <li><a href="#int_stackprotector">
305 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
306 <li><a href="#int_objectsize">
307 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
314 <div class="doc_author">
315 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
316 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
319 <!-- *********************************************************************** -->
320 <div class="doc_section"> <a name="abstract">Abstract </a></div>
321 <!-- *********************************************************************** -->
323 <div class="doc_text">
325 <p>This document is a reference manual for the LLVM assembly language. LLVM is
326 a Static Single Assignment (SSA) based representation that provides type
327 safety, low-level operations, flexibility, and the capability of representing
328 'all' high-level languages cleanly. It is the common code representation
329 used throughout all phases of the LLVM compilation strategy.</p>
333 <!-- *********************************************************************** -->
334 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
335 <!-- *********************************************************************** -->
337 <div class="doc_text">
339 <p>The LLVM code representation is designed to be used in three different forms:
340 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
341 for fast loading by a Just-In-Time compiler), and as a human readable
342 assembly language representation. This allows LLVM to provide a powerful
343 intermediate representation for efficient compiler transformations and
344 analysis, while providing a natural means to debug and visualize the
345 transformations. The three different forms of LLVM are all equivalent. This
346 document describes the human readable representation and notation.</p>
348 <p>The LLVM representation aims to be light-weight and low-level while being
349 expressive, typed, and extensible at the same time. It aims to be a
350 "universal IR" of sorts, by being at a low enough level that high-level ideas
351 may be cleanly mapped to it (similar to how microprocessors are "universal
352 IR's", allowing many source languages to be mapped to them). By providing
353 type information, LLVM can be used as the target of optimizations: for
354 example, through pointer analysis, it can be proven that a C automatic
355 variable is never accessed outside of the current function, allowing it to
356 be promoted to a simple SSA value instead of a memory location.</p>
360 <!-- _______________________________________________________________________ -->
361 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
363 <div class="doc_text">
365 <p>It is important to note that this document describes 'well formed' LLVM
366 assembly language. There is a difference between what the parser accepts and
367 what is considered 'well formed'. For example, the following instruction is
368 syntactically okay, but not well formed:</p>
370 <div class="doc_code">
372 %x = <a href="#i_add">add</a> i32 1, %x
376 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
377 LLVM infrastructure provides a verification pass that may be used to verify
378 that an LLVM module is well formed. This pass is automatically run by the
379 parser after parsing input assembly and by the optimizer before it outputs
380 bitcode. The violations pointed out by the verifier pass indicate bugs in
381 transformation passes or input to the parser.</p>
385 <!-- Describe the typesetting conventions here. -->
387 <!-- *********************************************************************** -->
388 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
389 <!-- *********************************************************************** -->
391 <div class="doc_text">
393 <p>LLVM identifiers come in two basic types: global and local. Global
394 identifiers (functions, global variables) begin with the <tt>'@'</tt>
395 character. Local identifiers (register names, types) begin with
396 the <tt>'%'</tt> character. Additionally, there are three different formats
397 for identifiers, for different purposes:</p>
400 <li>Named values are represented as a string of characters with their prefix.
401 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
402 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
403 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
404 other characters in their names can be surrounded with quotes. Special
405 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
406 ASCII code for the character in hexadecimal. In this way, any character
407 can be used in a name value, even quotes themselves.</li>
409 <li>Unnamed values are represented as an unsigned numeric value with their
410 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
412 <li>Constants, which are described in a <a href="#constants">section about
413 constants</a>, below.</li>
416 <p>LLVM requires that values start with a prefix for two reasons: Compilers
417 don't need to worry about name clashes with reserved words, and the set of
418 reserved words may be expanded in the future without penalty. Additionally,
419 unnamed identifiers allow a compiler to quickly come up with a temporary
420 variable without having to avoid symbol table conflicts.</p>
422 <p>Reserved words in LLVM are very similar to reserved words in other
423 languages. There are keywords for different opcodes
424 ('<tt><a href="#i_add">add</a></tt>',
425 '<tt><a href="#i_bitcast">bitcast</a></tt>',
426 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
427 ('<tt><a href="#t_void">void</a></tt>',
428 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
429 reserved words cannot conflict with variable names, because none of them
430 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
432 <p>Here is an example of LLVM code to multiply the integer variable
433 '<tt>%X</tt>' by 8:</p>
437 <div class="doc_code">
439 %result = <a href="#i_mul">mul</a> i32 %X, 8
443 <p>After strength reduction:</p>
445 <div class="doc_code">
447 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
451 <p>And the hard way:</p>
453 <div class="doc_code">
455 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
456 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
457 %result = <a href="#i_add">add</a> i32 %1, %1
461 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
462 lexical features of LLVM:</p>
465 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
468 <li>Unnamed temporaries are created when the result of a computation is not
469 assigned to a named value.</li>
471 <li>Unnamed temporaries are numbered sequentially</li>
474 <p>It also shows a convention that we follow in this document. When
475 demonstrating instructions, we will follow an instruction with a comment that
476 defines the type and name of value produced. Comments are shown in italic
481 <!-- *********************************************************************** -->
482 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
483 <!-- *********************************************************************** -->
485 <!-- ======================================================================= -->
486 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
489 <div class="doc_text">
491 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
492 of the input programs. Each module consists of functions, global variables,
493 and symbol table entries. Modules may be combined together with the LLVM
494 linker, which merges function (and global variable) definitions, resolves
495 forward declarations, and merges symbol table entries. Here is an example of
496 the "hello world" module:</p>
498 <div class="doc_code">
500 <i>; Declare the string constant as a global constant.</i>
501 <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>
503 <i>; External declaration of the puts function</i>
504 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
506 <i>; Definition of main function</i>
507 define i32 @main() { <i>; i32()* </i>
508 <i>; Convert [13 x i8]* to i8 *...</i>
509 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
511 <i>; Call puts function to write out the string to stdout.</i>
512 <a href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
513 <a href="#i_ret">ret</a> i32 0<br>}
515 <i>; Named metadata</i>
516 !1 = metadata !{i32 41}
521 <p>This example is made up of a <a href="#globalvars">global variable</a> named
522 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
523 a <a href="#functionstructure">function definition</a> for
524 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
527 <p>In general, a module is made up of a list of global values, where both
528 functions and global variables are global values. Global values are
529 represented by a pointer to a memory location (in this case, a pointer to an
530 array of char, and a pointer to a function), and have one of the
531 following <a href="#linkage">linkage types</a>.</p>
535 <!-- ======================================================================= -->
536 <div class="doc_subsection">
537 <a name="linkage">Linkage Types</a>
540 <div class="doc_text">
542 <p>All Global Variables and Functions have one of the following types of
546 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
547 <dd>Global values with private linkage are only directly accessible by objects
548 in the current module. In particular, linking code into a module with an
549 private global value may cause the private to be renamed as necessary to
550 avoid collisions. Because the symbol is private to the module, all
551 references can be updated. This doesn't show up in any symbol table in the
554 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
555 <dd>Similar to private, but the symbol is passed through the assembler and
556 removed by the linker after evaluation. Note that (unlike private
557 symbols) linker_private symbols are subject to coalescing by the linker:
558 weak symbols get merged and redefinitions are rejected. However, unlike
559 normal strong symbols, they are removed by the linker from the final
560 linked image (executable or dynamic library).</dd>
562 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
563 <dd>Similar to private, but the value shows as a local symbol
564 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
565 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
567 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
568 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
569 into the object file corresponding to the LLVM module. They exist to
570 allow inlining and other optimizations to take place given knowledge of
571 the definition of the global, which is known to be somewhere outside the
572 module. Globals with <tt>available_externally</tt> linkage are allowed to
573 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
574 This linkage type is only allowed on definitions, not declarations.</dd>
576 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
577 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
578 the same name when linkage occurs. This can be used to implement
579 some forms of inline functions, templates, or other code which must be
580 generated in each translation unit that uses it, but where the body may
581 be overridden with a more definitive definition later. Unreferenced
582 <tt>linkonce</tt> globals are allowed to be discarded. Note that
583 <tt>linkonce</tt> linkage does not actually allow the optimizer to
584 inline the body of this function into callers because it doesn't know if
585 this definition of the function is the definitive definition within the
586 program or whether it will be overridden by a stronger definition.
587 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
590 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
591 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
592 <tt>linkonce</tt> linkage, except that unreferenced globals with
593 <tt>weak</tt> linkage may not be discarded. This is used for globals that
594 are declared "weak" in C source code.</dd>
596 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
597 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
598 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
600 Symbols with "<tt>common</tt>" linkage are merged in the same way as
601 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
602 <tt>common</tt> symbols may not have an explicit section,
603 must have a zero initializer, and may not be marked '<a
604 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
605 have common linkage.</dd>
608 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
609 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
610 pointer to array type. When two global variables with appending linkage
611 are linked together, the two global arrays are appended together. This is
612 the LLVM, typesafe, equivalent of having the system linker append together
613 "sections" with identical names when .o files are linked.</dd>
615 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
616 <dd>The semantics of this linkage follow the ELF object file model: the symbol
617 is weak until linked, if not linked, the symbol becomes null instead of
618 being an undefined reference.</dd>
620 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
621 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
622 <dd>Some languages allow differing globals to be merged, such as two functions
623 with different semantics. Other languages, such as <tt>C++</tt>, ensure
624 that only equivalent globals are ever merged (the "one definition rule" -
625 "ODR"). Such languages can use the <tt>linkonce_odr</tt>
626 and <tt>weak_odr</tt> linkage types to indicate that the global will only
627 be merged with equivalent globals. These linkage types are otherwise the
628 same as their non-<tt>odr</tt> versions.</dd>
630 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
631 <dd>If none of the above identifiers are used, the global is externally
632 visible, meaning that it participates in linkage and can be used to
633 resolve external symbol references.</dd>
636 <p>The next two types of linkage are targeted for Microsoft Windows platform
637 only. They are designed to support importing (exporting) symbols from (to)
638 DLLs (Dynamic Link Libraries).</p>
641 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
642 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
643 or variable via a global pointer to a pointer that is set up by the DLL
644 exporting the symbol. On Microsoft Windows targets, the pointer name is
645 formed by combining <code>__imp_</code> and the function or variable
648 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
649 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
650 pointer to a pointer in a DLL, so that it can be referenced with the
651 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
652 name is formed by combining <code>__imp_</code> and the function or
656 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
657 another module defined a "<tt>.LC0</tt>" variable and was linked with this
658 one, one of the two would be renamed, preventing a collision. Since
659 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
660 declarations), they are accessible outside of the current module.</p>
662 <p>It is illegal for a function <i>declaration</i> to have any linkage type
663 other than "externally visible", <tt>dllimport</tt>
664 or <tt>extern_weak</tt>.</p>
666 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
667 or <tt>weak_odr</tt> linkages.</p>
671 <!-- ======================================================================= -->
672 <div class="doc_subsection">
673 <a name="callingconv">Calling Conventions</a>
676 <div class="doc_text">
678 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
679 and <a href="#i_invoke">invokes</a> can all have an optional calling
680 convention specified for the call. The calling convention of any pair of
681 dynamic caller/callee must match, or the behavior of the program is
682 undefined. The following calling conventions are supported by LLVM, and more
683 may be added in the future:</p>
686 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
687 <dd>This calling convention (the default if no other calling convention is
688 specified) matches the target C calling conventions. This calling
689 convention supports varargs function calls and tolerates some mismatch in
690 the declared prototype and implemented declaration of the function (as
693 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
694 <dd>This calling convention attempts to make calls as fast as possible
695 (e.g. by passing things in registers). This calling convention allows the
696 target to use whatever tricks it wants to produce fast code for the
697 target, without having to conform to an externally specified ABI
698 (Application Binary Interface).
699 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
700 when this or the GHC convention is used.</a> This calling convention
701 does not support varargs and requires the prototype of all callees to
702 exactly match the prototype of the function definition.</dd>
704 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
705 <dd>This calling convention attempts to make code in the caller as efficient
706 as possible under the assumption that the call is not commonly executed.
707 As such, these calls often preserve all registers so that the call does
708 not break any live ranges in the caller side. This calling convention
709 does not support varargs and requires the prototype of all callees to
710 exactly match the prototype of the function definition.</dd>
712 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
713 <dd>This calling convention has been implemented specifically for use by the
714 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
715 It passes everything in registers, going to extremes to achieve this by
716 disabling callee save registers. This calling convention should not be
717 used lightly but only for specific situations such as an alternative to
718 the <em>register pinning</em> performance technique often used when
719 implementing functional programming languages.At the moment only X86
720 supports this convention and it has the following limitations:
722 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
723 floating point types are supported.</li>
724 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
725 6 floating point parameters.</li>
727 This calling convention supports
728 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
729 requires both the caller and callee are using it.
732 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
733 <dd>Any calling convention may be specified by number, allowing
734 target-specific calling conventions to be used. Target specific calling
735 conventions start at 64.</dd>
738 <p>More calling conventions can be added/defined on an as-needed basis, to
739 support Pascal conventions or any other well-known target-independent
744 <!-- ======================================================================= -->
745 <div class="doc_subsection">
746 <a name="visibility">Visibility Styles</a>
749 <div class="doc_text">
751 <p>All Global Variables and Functions have one of the following visibility
755 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
756 <dd>On targets that use the ELF object file format, default visibility means
757 that the declaration is visible to other modules and, in shared libraries,
758 means that the declared entity may be overridden. On Darwin, default
759 visibility means that the declaration is visible to other modules. Default
760 visibility corresponds to "external linkage" in the language.</dd>
762 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
763 <dd>Two declarations of an object with hidden visibility refer to the same
764 object if they are in the same shared object. Usually, hidden visibility
765 indicates that the symbol will not be placed into the dynamic symbol
766 table, so no other module (executable or shared library) can reference it
769 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
770 <dd>On ELF, protected visibility indicates that the symbol will be placed in
771 the dynamic symbol table, but that references within the defining module
772 will bind to the local symbol. That is, the symbol cannot be overridden by
778 <!-- ======================================================================= -->
779 <div class="doc_subsection">
780 <a name="namedtypes">Named Types</a>
783 <div class="doc_text">
785 <p>LLVM IR allows you to specify name aliases for certain types. This can make
786 it easier to read the IR and make the IR more condensed (particularly when
787 recursive types are involved). An example of a name specification is:</p>
789 <div class="doc_code">
791 %mytype = type { %mytype*, i32 }
795 <p>You may give a name to any <a href="#typesystem">type</a> except
796 "<a href="t_void">void</a>". Type name aliases may be used anywhere a type
797 is expected with the syntax "%mytype".</p>
799 <p>Note that type names are aliases for the structural type that they indicate,
800 and that you can therefore specify multiple names for the same type. This
801 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
802 uses structural typing, the name is not part of the type. When printing out
803 LLVM IR, the printer will pick <em>one name</em> to render all types of a
804 particular shape. This means that if you have code where two different
805 source types end up having the same LLVM type, that the dumper will sometimes
806 print the "wrong" or unexpected type. This is an important design point and
807 isn't going to change.</p>
811 <!-- ======================================================================= -->
812 <div class="doc_subsection">
813 <a name="globalvars">Global Variables</a>
816 <div class="doc_text">
818 <p>Global variables define regions of memory allocated at compilation time
819 instead of run-time. Global variables may optionally be initialized, may
820 have an explicit section to be placed in, and may have an optional explicit
821 alignment specified. A variable may be defined as "thread_local", which
822 means that it will not be shared by threads (each thread will have a
823 separated copy of the variable). A variable may be defined as a global
824 "constant," which indicates that the contents of the variable
825 will <b>never</b> be modified (enabling better optimization, allowing the
826 global data to be placed in the read-only section of an executable, etc).
827 Note that variables that need runtime initialization cannot be marked
828 "constant" as there is a store to the variable.</p>
830 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
831 constant, even if the final definition of the global is not. This capability
832 can be used to enable slightly better optimization of the program, but
833 requires the language definition to guarantee that optimizations based on the
834 'constantness' are valid for the translation units that do not include the
837 <p>As SSA values, global variables define pointer values that are in scope
838 (i.e. they dominate) all basic blocks in the program. Global variables
839 always define a pointer to their "content" type because they describe a
840 region of memory, and all memory objects in LLVM are accessed through
843 <p>A global variable may be declared to reside in a target-specific numbered
844 address space. For targets that support them, address spaces may affect how
845 optimizations are performed and/or what target instructions are used to
846 access the variable. The default address space is zero. The address space
847 qualifier must precede any other attributes.</p>
849 <p>LLVM allows an explicit section to be specified for globals. If the target
850 supports it, it will emit globals to the section specified.</p>
852 <p>An explicit alignment may be specified for a global. If not present, or if
853 the alignment is set to zero, the alignment of the global is set by the
854 target to whatever it feels convenient. If an explicit alignment is
855 specified, the global is forced to have at least that much alignment. All
856 alignments must be a power of 2.</p>
858 <p>For example, the following defines a global in a numbered address space with
859 an initializer, section, and alignment:</p>
861 <div class="doc_code">
863 @G = addrspace(5) constant float 1.0, section "foo", align 4
870 <!-- ======================================================================= -->
871 <div class="doc_subsection">
872 <a name="functionstructure">Functions</a>
875 <div class="doc_text">
877 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
878 optional <a href="#linkage">linkage type</a>, an optional
879 <a href="#visibility">visibility style</a>, an optional
880 <a href="#callingconv">calling convention</a>, a return type, an optional
881 <a href="#paramattrs">parameter attribute</a> for the return type, a function
882 name, a (possibly empty) argument list (each with optional
883 <a href="#paramattrs">parameter attributes</a>), optional
884 <a href="#fnattrs">function attributes</a>, an optional section, an optional
885 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
886 curly brace, a list of basic blocks, and a closing curly brace.</p>
888 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
889 optional <a href="#linkage">linkage type</a>, an optional
890 <a href="#visibility">visibility style</a>, an optional
891 <a href="#callingconv">calling convention</a>, a return type, an optional
892 <a href="#paramattrs">parameter attribute</a> for the return type, a function
893 name, a possibly empty list of arguments, an optional alignment, and an
894 optional <a href="#gc">garbage collector name</a>.</p>
896 <p>A function definition contains a list of basic blocks, forming the CFG
897 (Control Flow Graph) for the function. Each basic block may optionally start
898 with a label (giving the basic block a symbol table entry), contains a list
899 of instructions, and ends with a <a href="#terminators">terminator</a>
900 instruction (such as a branch or function return).</p>
902 <p>The first basic block in a function is special in two ways: it is immediately
903 executed on entrance to the function, and it is not allowed to have
904 predecessor basic blocks (i.e. there can not be any branches to the entry
905 block of a function). Because the block can have no predecessors, it also
906 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
908 <p>LLVM allows an explicit section to be specified for functions. If the target
909 supports it, it will emit functions to the section specified.</p>
911 <p>An explicit alignment may be specified for a function. If not present, or if
912 the alignment is set to zero, the alignment of the function is set by the
913 target to whatever it feels convenient. If an explicit alignment is
914 specified, the function is forced to have at least that much alignment. All
915 alignments must be a power of 2.</p>
918 <div class="doc_code">
920 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
921 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
922 <ResultType> @<FunctionName> ([argument list])
923 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
924 [<a href="#gc">gc</a>] { ... }
930 <!-- ======================================================================= -->
931 <div class="doc_subsection">
932 <a name="aliasstructure">Aliases</a>
935 <div class="doc_text">
937 <p>Aliases act as "second name" for the aliasee value (which can be either
938 function, global variable, another alias or bitcast of global value). Aliases
939 may have an optional <a href="#linkage">linkage type</a>, and an
940 optional <a href="#visibility">visibility style</a>.</p>
943 <div class="doc_code">
945 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
951 <!-- ======================================================================= -->
952 <div class="doc_subsection">
953 <a name="namedmetadatastructure">Named Metadata</a>
956 <div class="doc_text">
958 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
959 nodes</a> (but not metadata strings) and null are the only valid operands for
960 a named metadata.</p>
963 <div class="doc_code">
965 !1 = metadata !{metadata !"one"}
972 <!-- ======================================================================= -->
973 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
975 <div class="doc_text">
977 <p>The return type and each parameter of a function type may have a set of
978 <i>parameter attributes</i> associated with them. Parameter attributes are
979 used to communicate additional information about the result or parameters of
980 a function. Parameter attributes are considered to be part of the function,
981 not of the function type, so functions with different parameter attributes
982 can have the same function type.</p>
984 <p>Parameter attributes are simple keywords that follow the type specified. If
985 multiple parameter attributes are needed, they are space separated. For
988 <div class="doc_code">
990 declare i32 @printf(i8* noalias nocapture, ...)
991 declare i32 @atoi(i8 zeroext)
992 declare signext i8 @returns_signed_char()
996 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
997 <tt>readonly</tt>) come immediately after the argument list.</p>
999 <p>Currently, only the following parameter attributes are defined:</p>
1002 <dt><tt><b>zeroext</b></tt></dt>
1003 <dd>This indicates to the code generator that the parameter or return value
1004 should be zero-extended to a 32-bit value by the caller (for a parameter)
1005 or the callee (for a return value).</dd>
1007 <dt><tt><b>signext</b></tt></dt>
1008 <dd>This indicates to the code generator that the parameter or return value
1009 should be sign-extended to a 32-bit value by the caller (for a parameter)
1010 or the callee (for a return value).</dd>
1012 <dt><tt><b>inreg</b></tt></dt>
1013 <dd>This indicates that this parameter or return value should be treated in a
1014 special target-dependent fashion during while emitting code for a function
1015 call or return (usually, by putting it in a register as opposed to memory,
1016 though some targets use it to distinguish between two different kinds of
1017 registers). Use of this attribute is target-specific.</dd>
1019 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1020 <dd>This indicates that the pointer parameter should really be passed by value
1021 to the function. The attribute implies that a hidden copy of the pointee
1022 is made between the caller and the callee, so the callee is unable to
1023 modify the value in the callee. This attribute is only valid on LLVM
1024 pointer arguments. It is generally used to pass structs and arrays by
1025 value, but is also valid on pointers to scalars. The copy is considered
1026 to belong to the caller not the callee (for example,
1027 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1028 <tt>byval</tt> parameters). This is not a valid attribute for return
1029 values. The byval attribute also supports specifying an alignment with
1030 the align attribute. This has a target-specific effect on the code
1031 generator that usually indicates a desired alignment for the synthesized
1034 <dt><tt><b>sret</b></tt></dt>
1035 <dd>This indicates that the pointer parameter specifies the address of a
1036 structure that is the return value of the function in the source program.
1037 This pointer must be guaranteed by the caller to be valid: loads and
1038 stores to the structure may be assumed by the callee to not to trap. This
1039 may only be applied to the first parameter. This is not a valid attribute
1040 for return values. </dd>
1042 <dt><tt><b>noalias</b></tt></dt>
1043 <dd>This indicates that the pointer does not alias any global or any other
1044 parameter. The caller is responsible for ensuring that this is the
1045 case. On a function return value, <tt>noalias</tt> additionally indicates
1046 that the pointer does not alias any other pointers visible to the
1047 caller. For further details, please see the discussion of the NoAlias
1049 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
1052 <dt><tt><b>nocapture</b></tt></dt>
1053 <dd>This indicates that the callee does not make any copies of the pointer
1054 that outlive the callee itself. This is not a valid attribute for return
1057 <dt><tt><b>nest</b></tt></dt>
1058 <dd>This indicates that the pointer parameter can be excised using the
1059 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1060 attribute for return values.</dd>
1065 <!-- ======================================================================= -->
1066 <div class="doc_subsection">
1067 <a name="gc">Garbage Collector Names</a>
1070 <div class="doc_text">
1072 <p>Each function may specify a garbage collector name, which is simply a
1075 <div class="doc_code">
1077 define void @f() gc "name" { ... }
1081 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1082 collector which will cause the compiler to alter its output in order to
1083 support the named garbage collection algorithm.</p>
1087 <!-- ======================================================================= -->
1088 <div class="doc_subsection">
1089 <a name="fnattrs">Function Attributes</a>
1092 <div class="doc_text">
1094 <p>Function attributes are set to communicate additional information about a
1095 function. Function attributes are considered to be part of the function, not
1096 of the function type, so functions with different parameter attributes can
1097 have the same function type.</p>
1099 <p>Function attributes are simple keywords that follow the type specified. If
1100 multiple attributes are needed, they are space separated. For example:</p>
1102 <div class="doc_code">
1104 define void @f() noinline { ... }
1105 define void @f() alwaysinline { ... }
1106 define void @f() alwaysinline optsize { ... }
1107 define void @f() optsize { ... }
1112 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1113 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1114 the backend should forcibly align the stack pointer. Specify the
1115 desired alignment, which must be a power of two, in parentheses.
1117 <dt><tt><b>alwaysinline</b></tt></dt>
1118 <dd>This attribute indicates that the inliner should attempt to inline this
1119 function into callers whenever possible, ignoring any active inlining size
1120 threshold for this caller.</dd>
1122 <dt><tt><b>inlinehint</b></tt></dt>
1123 <dd>This attribute indicates that the source code contained a hint that inlining
1124 this function is desirable (such as the "inline" keyword in C/C++). It
1125 is just a hint; it imposes no requirements on the inliner.</dd>
1127 <dt><tt><b>noinline</b></tt></dt>
1128 <dd>This attribute indicates that the inliner should never inline this
1129 function in any situation. This attribute may not be used together with
1130 the <tt>alwaysinline</tt> attribute.</dd>
1132 <dt><tt><b>optsize</b></tt></dt>
1133 <dd>This attribute suggests that optimization passes and code generator passes
1134 make choices that keep the code size of this function low, and otherwise
1135 do optimizations specifically to reduce code size.</dd>
1137 <dt><tt><b>noreturn</b></tt></dt>
1138 <dd>This function attribute indicates that the function never returns
1139 normally. This produces undefined behavior at runtime if the function
1140 ever does dynamically return.</dd>
1142 <dt><tt><b>nounwind</b></tt></dt>
1143 <dd>This function attribute indicates that the function never returns with an
1144 unwind or exceptional control flow. If the function does unwind, its
1145 runtime behavior is undefined.</dd>
1147 <dt><tt><b>readnone</b></tt></dt>
1148 <dd>This attribute indicates that the function computes its result (or decides
1149 to unwind an exception) based strictly on its arguments, without
1150 dereferencing any pointer arguments or otherwise accessing any mutable
1151 state (e.g. memory, control registers, etc) visible to caller functions.
1152 It does not write through any pointer arguments
1153 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1154 changes any state visible to callers. This means that it cannot unwind
1155 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1156 could use the <tt>unwind</tt> instruction.</dd>
1158 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1159 <dd>This attribute indicates that the function does not write through any
1160 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1161 arguments) or otherwise modify any state (e.g. memory, control registers,
1162 etc) visible to caller functions. It may dereference pointer arguments
1163 and read state that may be set in the caller. A readonly function always
1164 returns the same value (or unwinds an exception identically) when called
1165 with the same set of arguments and global state. It cannot unwind an
1166 exception by calling the <tt>C++</tt> exception throwing methods, but may
1167 use the <tt>unwind</tt> instruction.</dd>
1169 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1170 <dd>This attribute indicates that the function should emit a stack smashing
1171 protector. It is in the form of a "canary"—a random value placed on
1172 the stack before the local variables that's checked upon return from the
1173 function to see if it has been overwritten. A heuristic is used to
1174 determine if a function needs stack protectors or not.<br>
1176 If a function that has an <tt>ssp</tt> attribute is inlined into a
1177 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1178 function will have an <tt>ssp</tt> attribute.</dd>
1180 <dt><tt><b>sspreq</b></tt></dt>
1181 <dd>This attribute indicates that the function should <em>always</em> emit a
1182 stack smashing protector. This overrides
1183 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1185 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1186 function that doesn't have an <tt>sspreq</tt> attribute or which has
1187 an <tt>ssp</tt> attribute, then the resulting function will have
1188 an <tt>sspreq</tt> attribute.</dd>
1190 <dt><tt><b>noredzone</b></tt></dt>
1191 <dd>This attribute indicates that the code generator should not use a red
1192 zone, even if the target-specific ABI normally permits it.</dd>
1194 <dt><tt><b>noimplicitfloat</b></tt></dt>
1195 <dd>This attributes disables implicit floating point instructions.</dd>
1197 <dt><tt><b>naked</b></tt></dt>
1198 <dd>This attribute disables prologue / epilogue emission for the function.
1199 This can have very system-specific consequences.</dd>
1204 <!-- ======================================================================= -->
1205 <div class="doc_subsection">
1206 <a name="moduleasm">Module-Level Inline Assembly</a>
1209 <div class="doc_text">
1211 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1212 the GCC "file scope inline asm" blocks. These blocks are internally
1213 concatenated by LLVM and treated as a single unit, but may be separated in
1214 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1216 <div class="doc_code">
1218 module asm "inline asm code goes here"
1219 module asm "more can go here"
1223 <p>The strings can contain any character by escaping non-printable characters.
1224 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1227 <p>The inline asm code is simply printed to the machine code .s file when
1228 assembly code is generated.</p>
1232 <!-- ======================================================================= -->
1233 <div class="doc_subsection">
1234 <a name="datalayout">Data Layout</a>
1237 <div class="doc_text">
1239 <p>A module may specify a target specific data layout string that specifies how
1240 data is to be laid out in memory. The syntax for the data layout is
1243 <div class="doc_code">
1245 target datalayout = "<i>layout specification</i>"
1249 <p>The <i>layout specification</i> consists of a list of specifications
1250 separated by the minus sign character ('-'). Each specification starts with
1251 a letter and may include other information after the letter to define some
1252 aspect of the data layout. The specifications accepted are as follows:</p>
1256 <dd>Specifies that the target lays out data in big-endian form. That is, the
1257 bits with the most significance have the lowest address location.</dd>
1260 <dd>Specifies that the target lays out data in little-endian form. That is,
1261 the bits with the least significance have the lowest address
1264 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1265 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1266 <i>preferred</i> alignments. All sizes are in bits. Specifying
1267 the <i>pref</i> alignment is optional. If omitted, the
1268 preceding <tt>:</tt> should be omitted too.</dd>
1270 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1271 <dd>This specifies the alignment for an integer type of a given bit
1272 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1274 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1275 <dd>This specifies the alignment for a vector type of a given bit
1278 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1279 <dd>This specifies the alignment for a floating point type of a given bit
1280 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1283 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1284 <dd>This specifies the alignment for an aggregate type of a given bit
1287 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1288 <dd>This specifies the alignment for a stack object of a given bit
1291 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1292 <dd>This specifies a set of native integer widths for the target CPU
1293 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1294 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1295 this set are considered to support most general arithmetic
1296 operations efficiently.</dd>
1299 <p>When constructing the data layout for a given target, LLVM starts with a
1300 default set of specifications which are then (possibly) overriden by the
1301 specifications in the <tt>datalayout</tt> keyword. The default specifications
1302 are given in this list:</p>
1305 <li><tt>E</tt> - big endian</li>
1306 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1307 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1308 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1309 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1310 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1311 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1312 alignment of 64-bits</li>
1313 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1314 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1315 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1316 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1317 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1318 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1321 <p>When LLVM is determining the alignment for a given type, it uses the
1322 following rules:</p>
1325 <li>If the type sought is an exact match for one of the specifications, that
1326 specification is used.</li>
1328 <li>If no match is found, and the type sought is an integer type, then the
1329 smallest integer type that is larger than the bitwidth of the sought type
1330 is used. If none of the specifications are larger than the bitwidth then
1331 the the largest integer type is used. For example, given the default
1332 specifications above, the i7 type will use the alignment of i8 (next
1333 largest) while both i65 and i256 will use the alignment of i64 (largest
1336 <li>If no match is found, and the type sought is a vector type, then the
1337 largest vector type that is smaller than the sought vector type will be
1338 used as a fall back. This happens because <128 x double> can be
1339 implemented in terms of 64 <2 x double>, for example.</li>
1344 <!-- ======================================================================= -->
1345 <div class="doc_subsection">
1346 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1349 <div class="doc_text">
1351 <p>Any memory access must be done through a pointer value associated
1352 with an address range of the memory access, otherwise the behavior
1353 is undefined. Pointer values are associated with address ranges
1354 according to the following rules:</p>
1357 <li>A pointer value formed from a
1358 <tt><a href="#i_getelementptr">getelementptr</a></tt> instruction
1359 is associated with the addresses associated with the first operand
1360 of the <tt>getelementptr</tt>.</li>
1361 <li>An address of a global variable is associated with the address
1362 range of the variable's storage.</li>
1363 <li>The result value of an allocation instruction is associated with
1364 the address range of the allocated storage.</li>
1365 <li>A null pointer in the default address-space is associated with
1367 <li>A pointer value formed by an
1368 <tt><a href="#i_inttoptr">inttoptr</a></tt> is associated with all
1369 address ranges of all pointer values that contribute (directly or
1370 indirectly) to the computation of the pointer's value.</li>
1371 <li>The result value of a
1372 <tt><a href="#i_bitcast">bitcast</a></tt> is associated with all
1373 addresses associated with the operand of the <tt>bitcast</tt>.</li>
1374 <li>An integer constant other than zero or a pointer value returned
1375 from a function not defined within LLVM may be associated with address
1376 ranges allocated through mechanisms other than those provided by
1377 LLVM. Such ranges shall not overlap with any ranges of addresses
1378 allocated by mechanisms provided by LLVM.</li>
1381 <p>LLVM IR does not associate types with memory. The result type of a
1382 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1383 alignment of the memory from which to load, as well as the
1384 interpretation of the value. The first operand of a
1385 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1386 and alignment of the store.</p>
1388 <p>Consequently, type-based alias analysis, aka TBAA, aka
1389 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1390 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1391 additional information which specialized optimization passes may use
1392 to implement type-based alias analysis.</p>
1396 <!-- *********************************************************************** -->
1397 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1398 <!-- *********************************************************************** -->
1400 <div class="doc_text">
1402 <p>The LLVM type system is one of the most important features of the
1403 intermediate representation. Being typed enables a number of optimizations
1404 to be performed on the intermediate representation directly, without having
1405 to do extra analyses on the side before the transformation. A strong type
1406 system makes it easier to read the generated code and enables novel analyses
1407 and transformations that are not feasible to perform on normal three address
1408 code representations.</p>
1412 <!-- ======================================================================= -->
1413 <div class="doc_subsection"> <a name="t_classifications">Type
1414 Classifications</a> </div>
1416 <div class="doc_text">
1418 <p>The types fall into a few useful classifications:</p>
1420 <table border="1" cellspacing="0" cellpadding="4">
1422 <tr><th>Classification</th><th>Types</th></tr>
1424 <td><a href="#t_integer">integer</a></td>
1425 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1428 <td><a href="#t_floating">floating point</a></td>
1429 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1432 <td><a name="t_firstclass">first class</a></td>
1433 <td><a href="#t_integer">integer</a>,
1434 <a href="#t_floating">floating point</a>,
1435 <a href="#t_pointer">pointer</a>,
1436 <a href="#t_vector">vector</a>,
1437 <a href="#t_struct">structure</a>,
1438 <a href="#t_union">union</a>,
1439 <a href="#t_array">array</a>,
1440 <a href="#t_label">label</a>,
1441 <a href="#t_metadata">metadata</a>.
1445 <td><a href="#t_primitive">primitive</a></td>
1446 <td><a href="#t_label">label</a>,
1447 <a href="#t_void">void</a>,
1448 <a href="#t_floating">floating point</a>,
1449 <a href="#t_metadata">metadata</a>.</td>
1452 <td><a href="#t_derived">derived</a></td>
1453 <td><a href="#t_array">array</a>,
1454 <a href="#t_function">function</a>,
1455 <a href="#t_pointer">pointer</a>,
1456 <a href="#t_struct">structure</a>,
1457 <a href="#t_pstruct">packed structure</a>,
1458 <a href="#t_union">union</a>,
1459 <a href="#t_vector">vector</a>,
1460 <a href="#t_opaque">opaque</a>.
1466 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1467 important. Values of these types are the only ones which can be produced by
1472 <!-- ======================================================================= -->
1473 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1475 <div class="doc_text">
1477 <p>The primitive types are the fundamental building blocks of the LLVM
1482 <!-- _______________________________________________________________________ -->
1483 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1485 <div class="doc_text">
1488 <p>The integer type is a very simple type that simply specifies an arbitrary
1489 bit width for the integer type desired. Any bit width from 1 bit to
1490 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1497 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1501 <table class="layout">
1503 <td class="left"><tt>i1</tt></td>
1504 <td class="left">a single-bit integer.</td>
1507 <td class="left"><tt>i32</tt></td>
1508 <td class="left">a 32-bit integer.</td>
1511 <td class="left"><tt>i1942652</tt></td>
1512 <td class="left">a really big integer of over 1 million bits.</td>
1518 <!-- _______________________________________________________________________ -->
1519 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1521 <div class="doc_text">
1525 <tr><th>Type</th><th>Description</th></tr>
1526 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1527 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1528 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1529 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1530 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1536 <!-- _______________________________________________________________________ -->
1537 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1539 <div class="doc_text">
1542 <p>The void type does not represent any value and has no size.</p>
1551 <!-- _______________________________________________________________________ -->
1552 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1554 <div class="doc_text">
1557 <p>The label type represents code labels.</p>
1566 <!-- _______________________________________________________________________ -->
1567 <div class="doc_subsubsection"> <a name="t_metadata">Metadata Type</a> </div>
1569 <div class="doc_text">
1572 <p>The metadata type represents embedded metadata. No derived types may be
1573 created from metadata except for <a href="#t_function">function</a>
1584 <!-- ======================================================================= -->
1585 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1587 <div class="doc_text">
1589 <p>The real power in LLVM comes from the derived types in the system. This is
1590 what allows a programmer to represent arrays, functions, pointers, and other
1591 useful types. Each of these types contain one or more element types which
1592 may be a primitive type, or another derived type. For example, it is
1593 possible to have a two dimensional array, using an array as the element type
1594 of another array.</p>
1599 <!-- _______________________________________________________________________ -->
1600 <div class="doc_subsubsection"> <a name="t_aggregate">Aggregate Types</a> </div>
1602 <div class="doc_text">
1604 <p>Aggregate Types are a subset of derived types that can contain multiple
1605 member types. <a href="#t_array">Arrays</a>,
1606 <a href="#t_struct">structs</a>, <a href="#t_vector">vectors</a> and
1607 <a href="#t_union">unions</a> are aggregate types.</p>
1613 <!-- _______________________________________________________________________ -->
1614 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1616 <div class="doc_text">
1619 <p>The array type is a very simple derived type that arranges elements
1620 sequentially in memory. The array type requires a size (number of elements)
1621 and an underlying data type.</p>
1625 [<# elements> x <elementtype>]
1628 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1629 be any type with a size.</p>
1632 <table class="layout">
1634 <td class="left"><tt>[40 x i32]</tt></td>
1635 <td class="left">Array of 40 32-bit integer values.</td>
1638 <td class="left"><tt>[41 x i32]</tt></td>
1639 <td class="left">Array of 41 32-bit integer values.</td>
1642 <td class="left"><tt>[4 x i8]</tt></td>
1643 <td class="left">Array of 4 8-bit integer values.</td>
1646 <p>Here are some examples of multidimensional arrays:</p>
1647 <table class="layout">
1649 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1650 <td class="left">3x4 array of 32-bit integer values.</td>
1653 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1654 <td class="left">12x10 array of single precision floating point values.</td>
1657 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1658 <td class="left">2x3x4 array of 16-bit integer values.</td>
1662 <p>There is no restriction on indexing beyond the end of the array implied by
1663 a static type (though there are restrictions on indexing beyond the bounds
1664 of an allocated object in some cases). This means that single-dimension
1665 'variable sized array' addressing can be implemented in LLVM with a zero
1666 length array type. An implementation of 'pascal style arrays' in LLVM could
1667 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1671 <!-- _______________________________________________________________________ -->
1672 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1674 <div class="doc_text">
1677 <p>The function type can be thought of as a function signature. It consists of
1678 a return type and a list of formal parameter types. The return type of a
1679 function type is a scalar type, a void type, a struct type, or a union
1680 type. If the return type is a struct type then all struct elements must be
1681 of first class types, and the struct must have at least one element.</p>
1685 <returntype> (<parameter list>)
1688 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1689 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1690 which indicates that the function takes a variable number of arguments.
1691 Variable argument functions can access their arguments with
1692 the <a href="#int_varargs">variable argument handling intrinsic</a>
1693 functions. '<tt><returntype></tt>' is any type except
1694 <a href="#t_label">label</a>.</p>
1697 <table class="layout">
1699 <td class="left"><tt>i32 (i32)</tt></td>
1700 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1702 </tr><tr class="layout">
1703 <td class="left"><tt>float (i16, i32 *) *
1705 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1706 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
1707 returning <tt>float</tt>.
1709 </tr><tr class="layout">
1710 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1711 <td class="left">A vararg function that takes at least one
1712 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1713 which returns an integer. This is the signature for <tt>printf</tt> in
1716 </tr><tr class="layout">
1717 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1718 <td class="left">A function taking an <tt>i32</tt>, returning a
1719 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
1726 <!-- _______________________________________________________________________ -->
1727 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1729 <div class="doc_text">
1732 <p>The structure type is used to represent a collection of data members together
1733 in memory. The packing of the field types is defined to match the ABI of the
1734 underlying processor. The elements of a structure may be any type that has a
1737 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
1738 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
1739 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
1740 Structures in registers are accessed using the
1741 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
1742 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
1745 { <type list> }
1749 <table class="layout">
1751 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1752 <td class="left">A triple of three <tt>i32</tt> values</td>
1753 </tr><tr class="layout">
1754 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1755 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1756 second element is a <a href="#t_pointer">pointer</a> to a
1757 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1758 an <tt>i32</tt>.</td>
1764 <!-- _______________________________________________________________________ -->
1765 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1768 <div class="doc_text">
1771 <p>The packed structure type is used to represent a collection of data members
1772 together in memory. There is no padding between fields. Further, the
1773 alignment of a packed structure is 1 byte. The elements of a packed
1774 structure may be any type that has a size.</p>
1776 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1777 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1778 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1782 < { <type list> } >
1786 <table class="layout">
1788 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1789 <td class="left">A triple of three <tt>i32</tt> values</td>
1790 </tr><tr class="layout">
1792 <tt>< { float, i32 (i32)* } ></tt></td>
1793 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1794 second element is a <a href="#t_pointer">pointer</a> to a
1795 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1796 an <tt>i32</tt>.</td>
1802 <!-- _______________________________________________________________________ -->
1803 <div class="doc_subsubsection"> <a name="t_union">Union Type</a> </div>
1805 <div class="doc_text">
1808 <p>A union type describes an object with size and alignment suitable for
1809 an object of any one of a given set of types (also known as an "untagged"
1810 union). It is similar in concept and usage to a
1811 <a href="#t_struct">struct</a>, except that all members of the union
1812 have an offset of zero. The elements of a union may be any type that has a
1813 size. Unions must have at least one member - empty unions are not allowed.
1816 <p>The size of the union as a whole will be the size of its largest member,
1817 and the alignment requirements of the union as a whole will be the largest
1818 alignment requirement of any member.</p>
1820 <p>Union members are accessed using '<tt><a href="#i_load">load</a></tt> and
1821 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1822 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
1823 Since all members are at offset zero, the getelementptr instruction does
1824 not affect the address, only the type of the resulting pointer.</p>
1828 union { <type list> }
1832 <table class="layout">
1834 <td class="left"><tt>union { i32, i32*, float }</tt></td>
1835 <td class="left">A union of three types: an <tt>i32</tt>, a pointer to
1836 an <tt>i32</tt>, and a <tt>float</tt>.</td>
1837 </tr><tr class="layout">
1839 <tt>union { float, i32 (i32) * }</tt></td>
1840 <td class="left">A union, where the first element is a <tt>float</tt> and the
1841 second element is a <a href="#t_pointer">pointer</a> to a
1842 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1843 an <tt>i32</tt>.</td>
1849 <!-- _______________________________________________________________________ -->
1850 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1852 <div class="doc_text">
1855 <p>The pointer type is used to specify memory locations.
1856 Pointers are commonly used to reference objects in memory.</p>
1858 <p>Pointer types may have an optional address space attribute defining the
1859 numbered address space where the pointed-to object resides. The default
1860 address space is number zero. The semantics of non-zero address
1861 spaces are target-specific.</p>
1863 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
1864 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1872 <table class="layout">
1874 <td class="left"><tt>[4 x i32]*</tt></td>
1875 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1876 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1879 <td class="left"><tt>i32 (i32 *) *</tt></td>
1880 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1881 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1885 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1886 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1887 that resides in address space #5.</td>
1893 <!-- _______________________________________________________________________ -->
1894 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1896 <div class="doc_text">
1899 <p>A vector type is a simple derived type that represents a vector of elements.
1900 Vector types are used when multiple primitive data are operated in parallel
1901 using a single instruction (SIMD). A vector type requires a size (number of
1902 elements) and an underlying primitive data type. Vector types are considered
1903 <a href="#t_firstclass">first class</a>.</p>
1907 < <# elements> x <elementtype> >
1910 <p>The number of elements is a constant integer value; elementtype may be any
1911 integer or floating point type.</p>
1914 <table class="layout">
1916 <td class="left"><tt><4 x i32></tt></td>
1917 <td class="left">Vector of 4 32-bit integer values.</td>
1920 <td class="left"><tt><8 x float></tt></td>
1921 <td class="left">Vector of 8 32-bit floating-point values.</td>
1924 <td class="left"><tt><2 x i64></tt></td>
1925 <td class="left">Vector of 2 64-bit integer values.</td>
1931 <!-- _______________________________________________________________________ -->
1932 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1933 <div class="doc_text">
1936 <p>Opaque types are used to represent unknown types in the system. This
1937 corresponds (for example) to the C notion of a forward declared structure
1938 type. In LLVM, opaque types can eventually be resolved to any type (not just
1939 a structure type).</p>
1947 <table class="layout">
1949 <td class="left"><tt>opaque</tt></td>
1950 <td class="left">An opaque type.</td>
1956 <!-- ======================================================================= -->
1957 <div class="doc_subsection">
1958 <a name="t_uprefs">Type Up-references</a>
1961 <div class="doc_text">
1964 <p>An "up reference" allows you to refer to a lexically enclosing type without
1965 requiring it to have a name. For instance, a structure declaration may
1966 contain a pointer to any of the types it is lexically a member of. Example
1967 of up references (with their equivalent as named type declarations)
1971 { \2 * } %x = type { %x* }
1972 { \2 }* %y = type { %y }*
1976 <p>An up reference is needed by the asmprinter for printing out cyclic types
1977 when there is no declared name for a type in the cycle. Because the
1978 asmprinter does not want to print out an infinite type string, it needs a
1979 syntax to handle recursive types that have no names (all names are optional
1987 <p>The level is the count of the lexical type that is being referred to.</p>
1990 <table class="layout">
1992 <td class="left"><tt>\1*</tt></td>
1993 <td class="left">Self-referential pointer.</td>
1996 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1997 <td class="left">Recursive structure where the upref refers to the out-most
2004 <!-- *********************************************************************** -->
2005 <div class="doc_section"> <a name="constants">Constants</a> </div>
2006 <!-- *********************************************************************** -->
2008 <div class="doc_text">
2010 <p>LLVM has several different basic types of constants. This section describes
2011 them all and their syntax.</p>
2015 <!-- ======================================================================= -->
2016 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
2018 <div class="doc_text">
2021 <dt><b>Boolean constants</b></dt>
2022 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2023 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2025 <dt><b>Integer constants</b></dt>
2026 <dd>Standard integers (such as '4') are constants of
2027 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2028 with integer types.</dd>
2030 <dt><b>Floating point constants</b></dt>
2031 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2032 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2033 notation (see below). The assembler requires the exact decimal value of a
2034 floating-point constant. For example, the assembler accepts 1.25 but
2035 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2036 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2038 <dt><b>Null pointer constants</b></dt>
2039 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2040 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2043 <p>The one non-intuitive notation for constants is the hexadecimal form of
2044 floating point constants. For example, the form '<tt>double
2045 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2046 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2047 constants are required (and the only time that they are generated by the
2048 disassembler) is when a floating point constant must be emitted but it cannot
2049 be represented as a decimal floating point number in a reasonable number of
2050 digits. For example, NaN's, infinities, and other special values are
2051 represented in their IEEE hexadecimal format so that assembly and disassembly
2052 do not cause any bits to change in the constants.</p>
2054 <p>When using the hexadecimal form, constants of types float and double are
2055 represented using the 16-digit form shown above (which matches the IEEE754
2056 representation for double); float values must, however, be exactly
2057 representable as IEE754 single precision. Hexadecimal format is always used
2058 for long double, and there are three forms of long double. The 80-bit format
2059 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2060 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2061 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2062 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2063 currently supported target uses this format. Long doubles will only work if
2064 they match the long double format on your target. All hexadecimal formats
2065 are big-endian (sign bit at the left).</p>
2069 <!-- ======================================================================= -->
2070 <div class="doc_subsection">
2071 <a name="aggregateconstants"></a> <!-- old anchor -->
2072 <a name="complexconstants">Complex Constants</a>
2075 <div class="doc_text">
2077 <p>Complex constants are a (potentially recursive) combination of simple
2078 constants and smaller complex constants.</p>
2081 <dt><b>Structure constants</b></dt>
2082 <dd>Structure constants are represented with notation similar to structure
2083 type definitions (a comma separated list of elements, surrounded by braces
2084 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2085 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2086 Structure constants must have <a href="#t_struct">structure type</a>, and
2087 the number and types of elements must match those specified by the
2090 <dt><b>Union constants</b></dt>
2091 <dd>Union constants are represented with notation similar to a structure with
2092 a single element - that is, a single typed element surrounded
2093 by braces (<tt>{}</tt>)). For example: "<tt>{ i32 4 }</tt>". The
2094 <a href="#t_union">union type</a> can be initialized with a single-element
2095 struct as long as the type of the struct element matches the type of
2096 one of the union members.</dd>
2098 <dt><b>Array constants</b></dt>
2099 <dd>Array constants are represented with notation similar to array type
2100 definitions (a comma separated list of elements, surrounded by square
2101 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2102 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2103 the number and types of elements must match those specified by the
2106 <dt><b>Vector constants</b></dt>
2107 <dd>Vector constants are represented with notation similar to vector type
2108 definitions (a comma separated list of elements, surrounded by
2109 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2110 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2111 have <a href="#t_vector">vector type</a>, and the number and types of
2112 elements must match those specified by the type.</dd>
2114 <dt><b>Zero initialization</b></dt>
2115 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2116 value to zero of <em>any</em> type, including scalar and
2117 <a href="#t_aggregate">aggregate</a> types.
2118 This is often used to avoid having to print large zero initializers
2119 (e.g. for large arrays) and is always exactly equivalent to using explicit
2120 zero initializers.</dd>
2122 <dt><b>Metadata node</b></dt>
2123 <dd>A metadata node is a structure-like constant with
2124 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2125 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2126 be interpreted as part of the instruction stream, metadata is a place to
2127 attach additional information such as debug info.</dd>
2132 <!-- ======================================================================= -->
2133 <div class="doc_subsection">
2134 <a name="globalconstants">Global Variable and Function Addresses</a>
2137 <div class="doc_text">
2139 <p>The addresses of <a href="#globalvars">global variables</a>
2140 and <a href="#functionstructure">functions</a> are always implicitly valid
2141 (link-time) constants. These constants are explicitly referenced when
2142 the <a href="#identifiers">identifier for the global</a> is used and always
2143 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2144 legal LLVM file:</p>
2146 <div class="doc_code">
2150 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2156 <!-- ======================================================================= -->
2157 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
2158 <div class="doc_text">
2160 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2161 indicates that the user of the value may receive an unspecified bit-pattern.
2162 Undefined values may be of any type (other than label or void) and be used
2163 anywhere a constant is permitted.</p>
2165 <p>Undefined values are useful because they indicate to the compiler that the
2166 program is well defined no matter what value is used. This gives the
2167 compiler more freedom to optimize. Here are some examples of (potentially
2168 surprising) transformations that are valid (in pseudo IR):</p>
2171 <div class="doc_code">
2183 <p>This is safe because all of the output bits are affected by the undef bits.
2184 Any output bit can have a zero or one depending on the input bits.</p>
2186 <div class="doc_code">
2199 <p>These logical operations have bits that are not always affected by the input.
2200 For example, if "%X" has a zero bit, then the output of the 'and' operation will
2201 always be a zero, no matter what the corresponding bit from the undef is. As
2202 such, it is unsafe to optimize or assume that the result of the and is undef.
2203 However, it is safe to assume that all bits of the undef could be 0, and
2204 optimize the and to 0. Likewise, it is safe to assume that all the bits of
2205 the undef operand to the or could be set, allowing the or to be folded to
2208 <div class="doc_code">
2210 %A = select undef, %X, %Y
2211 %B = select undef, 42, %Y
2212 %C = select %X, %Y, undef
2224 <p>This set of examples show that undefined select (and conditional branch)
2225 conditions can go "either way" but they have to come from one of the two
2226 operands. In the %A example, if %X and %Y were both known to have a clear low
2227 bit, then %A would have to have a cleared low bit. However, in the %C example,
2228 the optimizer is allowed to assume that the undef operand could be the same as
2229 %Y, allowing the whole select to be eliminated.</p>
2232 <div class="doc_code">
2234 %A = xor undef, undef
2253 <p>This example points out that two undef operands are not necessarily the same.
2254 This can be surprising to people (and also matches C semantics) where they
2255 assume that "X^X" is always zero, even if X is undef. This isn't true for a
2256 number of reasons, but the short answer is that an undef "variable" can
2257 arbitrarily change its value over its "live range". This is true because the
2258 "variable" doesn't actually <em>have a live range</em>. Instead, the value is
2259 logically read from arbitrary registers that happen to be around when needed,
2260 so the value is not necessarily consistent over time. In fact, %A and %C need
2261 to have the same semantics or the core LLVM "replace all uses with" concept
2264 <div class="doc_code">
2274 <p>These examples show the crucial difference between an <em>undefined
2275 value</em> and <em>undefined behavior</em>. An undefined value (like undef) is
2276 allowed to have an arbitrary bit-pattern. This means that the %A operation
2277 can be constant folded to undef because the undef could be an SNaN, and fdiv is
2278 not (currently) defined on SNaN's. However, in the second example, we can make
2279 a more aggressive assumption: because the undef is allowed to be an arbitrary
2280 value, we are allowed to assume that it could be zero. Since a divide by zero
2281 has <em>undefined behavior</em>, we are allowed to assume that the operation
2282 does not execute at all. This allows us to delete the divide and all code after
2283 it: since the undefined operation "can't happen", the optimizer can assume that
2284 it occurs in dead code.
2287 <div class="doc_code">
2289 a: store undef -> %X
2290 b: store %X -> undef
2297 <p>These examples reiterate the fdiv example: a store "of" an undefined value
2298 can be assumed to not have any effect: we can assume that the value is
2299 overwritten with bits that happen to match what was already there. However, a
2300 store "to" an undefined location could clobber arbitrary memory, therefore, it
2301 has undefined behavior.</p>
2305 <!-- ======================================================================= -->
2306 <div class="doc_subsection"><a name="blockaddress">Addresses of Basic
2308 <div class="doc_text">
2310 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2312 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2313 basic block in the specified function, and always has an i8* type. Taking
2314 the address of the entry block is illegal.</p>
2316 <p>This value only has defined behavior when used as an operand to the
2317 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction or for comparisons
2318 against null. Pointer equality tests between labels addresses is undefined
2319 behavior - though, again, comparison against null is ok, and no label is
2320 equal to the null pointer. This may also be passed around as an opaque
2321 pointer sized value as long as the bits are not inspected. This allows
2322 <tt>ptrtoint</tt> and arithmetic to be performed on these values so long as
2323 the original value is reconstituted before the <tt>indirectbr</tt>.</p>
2325 <p>Finally, some targets may provide defined semantics when
2326 using the value as the operand to an inline assembly, but that is target
2333 <!-- ======================================================================= -->
2334 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
2337 <div class="doc_text">
2339 <p>Constant expressions are used to allow expressions involving other constants
2340 to be used as constants. Constant expressions may be of
2341 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2342 operation that does not have side effects (e.g. load and call are not
2343 supported). The following is the syntax for constant expressions:</p>
2346 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
2347 <dd>Truncate a constant to another type. The bit size of CST must be larger
2348 than the bit size of TYPE. Both types must be integers.</dd>
2350 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
2351 <dd>Zero extend a constant to another type. The bit size of CST must be
2352 smaller or equal to the bit size of TYPE. Both types must be
2355 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
2356 <dd>Sign extend a constant to another type. The bit size of CST must be
2357 smaller or equal to the bit size of TYPE. Both types must be
2360 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
2361 <dd>Truncate a floating point constant to another floating point type. The
2362 size of CST must be larger than the size of TYPE. Both types must be
2363 floating point.</dd>
2365 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
2366 <dd>Floating point extend a constant to another type. The size of CST must be
2367 smaller or equal to the size of TYPE. Both types must be floating
2370 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
2371 <dd>Convert a floating point constant to the corresponding unsigned integer
2372 constant. TYPE must be a scalar or vector integer type. CST must be of
2373 scalar or vector floating point type. Both CST and TYPE must be scalars,
2374 or vectors of the same number of elements. If the value won't fit in the
2375 integer type, the results are undefined.</dd>
2377 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
2378 <dd>Convert a floating point constant to the corresponding signed integer
2379 constant. TYPE must be a scalar or vector integer type. CST must be of
2380 scalar or vector floating point type. Both CST and TYPE must be scalars,
2381 or vectors of the same number of elements. If the value won't fit in the
2382 integer type, the results are undefined.</dd>
2384 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
2385 <dd>Convert an unsigned integer constant to the corresponding floating point
2386 constant. TYPE must be a scalar or vector floating point type. CST must be
2387 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2388 vectors of the same number of elements. If the value won't fit in the
2389 floating point type, the results are undefined.</dd>
2391 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
2392 <dd>Convert a signed integer constant to the corresponding floating point
2393 constant. TYPE must be a scalar or vector floating point type. CST must be
2394 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2395 vectors of the same number of elements. If the value won't fit in the
2396 floating point type, the results are undefined.</dd>
2398 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
2399 <dd>Convert a pointer typed constant to the corresponding integer constant
2400 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2401 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2402 make it fit in <tt>TYPE</tt>.</dd>
2404 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
2405 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2406 type. CST must be of integer type. The CST value is zero extended,
2407 truncated, or unchanged to make it fit in a pointer size. This one is
2408 <i>really</i> dangerous!</dd>
2410 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
2411 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2412 are the same as those for the <a href="#i_bitcast">bitcast
2413 instruction</a>.</dd>
2415 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2416 <dt><b><tt>getelementptr inbounds ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2417 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2418 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2419 instruction, the index list may have zero or more indexes, which are
2420 required to make sense for the type of "CSTPTR".</dd>
2422 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
2423 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2425 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
2426 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2428 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
2429 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2431 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2432 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2435 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2436 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2439 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2440 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2443 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2444 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2445 be any of the <a href="#binaryops">binary</a>
2446 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2447 on operands are the same as those for the corresponding instruction
2448 (e.g. no bitwise operations on floating point values are allowed).</dd>
2453 <!-- *********************************************************************** -->
2454 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2455 <!-- *********************************************************************** -->
2457 <!-- ======================================================================= -->
2458 <div class="doc_subsection">
2459 <a name="inlineasm">Inline Assembler Expressions</a>
2462 <div class="doc_text">
2464 <p>LLVM supports inline assembler expressions (as opposed
2465 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2466 a special value. This value represents the inline assembler as a string
2467 (containing the instructions to emit), a list of operand constraints (stored
2468 as a string), a flag that indicates whether or not the inline asm
2469 expression has side effects, and a flag indicating whether the function
2470 containing the asm needs to align its stack conservatively. An example
2471 inline assembler expression is:</p>
2473 <div class="doc_code">
2475 i32 (i32) asm "bswap $0", "=r,r"
2479 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2480 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2483 <div class="doc_code">
2485 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2489 <p>Inline asms with side effects not visible in the constraint list must be
2490 marked as having side effects. This is done through the use of the
2491 '<tt>sideeffect</tt>' keyword, like so:</p>
2493 <div class="doc_code">
2495 call void asm sideeffect "eieio", ""()
2499 <p>In some cases inline asms will contain code that will not work unless the
2500 stack is aligned in some way, such as calls or SSE instructions on x86,
2501 yet will not contain code that does that alignment within the asm.
2502 The compiler should make conservative assumptions about what the asm might
2503 contain and should generate its usual stack alignment code in the prologue
2504 if the '<tt>alignstack</tt>' keyword is present:</p>
2506 <div class="doc_code">
2508 call void asm alignstack "eieio", ""()
2512 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2515 <p>TODO: The format of the asm and constraints string still need to be
2516 documented here. Constraints on what can be done (e.g. duplication, moving,
2517 etc need to be documented). This is probably best done by reference to
2518 another document that covers inline asm from a holistic perspective.</p>
2522 <!-- ======================================================================= -->
2523 <div class="doc_subsection"><a name="metadata">Metadata Nodes and Metadata
2527 <div class="doc_text">
2529 <p>LLVM IR allows metadata to be attached to instructions in the program that
2530 can convey extra information about the code to the optimizers and code
2531 generator. One example application of metadata is source-level debug
2532 information. There are two metadata primitives: strings and nodes. All
2533 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2534 preceding exclamation point ('<tt>!</tt>').</p>
2536 <p>A metadata string is a string surrounded by double quotes. It can contain
2537 any character by escaping non-printable characters with "\xx" where "xx" is
2538 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2540 <p>Metadata nodes are represented with notation similar to structure constants
2541 (a comma separated list of elements, surrounded by braces and preceded by an
2542 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2543 10}</tt>". Metadata nodes can have any values as their operand.</p>
2545 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2546 metadata nodes, which can be looked up in the module symbol table. For
2547 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2549 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2550 function is using two metadata arguments.
2552 <div class="doc_code">
2554 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2558 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2559 attached with <tt>add</tt> instruction using <tt>!dbg</tt> identifier.
2561 <div class="doc_code">
2563 %indvar.next = add i64 %indvar, 1, !dbg !21
2569 <!-- *********************************************************************** -->
2570 <div class="doc_section">
2571 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2573 <!-- *********************************************************************** -->
2575 <p>LLVM has a number of "magic" global variables that contain data that affect
2576 code generation or other IR semantics. These are documented here. All globals
2577 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2578 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2581 <!-- ======================================================================= -->
2582 <div class="doc_subsection">
2583 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2586 <div class="doc_text">
2588 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2589 href="#linkage_appending">appending linkage</a>. This array contains a list of
2590 pointers to global variables and functions which may optionally have a pointer
2591 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2597 @llvm.used = appending global [2 x i8*] [
2599 i8* bitcast (i32* @Y to i8*)
2600 ], section "llvm.metadata"
2603 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2604 compiler, assembler, and linker are required to treat the symbol as if there is
2605 a reference to the global that it cannot see. For example, if a variable has
2606 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2607 list, it cannot be deleted. This is commonly used to represent references from
2608 inline asms and other things the compiler cannot "see", and corresponds to
2609 "attribute((used))" in GNU C.</p>
2611 <p>On some targets, the code generator must emit a directive to the assembler or
2612 object file to prevent the assembler and linker from molesting the symbol.</p>
2616 <!-- ======================================================================= -->
2617 <div class="doc_subsection">
2618 <a name="intg_compiler_used">The '<tt>llvm.compiler.used</tt>' Global Variable</a>
2621 <div class="doc_text">
2623 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2624 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2625 touching the symbol. On targets that support it, this allows an intelligent
2626 linker to optimize references to the symbol without being impeded as it would be
2627 by <tt>@llvm.used</tt>.</p>
2629 <p>This is a rare construct that should only be used in rare circumstances, and
2630 should not be exposed to source languages.</p>
2634 <!-- ======================================================================= -->
2635 <div class="doc_subsection">
2636 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2639 <div class="doc_text">
2641 <p>TODO: Describe this.</p>
2645 <!-- ======================================================================= -->
2646 <div class="doc_subsection">
2647 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2650 <div class="doc_text">
2652 <p>TODO: Describe this.</p>
2657 <!-- *********************************************************************** -->
2658 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2659 <!-- *********************************************************************** -->
2661 <div class="doc_text">
2663 <p>The LLVM instruction set consists of several different classifications of
2664 instructions: <a href="#terminators">terminator
2665 instructions</a>, <a href="#binaryops">binary instructions</a>,
2666 <a href="#bitwiseops">bitwise binary instructions</a>,
2667 <a href="#memoryops">memory instructions</a>, and
2668 <a href="#otherops">other instructions</a>.</p>
2672 <!-- ======================================================================= -->
2673 <div class="doc_subsection"> <a name="terminators">Terminator
2674 Instructions</a> </div>
2676 <div class="doc_text">
2678 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
2679 in a program ends with a "Terminator" instruction, which indicates which
2680 block should be executed after the current block is finished. These
2681 terminator instructions typically yield a '<tt>void</tt>' value: they produce
2682 control flow, not values (the one exception being the
2683 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2685 <p>There are six different terminator instructions: the
2686 '<a href="#i_ret"><tt>ret</tt></a>' instruction, the
2687 '<a href="#i_br"><tt>br</tt></a>' instruction, the
2688 '<a href="#i_switch"><tt>switch</tt></a>' instruction, the
2689 '<a href="#i_indirectbr">'<tt>indirectbr</tt></a>' Instruction, the
2690 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the
2691 '<a href="#i_unwind"><tt>unwind</tt></a>' instruction, and the
2692 '<a href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2696 <!-- _______________________________________________________________________ -->
2697 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2698 Instruction</a> </div>
2700 <div class="doc_text">
2704 ret <type> <value> <i>; Return a value from a non-void function</i>
2705 ret void <i>; Return from void function</i>
2709 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
2710 a value) from a function back to the caller.</p>
2712 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
2713 value and then causes control flow, and one that just causes control flow to
2717 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
2718 return value. The type of the return value must be a
2719 '<a href="#t_firstclass">first class</a>' type.</p>
2721 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
2722 non-void return type and contains a '<tt>ret</tt>' instruction with no return
2723 value or a return value with a type that does not match its type, or if it
2724 has a void return type and contains a '<tt>ret</tt>' instruction with a
2728 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
2729 the calling function's context. If the caller is a
2730 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
2731 instruction after the call. If the caller was an
2732 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
2733 the beginning of the "normal" destination block. If the instruction returns
2734 a value, that value shall set the call or invoke instruction's return
2739 ret i32 5 <i>; Return an integer value of 5</i>
2740 ret void <i>; Return from a void function</i>
2741 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2745 <!-- _______________________________________________________________________ -->
2746 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2748 <div class="doc_text">
2752 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2756 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
2757 different basic block in the current function. There are two forms of this
2758 instruction, corresponding to a conditional branch and an unconditional
2762 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
2763 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
2764 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
2768 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2769 argument is evaluated. If the value is <tt>true</tt>, control flows to the
2770 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2771 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2776 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
2777 br i1 %cond, label %IfEqual, label %IfUnequal
2779 <a href="#i_ret">ret</a> i32 1
2781 <a href="#i_ret">ret</a> i32 0
2786 <!-- _______________________________________________________________________ -->
2787 <div class="doc_subsubsection">
2788 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2791 <div class="doc_text">
2795 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2799 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2800 several different places. It is a generalization of the '<tt>br</tt>'
2801 instruction, allowing a branch to occur to one of many possible
2805 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2806 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
2807 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
2808 The table is not allowed to contain duplicate constant entries.</p>
2811 <p>The <tt>switch</tt> instruction specifies a table of values and
2812 destinations. When the '<tt>switch</tt>' instruction is executed, this table
2813 is searched for the given value. If the value is found, control flow is
2814 transferred to the corresponding destination; otherwise, control flow is
2815 transferred to the default destination.</p>
2817 <h5>Implementation:</h5>
2818 <p>Depending on properties of the target machine and the particular
2819 <tt>switch</tt> instruction, this instruction may be code generated in
2820 different ways. For example, it could be generated as a series of chained
2821 conditional branches or with a lookup table.</p>
2825 <i>; Emulate a conditional br instruction</i>
2826 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2827 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2829 <i>; Emulate an unconditional br instruction</i>
2830 switch i32 0, label %dest [ ]
2832 <i>; Implement a jump table:</i>
2833 switch i32 %val, label %otherwise [ i32 0, label %onzero
2835 i32 2, label %ontwo ]
2841 <!-- _______________________________________________________________________ -->
2842 <div class="doc_subsubsection">
2843 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
2846 <div class="doc_text">
2850 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
2855 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
2856 within the current function, whose address is specified by
2857 "<tt>address</tt>". Address must be derived from a <a
2858 href="#blockaddress">blockaddress</a> constant.</p>
2862 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
2863 rest of the arguments indicate the full set of possible destinations that the
2864 address may point to. Blocks are allowed to occur multiple times in the
2865 destination list, though this isn't particularly useful.</p>
2867 <p>This destination list is required so that dataflow analysis has an accurate
2868 understanding of the CFG.</p>
2872 <p>Control transfers to the block specified in the address argument. All
2873 possible destination blocks must be listed in the label list, otherwise this
2874 instruction has undefined behavior. This implies that jumps to labels
2875 defined in other functions have undefined behavior as well.</p>
2877 <h5>Implementation:</h5>
2879 <p>This is typically implemented with a jump through a register.</p>
2883 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
2889 <!-- _______________________________________________________________________ -->
2890 <div class="doc_subsubsection">
2891 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2894 <div class="doc_text">
2898 <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>]
2899 to label <normal label> unwind label <exception label>
2903 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2904 function, with the possibility of control flow transfer to either the
2905 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
2906 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
2907 control flow will return to the "normal" label. If the callee (or any
2908 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
2909 instruction, control is interrupted and continued at the dynamically nearest
2910 "exception" label.</p>
2913 <p>This instruction requires several arguments:</p>
2916 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
2917 convention</a> the call should use. If none is specified, the call
2918 defaults to using C calling conventions.</li>
2920 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2921 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
2922 '<tt>inreg</tt>' attributes are valid here.</li>
2924 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2925 function value being invoked. In most cases, this is a direct function
2926 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
2927 off an arbitrary pointer to function value.</li>
2929 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2930 function to be invoked. </li>
2932 <li>'<tt>function args</tt>': argument list whose types match the function
2933 signature argument types and parameter attributes. All arguments must be
2934 of <a href="#t_firstclass">first class</a> type. If the function
2935 signature indicates the function accepts a variable number of arguments,
2936 the extra arguments can be specified.</li>
2938 <li>'<tt>normal label</tt>': the label reached when the called function
2939 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2941 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2942 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2944 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2945 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2946 '<tt>readnone</tt>' attributes are valid here.</li>
2950 <p>This instruction is designed to operate as a standard
2951 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
2952 primary difference is that it establishes an association with a label, which
2953 is used by the runtime library to unwind the stack.</p>
2955 <p>This instruction is used in languages with destructors to ensure that proper
2956 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2957 exception. Additionally, this is important for implementation of
2958 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2960 <p>For the purposes of the SSA form, the definition of the value returned by the
2961 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
2962 block to the "normal" label. If the callee unwinds then no return value is
2965 <p>Note that the code generator does not yet completely support unwind, and
2966 that the invoke/unwind semantics are likely to change in future versions.</p>
2970 %retval = invoke i32 @Test(i32 15) to label %Continue
2971 unwind label %TestCleanup <i>; {i32}:retval set</i>
2972 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2973 unwind label %TestCleanup <i>; {i32}:retval set</i>
2978 <!-- _______________________________________________________________________ -->
2980 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2981 Instruction</a> </div>
2983 <div class="doc_text">
2991 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2992 at the first callee in the dynamic call stack which used
2993 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
2994 This is primarily used to implement exception handling.</p>
2997 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2998 immediately halt. The dynamic call stack is then searched for the
2999 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
3000 Once found, execution continues at the "exceptional" destination block
3001 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
3002 instruction in the dynamic call chain, undefined behavior results.</p>
3004 <p>Note that the code generator does not yet completely support unwind, and
3005 that the invoke/unwind semantics are likely to change in future versions.</p>
3009 <!-- _______________________________________________________________________ -->
3011 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
3012 Instruction</a> </div>
3014 <div class="doc_text">
3022 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3023 instruction is used to inform the optimizer that a particular portion of the
3024 code is not reachable. This can be used to indicate that the code after a
3025 no-return function cannot be reached, and other facts.</p>
3028 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3032 <!-- ======================================================================= -->
3033 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
3035 <div class="doc_text">
3037 <p>Binary operators are used to do most of the computation in a program. They
3038 require two operands of the same type, execute an operation on them, and
3039 produce a single value. The operands might represent multiple data, as is
3040 the case with the <a href="#t_vector">vector</a> data type. The result value
3041 has the same type as its operands.</p>
3043 <p>There are several different binary operators:</p>
3047 <!-- _______________________________________________________________________ -->
3048 <div class="doc_subsubsection">
3049 <a name="i_add">'<tt>add</tt>' Instruction</a>
3052 <div class="doc_text">
3056 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3057 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3058 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3059 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3063 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3066 <p>The two arguments to the '<tt>add</tt>' instruction must
3067 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3068 integer values. Both arguments must have identical types.</p>
3071 <p>The value produced is the integer sum of the two operands.</p>
3073 <p>If the sum has unsigned overflow, the result returned is the mathematical
3074 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3076 <p>Because LLVM integers use a two's complement representation, this instruction
3077 is appropriate for both signed and unsigned integers.</p>
3079 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3080 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3081 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3082 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
3086 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3091 <!-- _______________________________________________________________________ -->
3092 <div class="doc_subsubsection">
3093 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3096 <div class="doc_text">
3100 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3104 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3107 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3108 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3109 floating point values. Both arguments must have identical types.</p>
3112 <p>The value produced is the floating point sum of the two operands.</p>
3116 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3121 <!-- _______________________________________________________________________ -->
3122 <div class="doc_subsubsection">
3123 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3126 <div class="doc_text">
3130 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3131 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3132 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3133 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3137 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3140 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3141 '<tt>neg</tt>' instruction present in most other intermediate
3142 representations.</p>
3145 <p>The two arguments to the '<tt>sub</tt>' instruction must
3146 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3147 integer values. Both arguments must have identical types.</p>
3150 <p>The value produced is the integer difference of the two operands.</p>
3152 <p>If the difference has unsigned overflow, the result returned is the
3153 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3156 <p>Because LLVM integers use a two's complement representation, this instruction
3157 is appropriate for both signed and unsigned integers.</p>
3159 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3160 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3161 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3162 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
3166 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3167 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3172 <!-- _______________________________________________________________________ -->
3173 <div class="doc_subsubsection">
3174 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3177 <div class="doc_text">
3181 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3185 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3188 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3189 '<tt>fneg</tt>' instruction present in most other intermediate
3190 representations.</p>
3193 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3194 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3195 floating point values. Both arguments must have identical types.</p>
3198 <p>The value produced is the floating point difference of the two operands.</p>
3202 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3203 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3208 <!-- _______________________________________________________________________ -->
3209 <div class="doc_subsubsection">
3210 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3213 <div class="doc_text">
3217 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3218 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3219 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3220 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3224 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3227 <p>The two arguments to the '<tt>mul</tt>' instruction must
3228 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3229 integer values. Both arguments must have identical types.</p>
3232 <p>The value produced is the integer product of the two operands.</p>
3234 <p>If the result of the multiplication has unsigned overflow, the result
3235 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3236 width of the result.</p>
3238 <p>Because LLVM integers use a two's complement representation, and the result
3239 is the same width as the operands, this instruction returns the correct
3240 result for both signed and unsigned integers. If a full product
3241 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3242 be sign-extended or zero-extended as appropriate to the width of the full
3245 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3246 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3247 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3248 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
3252 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3257 <!-- _______________________________________________________________________ -->
3258 <div class="doc_subsubsection">
3259 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3262 <div class="doc_text">
3266 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3270 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3273 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3274 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3275 floating point values. Both arguments must have identical types.</p>
3278 <p>The value produced is the floating point product of the two operands.</p>
3282 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3287 <!-- _______________________________________________________________________ -->
3288 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
3291 <div class="doc_text">
3295 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3299 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3302 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3303 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3304 values. Both arguments must have identical types.</p>
3307 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3309 <p>Note that unsigned integer division and signed integer division are distinct
3310 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3312 <p>Division by zero leads to undefined behavior.</p>
3316 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3321 <!-- _______________________________________________________________________ -->
3322 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
3325 <div class="doc_text">
3329 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3330 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3334 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3337 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3338 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3339 values. Both arguments must have identical types.</p>
3342 <p>The value produced is the signed integer quotient of the two operands rounded
3345 <p>Note that signed integer division and unsigned integer division are distinct
3346 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3348 <p>Division by zero leads to undefined behavior. Overflow also leads to
3349 undefined behavior; this is a rare case, but can occur, for example, by doing
3350 a 32-bit division of -2147483648 by -1.</p>
3352 <p>If the <tt>exact</tt> keyword is present, the result value of the
3353 <tt>sdiv</tt> is undefined if the result would be rounded or if overflow
3358 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3363 <!-- _______________________________________________________________________ -->
3364 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
3365 Instruction</a> </div>
3367 <div class="doc_text">
3371 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3375 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3378 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3379 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3380 floating point values. Both arguments must have identical types.</p>
3383 <p>The value produced is the floating point quotient of the two operands.</p>
3387 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3392 <!-- _______________________________________________________________________ -->
3393 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3396 <div class="doc_text">
3400 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3404 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3405 division of its two arguments.</p>
3408 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3409 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3410 values. Both arguments must have identical types.</p>
3413 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3414 This instruction always performs an unsigned division to get the
3417 <p>Note that unsigned integer remainder and signed integer remainder are
3418 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3420 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3424 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3429 <!-- _______________________________________________________________________ -->
3430 <div class="doc_subsubsection">
3431 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3434 <div class="doc_text">
3438 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3442 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3443 division of its two operands. This instruction can also take
3444 <a href="#t_vector">vector</a> versions of the values in which case the
3445 elements must be integers.</p>
3448 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3449 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3450 values. Both arguments must have identical types.</p>
3453 <p>This instruction returns the <i>remainder</i> of a division (where the result
3454 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
3455 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
3456 a value. For more information about the difference,
3457 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3458 Math Forum</a>. For a table of how this is implemented in various languages,
3459 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3460 Wikipedia: modulo operation</a>.</p>
3462 <p>Note that signed integer remainder and unsigned integer remainder are
3463 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3465 <p>Taking the remainder of a division by zero leads to undefined behavior.
3466 Overflow also leads to undefined behavior; this is a rare case, but can
3467 occur, for example, by taking the remainder of a 32-bit division of
3468 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3469 lets srem be implemented using instructions that return both the result of
3470 the division and the remainder.)</p>
3474 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3479 <!-- _______________________________________________________________________ -->
3480 <div class="doc_subsubsection">
3481 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
3483 <div class="doc_text">
3487 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3491 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3492 its two operands.</p>
3495 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3496 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3497 floating point values. Both arguments must have identical types.</p>
3500 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3501 has the same sign as the dividend.</p>
3505 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3510 <!-- ======================================================================= -->
3511 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
3512 Operations</a> </div>
3514 <div class="doc_text">
3516 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3517 program. They are generally very efficient instructions and can commonly be
3518 strength reduced from other instructions. They require two operands of the
3519 same type, execute an operation on them, and produce a single value. The
3520 resulting value is the same type as its operands.</p>
3524 <!-- _______________________________________________________________________ -->
3525 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
3526 Instruction</a> </div>
3528 <div class="doc_text">
3532 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3536 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3537 a specified number of bits.</p>
3540 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3541 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3542 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3545 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3546 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3547 is (statically or dynamically) negative or equal to or larger than the number
3548 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3549 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3550 shift amount in <tt>op2</tt>.</p>
3554 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3555 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3556 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3557 <result> = shl i32 1, 32 <i>; undefined</i>
3558 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3563 <!-- _______________________________________________________________________ -->
3564 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
3565 Instruction</a> </div>
3567 <div class="doc_text">
3571 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3575 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
3576 operand shifted to the right a specified number of bits with zero fill.</p>
3579 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
3580 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3581 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3584 <p>This instruction always performs a logical shift right operation. The most
3585 significant bits of the result will be filled with zero bits after the shift.
3586 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
3587 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3588 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3589 shift amount in <tt>op2</tt>.</p>
3593 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
3594 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
3595 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
3596 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
3597 <result> = lshr i32 1, 32 <i>; undefined</i>
3598 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3603 <!-- _______________________________________________________________________ -->
3604 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
3605 Instruction</a> </div>
3606 <div class="doc_text">
3610 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3614 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
3615 operand shifted to the right a specified number of bits with sign
3619 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
3620 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3621 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3624 <p>This instruction always performs an arithmetic shift right operation, The
3625 most significant bits of the result will be filled with the sign bit
3626 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
3627 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
3628 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
3629 the corresponding shift amount in <tt>op2</tt>.</p>
3633 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
3634 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
3635 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
3636 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
3637 <result> = ashr i32 1, 32 <i>; undefined</i>
3638 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3643 <!-- _______________________________________________________________________ -->
3644 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
3645 Instruction</a> </div>
3647 <div class="doc_text">
3651 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3655 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
3659 <p>The two arguments to the '<tt>and</tt>' instruction must be
3660 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3661 values. Both arguments must have identical types.</p>
3664 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
3666 <table border="1" cellspacing="0" cellpadding="4">
3698 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
3699 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
3700 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
3703 <!-- _______________________________________________________________________ -->
3704 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
3706 <div class="doc_text">
3710 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3714 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
3718 <p>The two arguments to the '<tt>or</tt>' instruction must be
3719 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3720 values. Both arguments must have identical types.</p>
3723 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
3725 <table border="1" cellspacing="0" cellpadding="4">
3757 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3758 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3759 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3764 <!-- _______________________________________________________________________ -->
3765 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3766 Instruction</a> </div>
3768 <div class="doc_text">
3772 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3776 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
3777 its two operands. The <tt>xor</tt> is used to implement the "one's
3778 complement" operation, which is the "~" operator in C.</p>
3781 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3782 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3783 values. Both arguments must have identical types.</p>
3786 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3788 <table border="1" cellspacing="0" cellpadding="4">
3820 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3821 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3822 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3823 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3828 <!-- ======================================================================= -->
3829 <div class="doc_subsection">
3830 <a name="vectorops">Vector Operations</a>
3833 <div class="doc_text">
3835 <p>LLVM supports several instructions to represent vector operations in a
3836 target-independent manner. These instructions cover the element-access and
3837 vector-specific operations needed to process vectors effectively. While LLVM
3838 does directly support these vector operations, many sophisticated algorithms
3839 will want to use target-specific intrinsics to take full advantage of a
3840 specific target.</p>
3844 <!-- _______________________________________________________________________ -->
3845 <div class="doc_subsubsection">
3846 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3849 <div class="doc_text">
3853 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3857 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
3858 from a vector at a specified index.</p>
3862 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
3863 of <a href="#t_vector">vector</a> type. The second operand is an index
3864 indicating the position from which to extract the element. The index may be
3868 <p>The result is a scalar of the same type as the element type of
3869 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3870 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3871 results are undefined.</p>
3875 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3880 <!-- _______________________________________________________________________ -->
3881 <div class="doc_subsubsection">
3882 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3885 <div class="doc_text">
3889 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3893 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
3894 vector at a specified index.</p>
3897 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
3898 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
3899 whose type must equal the element type of the first operand. The third
3900 operand is an index indicating the position at which to insert the value.
3901 The index may be a variable.</p>
3904 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
3905 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
3906 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3907 results are undefined.</p>
3911 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3916 <!-- _______________________________________________________________________ -->
3917 <div class="doc_subsubsection">
3918 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3921 <div class="doc_text">
3925 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3929 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3930 from two input vectors, returning a vector with the same element type as the
3931 input and length that is the same as the shuffle mask.</p>
3934 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3935 with types that match each other. The third argument is a shuffle mask whose
3936 element type is always 'i32'. The result of the instruction is a vector
3937 whose length is the same as the shuffle mask and whose element type is the
3938 same as the element type of the first two operands.</p>
3940 <p>The shuffle mask operand is required to be a constant vector with either
3941 constant integer or undef values.</p>
3944 <p>The elements of the two input vectors are numbered from left to right across
3945 both of the vectors. The shuffle mask operand specifies, for each element of
3946 the result vector, which element of the two input vectors the result element
3947 gets. The element selector may be undef (meaning "don't care") and the
3948 second operand may be undef if performing a shuffle from only one vector.</p>
3952 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3953 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3954 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
3955 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3956 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
3957 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3958 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3959 <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>
3964 <!-- ======================================================================= -->
3965 <div class="doc_subsection">
3966 <a name="aggregateops">Aggregate Operations</a>
3969 <div class="doc_text">
3971 <p>LLVM supports several instructions for working with
3972 <a href="#t_aggregate">aggregate</a> values.</p>
3976 <!-- _______________________________________________________________________ -->
3977 <div class="doc_subsubsection">
3978 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3981 <div class="doc_text">
3985 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3989 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
3990 from an <a href="#t_aggregate">aggregate</a> value.</p>
3993 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
3994 of <a href="#t_struct">struct</a>, <a href="#t_union">union</a> or
3995 <a href="#t_array">array</a> type. The operands are constant indices to
3996 specify which value to extract in a similar manner as indices in a
3997 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4000 <p>The result is the value at the position in the aggregate specified by the
4005 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4010 <!-- _______________________________________________________________________ -->
4011 <div class="doc_subsubsection">
4012 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4015 <div class="doc_text">
4019 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx> <i>; yields <aggregate type></i>
4023 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4024 in an <a href="#t_aggregate">aggregate</a> value.</p>
4027 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4028 of <a href="#t_struct">struct</a>, <a href="#t_union">union</a> or
4029 <a href="#t_array">array</a> type. The second operand is a first-class
4030 value to insert. The following operands are constant indices indicating
4031 the position at which to insert the value in a similar manner as indices in a
4032 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction. The
4033 value to insert must have the same type as the value identified by the
4037 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4038 that of <tt>val</tt> except that the value at the position specified by the
4039 indices is that of <tt>elt</tt>.</p>
4043 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4044 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4050 <!-- ======================================================================= -->
4051 <div class="doc_subsection">
4052 <a name="memoryops">Memory Access and Addressing Operations</a>
4055 <div class="doc_text">
4057 <p>A key design point of an SSA-based representation is how it represents
4058 memory. In LLVM, no memory locations are in SSA form, which makes things
4059 very simple. This section describes how to read, write, and allocate
4064 <!-- _______________________________________________________________________ -->
4065 <div class="doc_subsubsection">
4066 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4069 <div class="doc_text">
4073 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4077 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4078 currently executing function, to be automatically released when this function
4079 returns to its caller. The object is always allocated in the generic address
4080 space (address space zero).</p>
4083 <p>The '<tt>alloca</tt>' instruction
4084 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4085 runtime stack, returning a pointer of the appropriate type to the program.
4086 If "NumElements" is specified, it is the number of elements allocated,
4087 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4088 specified, the value result of the allocation is guaranteed to be aligned to
4089 at least that boundary. If not specified, or if zero, the target can choose
4090 to align the allocation on any convenient boundary compatible with the
4093 <p>'<tt>type</tt>' may be any sized type.</p>
4096 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4097 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4098 memory is automatically released when the function returns. The
4099 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4100 variables that must have an address available. When the function returns
4101 (either with the <tt><a href="#i_ret">ret</a></tt>
4102 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
4103 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4107 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4108 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4109 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4110 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4115 <!-- _______________________________________________________________________ -->
4116 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
4117 Instruction</a> </div>
4119 <div class="doc_text">
4123 <result> = load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4124 <result> = volatile load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4125 !<index> = !{ i32 1 }
4129 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4132 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4133 from which to load. The pointer must point to
4134 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4135 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4136 number or order of execution of this <tt>load</tt> with other
4137 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
4140 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4141 operation (that is, the alignment of the memory address). A value of 0 or an
4142 omitted <tt>align</tt> argument means that the operation has the preferential
4143 alignment for the target. It is the responsibility of the code emitter to
4144 ensure that the alignment information is correct. Overestimating the
4145 alignment results in undefined behavior. Underestimating the alignment may
4146 produce less efficient code. An alignment of 1 is always safe.</p>
4148 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4149 metatadata name <index> corresponding to a metadata node with
4150 one <tt>i32</tt> entry of value 1. The existence of
4151 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4152 and code generator that this load is not expected to be reused in the cache.
4153 The code generator may select special instructions to save cache bandwidth,
4154 such as the <tt>MOVNT</tt> instruction on x86.</p>
4157 <p>The location of memory pointed to is loaded. If the value being loaded is of
4158 scalar type then the number of bytes read does not exceed the minimum number
4159 of bytes needed to hold all bits of the type. For example, loading an
4160 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4161 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4162 is undefined if the value was not originally written using a store of the
4167 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4168 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4169 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4174 <!-- _______________________________________________________________________ -->
4175 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
4176 Instruction</a> </div>
4178 <div class="doc_text">
4182 store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4183 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4187 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4190 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4191 and an address at which to store it. The type of the
4192 '<tt><pointer></tt>' operand must be a pointer to
4193 the <a href="#t_firstclass">first class</a> type of the
4194 '<tt><value></tt>' operand. If the <tt>store</tt> is marked
4195 as <tt>volatile</tt>, then the optimizer is not allowed to modify the number
4196 or order of execution of this <tt>store</tt> with other
4197 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
4200 <p>The optional constant "align" argument specifies the alignment of the
4201 operation (that is, the alignment of the memory address). A value of 0 or an
4202 omitted "align" argument means that the operation has the preferential
4203 alignment for the target. It is the responsibility of the code emitter to
4204 ensure that the alignment information is correct. Overestimating the
4205 alignment results in an undefined behavior. Underestimating the alignment may
4206 produce less efficient code. An alignment of 1 is always safe.</p>
4208 <p>The optional !nontemporal metadata must reference a single metatadata
4209 name <index> corresponding to a metadata node with one i32 entry of
4210 value 1. The existence of the !nontemporal metatadata on the
4211 instruction tells the optimizer and code generator that this load is
4212 not expected to be reused in the cache. The code generator may
4213 select special instructions to save cache bandwidth, such as the
4214 MOVNT instruction on x86.</p>
4218 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4219 location specified by the '<tt><pointer></tt>' operand. If
4220 '<tt><value></tt>' is of scalar type then the number of bytes written
4221 does not exceed the minimum number of bytes needed to hold all bits of the
4222 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4223 writing a value of a type like <tt>i20</tt> with a size that is not an
4224 integral number of bytes, it is unspecified what happens to the extra bits
4225 that do not belong to the type, but they will typically be overwritten.</p>
4229 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4230 store i32 3, i32* %ptr <i>; yields {void}</i>
4231 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4236 <!-- _______________________________________________________________________ -->
4237 <div class="doc_subsubsection">
4238 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
4241 <div class="doc_text">
4245 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4246 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4250 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4251 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
4252 It performs address calculation only and does not access memory.</p>
4255 <p>The first argument is always a pointer, and forms the basis of the
4256 calculation. The remaining arguments are indices that indicate which of the
4257 elements of the aggregate object are indexed. The interpretation of each
4258 index is dependent on the type being indexed into. The first index always
4259 indexes the pointer value given as the first argument, the second index
4260 indexes a value of the type pointed to (not necessarily the value directly
4261 pointed to, since the first index can be non-zero), etc. The first type
4262 indexed into must be a pointer value, subsequent types can be arrays,
4263 vectors, structs and unions. Note that subsequent types being indexed into
4264 can never be pointers, since that would require loading the pointer before
4265 continuing calculation.</p>
4267 <p>The type of each index argument depends on the type it is indexing into.
4268 When indexing into a (optionally packed) structure or union, only <tt>i32</tt>
4269 integer <b>constants</b> are allowed. When indexing into an array, pointer
4270 or vector, integers of any width are allowed, and they are not required to be
4273 <p>For example, let's consider a C code fragment and how it gets compiled to
4276 <div class="doc_code">
4289 int *foo(struct ST *s) {
4290 return &s[1].Z.B[5][13];
4295 <p>The LLVM code generated by the GCC frontend is:</p>
4297 <div class="doc_code">
4299 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4300 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
4302 define i32* @foo(%ST* %s) {
4304 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
4311 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
4312 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
4313 }</tt>' type, a structure. The second index indexes into the third element
4314 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
4315 i8 }</tt>' type, another structure. The third index indexes into the second
4316 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
4317 array. The two dimensions of the array are subscripted into, yielding an
4318 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
4319 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
4321 <p>Note that it is perfectly legal to index partially through a structure,
4322 returning a pointer to an inner element. Because of this, the LLVM code for
4323 the given testcase is equivalent to:</p>
4326 define i32* @foo(%ST* %s) {
4327 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
4328 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
4329 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
4330 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
4331 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
4336 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
4337 <tt>getelementptr</tt> is undefined if the base pointer is not an
4338 <i>in bounds</i> address of an allocated object, or if any of the addresses
4339 that would be formed by successive addition of the offsets implied by the
4340 indices to the base address with infinitely precise arithmetic are not an
4341 <i>in bounds</i> address of that allocated object.
4342 The <i>in bounds</i> addresses for an allocated object are all the addresses
4343 that point into the object, plus the address one byte past the end.</p>
4345 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
4346 the base address with silently-wrapping two's complement arithmetic, and
4347 the result value of the <tt>getelementptr</tt> may be outside the object
4348 pointed to by the base pointer. The result value may not necessarily be
4349 used to access memory though, even if it happens to point into allocated
4350 storage. See the <a href="#pointeraliasing">Pointer Aliasing Rules</a>
4351 section for more information.</p>
4353 <p>The getelementptr instruction is often confusing. For some more insight into
4354 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
4358 <i>; yields [12 x i8]*:aptr</i>
4359 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4360 <i>; yields i8*:vptr</i>
4361 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4362 <i>; yields i8*:eptr</i>
4363 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4364 <i>; yields i32*:iptr</i>
4365 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4370 <!-- ======================================================================= -->
4371 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
4374 <div class="doc_text">
4376 <p>The instructions in this category are the conversion instructions (casting)
4377 which all take a single operand and a type. They perform various bit
4378 conversions on the operand.</p>
4382 <!-- _______________________________________________________________________ -->
4383 <div class="doc_subsubsection">
4384 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
4386 <div class="doc_text">
4390 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
4394 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
4395 type <tt>ty2</tt>.</p>
4398 <p>The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
4399 be an <a href="#t_integer">integer</a> type, and a type that specifies the
4400 size and type of the result, which must be
4401 an <a href="#t_integer">integer</a> type. The bit size of <tt>value</tt> must
4402 be larger than the bit size of <tt>ty2</tt>. Equal sized types are not
4406 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
4407 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
4408 source size must be larger than the destination size, <tt>trunc</tt> cannot
4409 be a <i>no-op cast</i>. It will always truncate bits.</p>
4413 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
4414 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
4415 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
4420 <!-- _______________________________________________________________________ -->
4421 <div class="doc_subsubsection">
4422 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
4424 <div class="doc_text">
4428 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
4432 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
4437 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
4438 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4439 also be of <a href="#t_integer">integer</a> type. The bit size of the
4440 <tt>value</tt> must be smaller than the bit size of the destination type,
4444 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
4445 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
4447 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
4451 %X = zext i32 257 to i64 <i>; yields i64:257</i>
4452 %Y = zext i1 true to i32 <i>; yields i32:1</i>
4457 <!-- _______________________________________________________________________ -->
4458 <div class="doc_subsubsection">
4459 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
4461 <div class="doc_text">
4465 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
4469 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
4472 <p>The '<tt>sext</tt>' instruction takes a value to cast, which must be of
4473 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4474 also be of <a href="#t_integer">integer</a> type. The bit size of the
4475 <tt>value</tt> must be smaller than the bit size of the destination type,
4479 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
4480 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
4481 of the type <tt>ty2</tt>.</p>
4483 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
4487 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
4488 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
4493 <!-- _______________________________________________________________________ -->
4494 <div class="doc_subsubsection">
4495 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
4498 <div class="doc_text">
4502 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
4506 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
4510 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
4511 point</a> value to cast and a <a href="#t_floating">floating point</a> type
4512 to cast it to. The size of <tt>value</tt> must be larger than the size of
4513 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
4514 <i>no-op cast</i>.</p>
4517 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
4518 <a href="#t_floating">floating point</a> type to a smaller
4519 <a href="#t_floating">floating point</a> type. If the value cannot fit
4520 within the destination type, <tt>ty2</tt>, then the results are
4525 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
4526 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
4531 <!-- _______________________________________________________________________ -->
4532 <div class="doc_subsubsection">
4533 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
4535 <div class="doc_text">
4539 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
4543 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
4544 floating point value.</p>
4547 <p>The '<tt>fpext</tt>' instruction takes a
4548 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
4549 a <a href="#t_floating">floating point</a> type to cast it to. The source
4550 type must be smaller than the destination type.</p>
4553 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
4554 <a href="#t_floating">floating point</a> type to a larger
4555 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
4556 used to make a <i>no-op cast</i> because it always changes bits. Use
4557 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
4561 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
4562 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
4567 <!-- _______________________________________________________________________ -->
4568 <div class="doc_subsubsection">
4569 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
4571 <div class="doc_text">
4575 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
4579 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
4580 unsigned integer equivalent of type <tt>ty2</tt>.</p>
4583 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
4584 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4585 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4586 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4587 vector integer type with the same number of elements as <tt>ty</tt></p>
4590 <p>The '<tt>fptoui</tt>' instruction converts its
4591 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4592 towards zero) unsigned integer value. If the value cannot fit
4593 in <tt>ty2</tt>, the results are undefined.</p>
4597 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
4598 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
4599 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
4604 <!-- _______________________________________________________________________ -->
4605 <div class="doc_subsubsection">
4606 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
4608 <div class="doc_text">
4612 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
4616 <p>The '<tt>fptosi</tt>' instruction converts
4617 <a href="#t_floating">floating point</a> <tt>value</tt> to
4618 type <tt>ty2</tt>.</p>
4621 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
4622 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4623 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4624 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4625 vector integer type with the same number of elements as <tt>ty</tt></p>
4628 <p>The '<tt>fptosi</tt>' instruction converts its
4629 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4630 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
4631 the results are undefined.</p>
4635 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
4636 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
4637 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
4642 <!-- _______________________________________________________________________ -->
4643 <div class="doc_subsubsection">
4644 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
4646 <div class="doc_text">
4650 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4654 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4655 integer and converts that value to the <tt>ty2</tt> type.</p>
4658 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4659 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4660 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4661 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4662 floating point type with the same number of elements as <tt>ty</tt></p>
4665 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4666 integer quantity and converts it to the corresponding floating point
4667 value. If the value cannot fit in the floating point value, the results are
4672 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4673 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4678 <!-- _______________________________________________________________________ -->
4679 <div class="doc_subsubsection">
4680 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4682 <div class="doc_text">
4686 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4690 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
4691 and converts that value to the <tt>ty2</tt> type.</p>
4694 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4695 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4696 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4697 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4698 floating point type with the same number of elements as <tt>ty</tt></p>
4701 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
4702 quantity and converts it to the corresponding floating point value. If the
4703 value cannot fit in the floating point value, the results are undefined.</p>
4707 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4708 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4713 <!-- _______________________________________________________________________ -->
4714 <div class="doc_subsubsection">
4715 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4717 <div class="doc_text">
4721 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4725 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4726 the integer type <tt>ty2</tt>.</p>
4729 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4730 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4731 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4734 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4735 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4736 truncating or zero extending that value to the size of the integer type. If
4737 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4738 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4739 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4744 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4745 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4750 <!-- _______________________________________________________________________ -->
4751 <div class="doc_subsubsection">
4752 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4754 <div class="doc_text">
4758 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4762 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
4763 pointer type, <tt>ty2</tt>.</p>
4766 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4767 value to cast, and a type to cast it to, which must be a
4768 <a href="#t_pointer">pointer</a> type.</p>
4771 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4772 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4773 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4774 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
4775 than the size of a pointer then a zero extension is done. If they are the
4776 same size, nothing is done (<i>no-op cast</i>).</p>
4780 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4781 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4782 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4787 <!-- _______________________________________________________________________ -->
4788 <div class="doc_subsubsection">
4789 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4791 <div class="doc_text">
4795 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4799 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4800 <tt>ty2</tt> without changing any bits.</p>
4803 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
4804 non-aggregate first class value, and a type to cast it to, which must also be
4805 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
4806 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
4807 identical. If the source type is a pointer, the destination type must also be
4808 a pointer. This instruction supports bitwise conversion of vectors to
4809 integers and to vectors of other types (as long as they have the same
4813 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4814 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4815 this conversion. The conversion is done as if the <tt>value</tt> had been
4816 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
4817 be converted to other pointer types with this instruction. To convert
4818 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
4819 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4823 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4824 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4825 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4830 <!-- ======================================================================= -->
4831 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4833 <div class="doc_text">
4835 <p>The instructions in this category are the "miscellaneous" instructions, which
4836 defy better classification.</p>
4840 <!-- _______________________________________________________________________ -->
4841 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4844 <div class="doc_text">
4848 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4852 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
4853 boolean values based on comparison of its two integer, integer vector, or
4854 pointer operands.</p>
4857 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4858 the condition code indicating the kind of comparison to perform. It is not a
4859 value, just a keyword. The possible condition code are:</p>
4862 <li><tt>eq</tt>: equal</li>
4863 <li><tt>ne</tt>: not equal </li>
4864 <li><tt>ugt</tt>: unsigned greater than</li>
4865 <li><tt>uge</tt>: unsigned greater or equal</li>
4866 <li><tt>ult</tt>: unsigned less than</li>
4867 <li><tt>ule</tt>: unsigned less or equal</li>
4868 <li><tt>sgt</tt>: signed greater than</li>
4869 <li><tt>sge</tt>: signed greater or equal</li>
4870 <li><tt>slt</tt>: signed less than</li>
4871 <li><tt>sle</tt>: signed less or equal</li>
4874 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4875 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
4876 typed. They must also be identical types.</p>
4879 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
4880 condition code given as <tt>cond</tt>. The comparison performed always yields
4881 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
4882 result, as follows:</p>
4885 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4886 <tt>false</tt> otherwise. No sign interpretation is necessary or
4889 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4890 <tt>false</tt> otherwise. No sign interpretation is necessary or
4893 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4894 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4896 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4897 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4898 to <tt>op2</tt>.</li>
4900 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4901 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4903 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4904 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4906 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4907 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4909 <li><tt>sge</tt>: interprets the operands as signed values and yields
4910 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4911 to <tt>op2</tt>.</li>
4913 <li><tt>slt</tt>: interprets the operands as signed values and yields
4914 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4916 <li><tt>sle</tt>: interprets the operands as signed values and yields
4917 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4920 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4921 values are compared as if they were integers.</p>
4923 <p>If the operands are integer vectors, then they are compared element by
4924 element. The result is an <tt>i1</tt> vector with the same number of elements
4925 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
4929 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4930 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4931 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4932 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4933 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4934 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4937 <p>Note that the code generator does not yet support vector types with
4938 the <tt>icmp</tt> instruction.</p>
4942 <!-- _______________________________________________________________________ -->
4943 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4946 <div class="doc_text">
4950 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4954 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
4955 values based on comparison of its operands.</p>
4957 <p>If the operands are floating point scalars, then the result type is a boolean
4958 (<a href="#t_integer"><tt>i1</tt></a>).</p>
4960 <p>If the operands are floating point vectors, then the result type is a vector
4961 of boolean with the same number of elements as the operands being
4965 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4966 the condition code indicating the kind of comparison to perform. It is not a
4967 value, just a keyword. The possible condition code are:</p>
4970 <li><tt>false</tt>: no comparison, always returns false</li>
4971 <li><tt>oeq</tt>: ordered and equal</li>
4972 <li><tt>ogt</tt>: ordered and greater than </li>
4973 <li><tt>oge</tt>: ordered and greater than or equal</li>
4974 <li><tt>olt</tt>: ordered and less than </li>
4975 <li><tt>ole</tt>: ordered and less than or equal</li>
4976 <li><tt>one</tt>: ordered and not equal</li>
4977 <li><tt>ord</tt>: ordered (no nans)</li>
4978 <li><tt>ueq</tt>: unordered or equal</li>
4979 <li><tt>ugt</tt>: unordered or greater than </li>
4980 <li><tt>uge</tt>: unordered or greater than or equal</li>
4981 <li><tt>ult</tt>: unordered or less than </li>
4982 <li><tt>ule</tt>: unordered or less than or equal</li>
4983 <li><tt>une</tt>: unordered or not equal</li>
4984 <li><tt>uno</tt>: unordered (either nans)</li>
4985 <li><tt>true</tt>: no comparison, always returns true</li>
4988 <p><i>Ordered</i> means that neither operand is a QNAN while
4989 <i>unordered</i> means that either operand may be a QNAN.</p>
4991 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
4992 a <a href="#t_floating">floating point</a> type or
4993 a <a href="#t_vector">vector</a> of floating point type. They must have
4994 identical types.</p>
4997 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4998 according to the condition code given as <tt>cond</tt>. If the operands are
4999 vectors, then the vectors are compared element by element. Each comparison
5000 performed always yields an <a href="#t_integer">i1</a> result, as
5004 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
5006 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5007 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5009 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5010 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5012 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5013 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5015 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5016 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5018 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5019 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5021 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5022 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5024 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
5026 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
5027 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5029 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
5030 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5032 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
5033 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5035 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
5036 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5038 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
5039 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5041 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
5042 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5044 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
5046 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5051 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
5052 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5053 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5054 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
5057 <p>Note that the code generator does not yet support vector types with
5058 the <tt>fcmp</tt> instruction.</p>
5062 <!-- _______________________________________________________________________ -->
5063 <div class="doc_subsubsection">
5064 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
5067 <div class="doc_text">
5071 <result> = phi <ty> [ <val0>, <label0>], ...
5075 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
5076 SSA graph representing the function.</p>
5079 <p>The type of the incoming values is specified with the first type field. After
5080 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5081 one pair for each predecessor basic block of the current block. Only values
5082 of <a href="#t_firstclass">first class</a> type may be used as the value
5083 arguments to the PHI node. Only labels may be used as the label
5086 <p>There must be no non-phi instructions between the start of a basic block and
5087 the PHI instructions: i.e. PHI instructions must be first in a basic
5090 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5091 occur on the edge from the corresponding predecessor block to the current
5092 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5093 value on the same edge).</p>
5096 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
5097 specified by the pair corresponding to the predecessor basic block that
5098 executed just prior to the current block.</p>
5102 Loop: ; Infinite loop that counts from 0 on up...
5103 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5104 %nextindvar = add i32 %indvar, 1
5110 <!-- _______________________________________________________________________ -->
5111 <div class="doc_subsubsection">
5112 <a name="i_select">'<tt>select</tt>' Instruction</a>
5115 <div class="doc_text">
5119 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5121 <i>selty</i> is either i1 or {<N x i1>}
5125 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
5126 condition, without branching.</p>
5130 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5131 values indicating the condition, and two values of the
5132 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
5133 vectors and the condition is a scalar, then entire vectors are selected, not
5134 individual elements.</p>
5137 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
5138 first value argument; otherwise, it returns the second value argument.</p>
5140 <p>If the condition is a vector of i1, then the value arguments must be vectors
5141 of the same size, and the selection is done element by element.</p>
5145 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
5148 <p>Note that the code generator does not yet support conditions
5149 with vector type.</p>
5153 <!-- _______________________________________________________________________ -->
5154 <div class="doc_subsubsection">
5155 <a name="i_call">'<tt>call</tt>' Instruction</a>
5158 <div class="doc_text">
5162 <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>]
5166 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
5169 <p>This instruction requires several arguments:</p>
5172 <li>The optional "tail" marker indicates that the callee function does not
5173 access any allocas or varargs in the caller. Note that calls may be
5174 marked "tail" even if they do not occur before
5175 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
5176 present, the function call is eligible for tail call optimization,
5177 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
5178 optimized into a jump</a>. The code generator may optimize calls marked
5179 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
5180 sibling call optimization</a> when the caller and callee have
5181 matching signatures, or 2) forced tail call optimization when the
5182 following extra requirements are met:
5184 <li>Caller and callee both have the calling
5185 convention <tt>fastcc</tt>.</li>
5186 <li>The call is in tail position (ret immediately follows call and ret
5187 uses value of call or is void).</li>
5188 <li>Option <tt>-tailcallopt</tt> is enabled,
5189 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
5190 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
5191 constraints are met.</a></li>
5195 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5196 convention</a> the call should use. If none is specified, the call
5197 defaults to using C calling conventions. The calling convention of the
5198 call must match the calling convention of the target function, or else the
5199 behavior is undefined.</li>
5201 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5202 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5203 '<tt>inreg</tt>' attributes are valid here.</li>
5205 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5206 type of the return value. Functions that return no value are marked
5207 <tt><a href="#t_void">void</a></tt>.</li>
5209 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5210 being invoked. The argument types must match the types implied by this
5211 signature. This type can be omitted if the function is not varargs and if
5212 the function type does not return a pointer to a function.</li>
5214 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5215 be invoked. In most cases, this is a direct function invocation, but
5216 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5217 to function value.</li>
5219 <li>'<tt>function args</tt>': argument list whose types match the function
5220 signature argument types and parameter attributes. All arguments must be
5221 of <a href="#t_firstclass">first class</a> type. If the function
5222 signature indicates the function accepts a variable number of arguments,
5223 the extra arguments can be specified.</li>
5225 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5226 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5227 '<tt>readnone</tt>' attributes are valid here.</li>
5231 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5232 a specified function, with its incoming arguments bound to the specified
5233 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5234 function, control flow continues with the instruction after the function
5235 call, and the return value of the function is bound to the result
5240 %retval = call i32 @test(i32 %argc)
5241 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
5242 %X = tail call i32 @foo() <i>; yields i32</i>
5243 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5244 call void %foo(i8 97 signext)
5246 %struct.A = type { i32, i8 }
5247 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
5248 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5249 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
5250 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
5251 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
5254 <p>llvm treats calls to some functions with names and arguments that match the
5255 standard C99 library as being the C99 library functions, and may perform
5256 optimizations or generate code for them under that assumption. This is
5257 something we'd like to change in the future to provide better support for
5258 freestanding environments and non-C-based languages.</p>
5262 <!-- _______________________________________________________________________ -->
5263 <div class="doc_subsubsection">
5264 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5267 <div class="doc_text">
5271 <resultval> = va_arg <va_list*> <arglist>, <argty>
5275 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
5276 the "variable argument" area of a function call. It is used to implement the
5277 <tt>va_arg</tt> macro in C.</p>
5280 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
5281 argument. It returns a value of the specified argument type and increments
5282 the <tt>va_list</tt> to point to the next argument. The actual type
5283 of <tt>va_list</tt> is target specific.</p>
5286 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
5287 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
5288 to the next argument. For more information, see the variable argument
5289 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
5291 <p>It is legal for this instruction to be called in a function which does not
5292 take a variable number of arguments, for example, the <tt>vfprintf</tt>
5295 <p><tt>va_arg</tt> is an LLVM instruction instead of
5296 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
5300 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
5302 <p>Note that the code generator does not yet fully support va_arg on many
5303 targets. Also, it does not currently support va_arg with aggregate types on
5308 <!-- *********************************************************************** -->
5309 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
5310 <!-- *********************************************************************** -->
5312 <div class="doc_text">
5314 <p>LLVM supports the notion of an "intrinsic function". These functions have
5315 well known names and semantics and are required to follow certain
5316 restrictions. Overall, these intrinsics represent an extension mechanism for
5317 the LLVM language that does not require changing all of the transformations
5318 in LLVM when adding to the language (or the bitcode reader/writer, the
5319 parser, etc...).</p>
5321 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
5322 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5323 begin with this prefix. Intrinsic functions must always be external
5324 functions: you cannot define the body of intrinsic functions. Intrinsic
5325 functions may only be used in call or invoke instructions: it is illegal to
5326 take the address of an intrinsic function. Additionally, because intrinsic
5327 functions are part of the LLVM language, it is required if any are added that
5328 they be documented here.</p>
5330 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5331 family of functions that perform the same operation but on different data
5332 types. Because LLVM can represent over 8 million different integer types,
5333 overloading is used commonly to allow an intrinsic function to operate on any
5334 integer type. One or more of the argument types or the result type can be
5335 overloaded to accept any integer type. Argument types may also be defined as
5336 exactly matching a previous argument's type or the result type. This allows
5337 an intrinsic function which accepts multiple arguments, but needs all of them
5338 to be of the same type, to only be overloaded with respect to a single
5339 argument or the result.</p>
5341 <p>Overloaded intrinsics will have the names of its overloaded argument types
5342 encoded into its function name, each preceded by a period. Only those types
5343 which are overloaded result in a name suffix. Arguments whose type is matched
5344 against another type do not. For example, the <tt>llvm.ctpop</tt> function
5345 can take an integer of any width and returns an integer of exactly the same
5346 integer width. This leads to a family of functions such as
5347 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
5348 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5349 suffix is required. Because the argument's type is matched against the return
5350 type, it does not require its own name suffix.</p>
5352 <p>To learn how to add an intrinsic function, please see the
5353 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
5357 <!-- ======================================================================= -->
5358 <div class="doc_subsection">
5359 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
5362 <div class="doc_text">
5364 <p>Variable argument support is defined in LLVM with
5365 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
5366 intrinsic functions. These functions are related to the similarly named
5367 macros defined in the <tt><stdarg.h></tt> header file.</p>
5369 <p>All of these functions operate on arguments that use a target-specific value
5370 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5371 not define what this type is, so all transformations should be prepared to
5372 handle these functions regardless of the type used.</p>
5374 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
5375 instruction and the variable argument handling intrinsic functions are
5378 <div class="doc_code">
5380 define i32 @test(i32 %X, ...) {
5381 ; Initialize variable argument processing
5383 %ap2 = bitcast i8** %ap to i8*
5384 call void @llvm.va_start(i8* %ap2)
5386 ; Read a single integer argument
5387 %tmp = va_arg i8** %ap, i32
5389 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5391 %aq2 = bitcast i8** %aq to i8*
5392 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5393 call void @llvm.va_end(i8* %aq2)
5395 ; Stop processing of arguments.
5396 call void @llvm.va_end(i8* %ap2)
5400 declare void @llvm.va_start(i8*)
5401 declare void @llvm.va_copy(i8*, i8*)
5402 declare void @llvm.va_end(i8*)
5408 <!-- _______________________________________________________________________ -->
5409 <div class="doc_subsubsection">
5410 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
5414 <div class="doc_text">
5418 declare void %llvm.va_start(i8* <arglist>)
5422 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
5423 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
5426 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
5429 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
5430 macro available in C. In a target-dependent way, it initializes
5431 the <tt>va_list</tt> element to which the argument points, so that the next
5432 call to <tt>va_arg</tt> will produce the first variable argument passed to
5433 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
5434 need to know the last argument of the function as the compiler can figure
5439 <!-- _______________________________________________________________________ -->
5440 <div class="doc_subsubsection">
5441 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
5444 <div class="doc_text">
5448 declare void @llvm.va_end(i8* <arglist>)
5452 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
5453 which has been initialized previously
5454 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
5455 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
5458 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
5461 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
5462 macro available in C. In a target-dependent way, it destroys
5463 the <tt>va_list</tt> element to which the argument points. Calls
5464 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
5465 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
5466 with calls to <tt>llvm.va_end</tt>.</p>
5470 <!-- _______________________________________________________________________ -->
5471 <div class="doc_subsubsection">
5472 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
5475 <div class="doc_text">
5479 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
5483 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
5484 from the source argument list to the destination argument list.</p>
5487 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
5488 The second argument is a pointer to a <tt>va_list</tt> element to copy
5492 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
5493 macro available in C. In a target-dependent way, it copies the
5494 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
5495 element. This intrinsic is necessary because
5496 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
5497 arbitrarily complex and require, for example, memory allocation.</p>
5501 <!-- ======================================================================= -->
5502 <div class="doc_subsection">
5503 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
5506 <div class="doc_text">
5508 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
5509 Collection</a> (GC) requires the implementation and generation of these
5510 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
5511 roots on the stack</a>, as well as garbage collector implementations that
5512 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
5513 barriers. Front-ends for type-safe garbage collected languages should generate
5514 these intrinsics to make use of the LLVM garbage collectors. For more details,
5515 see <a href="GarbageCollection.html">Accurate Garbage Collection with
5518 <p>The garbage collection intrinsics only operate on objects in the generic
5519 address space (address space zero).</p>
5523 <!-- _______________________________________________________________________ -->
5524 <div class="doc_subsubsection">
5525 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
5528 <div class="doc_text">
5532 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5536 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
5537 the code generator, and allows some metadata to be associated with it.</p>
5540 <p>The first argument specifies the address of a stack object that contains the
5541 root pointer. The second pointer (which must be either a constant or a
5542 global value address) contains the meta-data to be associated with the
5546 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
5547 location. At compile-time, the code generator generates information to allow
5548 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
5549 intrinsic may only be used in a function which <a href="#gc">specifies a GC
5554 <!-- _______________________________________________________________________ -->
5555 <div class="doc_subsubsection">
5556 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
5559 <div class="doc_text">
5563 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5567 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
5568 locations, allowing garbage collector implementations that require read
5572 <p>The second argument is the address to read from, which should be an address
5573 allocated from the garbage collector. The first object is a pointer to the
5574 start of the referenced object, if needed by the language runtime (otherwise
5578 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
5579 instruction, but may be replaced with substantially more complex code by the
5580 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
5581 may only be used in a function which <a href="#gc">specifies a GC
5586 <!-- _______________________________________________________________________ -->
5587 <div class="doc_subsubsection">
5588 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5591 <div class="doc_text">
5595 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5599 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5600 locations, allowing garbage collector implementations that require write
5601 barriers (such as generational or reference counting collectors).</p>
5604 <p>The first argument is the reference to store, the second is the start of the
5605 object to store it to, and the third is the address of the field of Obj to
5606 store to. If the runtime does not require a pointer to the object, Obj may
5610 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5611 instruction, but may be replaced with substantially more complex code by the
5612 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5613 may only be used in a function which <a href="#gc">specifies a GC
5618 <!-- ======================================================================= -->
5619 <div class="doc_subsection">
5620 <a name="int_codegen">Code Generator Intrinsics</a>
5623 <div class="doc_text">
5625 <p>These intrinsics are provided by LLVM to expose special features that may
5626 only be implemented with code generator support.</p>
5630 <!-- _______________________________________________________________________ -->
5631 <div class="doc_subsubsection">
5632 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5635 <div class="doc_text">
5639 declare i8 *@llvm.returnaddress(i32 <level>)
5643 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5644 target-specific value indicating the return address of the current function
5645 or one of its callers.</p>
5648 <p>The argument to this intrinsic indicates which function to return the address
5649 for. Zero indicates the calling function, one indicates its caller, etc.
5650 The argument is <b>required</b> to be a constant integer value.</p>
5653 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
5654 indicating the return address of the specified call frame, or zero if it
5655 cannot be identified. The value returned by this intrinsic is likely to be
5656 incorrect or 0 for arguments other than zero, so it should only be used for
5657 debugging purposes.</p>
5659 <p>Note that calling this intrinsic does not prevent function inlining or other
5660 aggressive transformations, so the value returned may not be that of the
5661 obvious source-language caller.</p>
5665 <!-- _______________________________________________________________________ -->
5666 <div class="doc_subsubsection">
5667 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5670 <div class="doc_text">
5674 declare i8 *@llvm.frameaddress(i32 <level>)
5678 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5679 target-specific frame pointer value for the specified stack frame.</p>
5682 <p>The argument to this intrinsic indicates which function to return the frame
5683 pointer for. Zero indicates the calling function, one indicates its caller,
5684 etc. The argument is <b>required</b> to be a constant integer value.</p>
5687 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
5688 indicating the frame address of the specified call frame, or zero if it
5689 cannot be identified. The value returned by this intrinsic is likely to be
5690 incorrect or 0 for arguments other than zero, so it should only be used for
5691 debugging purposes.</p>
5693 <p>Note that calling this intrinsic does not prevent function inlining or other
5694 aggressive transformations, so the value returned may not be that of the
5695 obvious source-language caller.</p>
5699 <!-- _______________________________________________________________________ -->
5700 <div class="doc_subsubsection">
5701 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5704 <div class="doc_text">
5708 declare i8 *@llvm.stacksave()
5712 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
5713 of the function stack, for use
5714 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
5715 useful for implementing language features like scoped automatic variable
5716 sized arrays in C99.</p>
5719 <p>This intrinsic returns a opaque pointer value that can be passed
5720 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
5721 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
5722 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
5723 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
5724 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
5725 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
5729 <!-- _______________________________________________________________________ -->
5730 <div class="doc_subsubsection">
5731 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5734 <div class="doc_text">
5738 declare void @llvm.stackrestore(i8 * %ptr)
5742 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5743 the function stack to the state it was in when the
5744 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
5745 executed. This is useful for implementing language features like scoped
5746 automatic variable sized arrays in C99.</p>
5749 <p>See the description
5750 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
5754 <!-- _______________________________________________________________________ -->
5755 <div class="doc_subsubsection">
5756 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5759 <div class="doc_text">
5763 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5767 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
5768 insert a prefetch instruction if supported; otherwise, it is a noop.
5769 Prefetches have no effect on the behavior of the program but can change its
5770 performance characteristics.</p>
5773 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
5774 specifier determining if the fetch should be for a read (0) or write (1),
5775 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5776 locality, to (3) - extremely local keep in cache. The <tt>rw</tt>
5777 and <tt>locality</tt> arguments must be constant integers.</p>
5780 <p>This intrinsic does not modify the behavior of the program. In particular,
5781 prefetches cannot trap and do not produce a value. On targets that support
5782 this intrinsic, the prefetch can provide hints to the processor cache for
5783 better performance.</p>
5787 <!-- _______________________________________________________________________ -->
5788 <div class="doc_subsubsection">
5789 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5792 <div class="doc_text">
5796 declare void @llvm.pcmarker(i32 <id>)
5800 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
5801 Counter (PC) in a region of code to simulators and other tools. The method
5802 is target specific, but it is expected that the marker will use exported
5803 symbols to transmit the PC of the marker. The marker makes no guarantees
5804 that it will remain with any specific instruction after optimizations. It is
5805 possible that the presence of a marker will inhibit optimizations. The
5806 intended use is to be inserted after optimizations to allow correlations of
5807 simulation runs.</p>
5810 <p><tt>id</tt> is a numerical id identifying the marker.</p>
5813 <p>This intrinsic does not modify the behavior of the program. Backends that do
5814 not support this intrinsic may ignore it.</p>
5818 <!-- _______________________________________________________________________ -->
5819 <div class="doc_subsubsection">
5820 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5823 <div class="doc_text">
5827 declare i64 @llvm.readcyclecounter( )
5831 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5832 counter register (or similar low latency, high accuracy clocks) on those
5833 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5834 should map to RPCC. As the backing counters overflow quickly (on the order
5835 of 9 seconds on alpha), this should only be used for small timings.</p>
5838 <p>When directly supported, reading the cycle counter should not modify any
5839 memory. Implementations are allowed to either return a application specific
5840 value or a system wide value. On backends without support, this is lowered
5841 to a constant 0.</p>
5845 <!-- ======================================================================= -->
5846 <div class="doc_subsection">
5847 <a name="int_libc">Standard C Library Intrinsics</a>
5850 <div class="doc_text">
5852 <p>LLVM provides intrinsics for a few important standard C library functions.
5853 These intrinsics allow source-language front-ends to pass information about
5854 the alignment of the pointer arguments to the code generator, providing
5855 opportunity for more efficient code generation.</p>
5859 <!-- _______________________________________________________________________ -->
5860 <div class="doc_subsubsection">
5861 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5864 <div class="doc_text">
5867 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
5868 integer bit width. Not all targets support all bit widths however.</p>
5871 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5872 i8 <len>, i32 <align>)
5873 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5874 i16 <len>, i32 <align>)
5875 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5876 i32 <len>, i32 <align>)
5877 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5878 i64 <len>, i32 <align>)
5882 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5883 source location to the destination location.</p>
5885 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5886 intrinsics do not return a value, and takes an extra alignment argument.</p>
5889 <p>The first argument is a pointer to the destination, the second is a pointer
5890 to the source. The third argument is an integer argument specifying the
5891 number of bytes to copy, and the fourth argument is the alignment of the
5892 source and destination locations.</p>
5894 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
5895 then the caller guarantees that both the source and destination pointers are
5896 aligned to that boundary.</p>
5899 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5900 source location to the destination location, which are not allowed to
5901 overlap. It copies "len" bytes of memory over. If the argument is known to
5902 be aligned to some boundary, this can be specified as the fourth argument,
5903 otherwise it should be set to 0 or 1.</p>
5907 <!-- _______________________________________________________________________ -->
5908 <div class="doc_subsubsection">
5909 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5912 <div class="doc_text">
5915 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5916 width. Not all targets support all bit widths however.</p>
5919 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5920 i8 <len>, i32 <align>)
5921 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5922 i16 <len>, i32 <align>)
5923 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5924 i32 <len>, i32 <align>)
5925 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5926 i64 <len>, i32 <align>)
5930 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
5931 source location to the destination location. It is similar to the
5932 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
5935 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5936 intrinsics do not return a value, and takes an extra alignment argument.</p>
5939 <p>The first argument is a pointer to the destination, the second is a pointer
5940 to the source. The third argument is an integer argument specifying the
5941 number of bytes to copy, and the fourth argument is the alignment of the
5942 source and destination locations.</p>
5944 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
5945 then the caller guarantees that the source and destination pointers are
5946 aligned to that boundary.</p>
5949 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
5950 source location to the destination location, which may overlap. It copies
5951 "len" bytes of memory over. If the argument is known to be aligned to some
5952 boundary, this can be specified as the fourth argument, otherwise it should
5953 be set to 0 or 1.</p>
5957 <!-- _______________________________________________________________________ -->
5958 <div class="doc_subsubsection">
5959 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5962 <div class="doc_text">
5965 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5966 width. Not all targets support all bit widths however.</p>
5969 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5970 i8 <len>, i32 <align>)
5971 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5972 i16 <len>, i32 <align>)
5973 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5974 i32 <len>, i32 <align>)
5975 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5976 i64 <len>, i32 <align>)
5980 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
5981 particular byte value.</p>
5983 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
5984 intrinsic does not return a value, and takes an extra alignment argument.</p>
5987 <p>The first argument is a pointer to the destination to fill, the second is the
5988 byte value to fill it with, the third argument is an integer argument
5989 specifying the number of bytes to fill, and the fourth argument is the known
5990 alignment of destination location.</p>
5992 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
5993 then the caller guarantees that the destination pointer is aligned to that
5997 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
5998 at the destination location. If the argument is known to be aligned to some
5999 boundary, this can be specified as the fourth argument, otherwise it should
6000 be set to 0 or 1.</p>
6004 <!-- _______________________________________________________________________ -->
6005 <div class="doc_subsubsection">
6006 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
6009 <div class="doc_text">
6012 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
6013 floating point or vector of floating point type. Not all targets support all
6017 declare float @llvm.sqrt.f32(float %Val)
6018 declare double @llvm.sqrt.f64(double %Val)
6019 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6020 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6021 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6025 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
6026 returning the same value as the libm '<tt>sqrt</tt>' functions would.
6027 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
6028 behavior for negative numbers other than -0.0 (which allows for better
6029 optimization, because there is no need to worry about errno being
6030 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
6033 <p>The argument and return value are floating point numbers of the same
6037 <p>This function returns the sqrt of the specified operand if it is a
6038 nonnegative floating point number.</p>
6042 <!-- _______________________________________________________________________ -->
6043 <div class="doc_subsubsection">
6044 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
6047 <div class="doc_text">
6050 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
6051 floating point or vector of floating point type. Not all targets support all
6055 declare float @llvm.powi.f32(float %Val, i32 %power)
6056 declare double @llvm.powi.f64(double %Val, i32 %power)
6057 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6058 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6059 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6063 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
6064 specified (positive or negative) power. The order of evaluation of
6065 multiplications is not defined. When a vector of floating point type is
6066 used, the second argument remains a scalar integer value.</p>
6069 <p>The second argument is an integer power, and the first is a value to raise to
6073 <p>This function returns the first value raised to the second power with an
6074 unspecified sequence of rounding operations.</p>
6078 <!-- _______________________________________________________________________ -->
6079 <div class="doc_subsubsection">
6080 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
6083 <div class="doc_text">
6086 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
6087 floating point or vector of floating point type. Not all targets support all
6091 declare float @llvm.sin.f32(float %Val)
6092 declare double @llvm.sin.f64(double %Val)
6093 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6094 declare fp128 @llvm.sin.f128(fp128 %Val)
6095 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6099 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
6102 <p>The argument and return value are floating point numbers of the same
6106 <p>This function returns the sine of the specified operand, returning the same
6107 values as the libm <tt>sin</tt> functions would, and handles error conditions
6108 in the same way.</p>
6112 <!-- _______________________________________________________________________ -->
6113 <div class="doc_subsubsection">
6114 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
6117 <div class="doc_text">
6120 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
6121 floating point or vector of floating point type. Not all targets support all
6125 declare float @llvm.cos.f32(float %Val)
6126 declare double @llvm.cos.f64(double %Val)
6127 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6128 declare fp128 @llvm.cos.f128(fp128 %Val)
6129 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6133 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
6136 <p>The argument and return value are floating point numbers of the same
6140 <p>This function returns the cosine of the specified operand, returning the same
6141 values as the libm <tt>cos</tt> functions would, and handles error conditions
6142 in the same way.</p>
6146 <!-- _______________________________________________________________________ -->
6147 <div class="doc_subsubsection">
6148 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
6151 <div class="doc_text">
6154 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
6155 floating point or vector of floating point type. Not all targets support all
6159 declare float @llvm.pow.f32(float %Val, float %Power)
6160 declare double @llvm.pow.f64(double %Val, double %Power)
6161 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6162 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6163 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6167 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
6168 specified (positive or negative) power.</p>
6171 <p>The second argument is a floating point power, and the first is a value to
6172 raise to that power.</p>
6175 <p>This function returns the first value raised to the second power, returning
6176 the same values as the libm <tt>pow</tt> functions would, and handles error
6177 conditions in the same way.</p>
6181 <!-- ======================================================================= -->
6182 <div class="doc_subsection">
6183 <a name="int_manip">Bit Manipulation Intrinsics</a>
6186 <div class="doc_text">
6188 <p>LLVM provides intrinsics for a few important bit manipulation operations.
6189 These allow efficient code generation for some algorithms.</p>
6193 <!-- _______________________________________________________________________ -->
6194 <div class="doc_subsubsection">
6195 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
6198 <div class="doc_text">
6201 <p>This is an overloaded intrinsic function. You can use bswap on any integer
6202 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
6205 declare i16 @llvm.bswap.i16(i16 <id>)
6206 declare i32 @llvm.bswap.i32(i32 <id>)
6207 declare i64 @llvm.bswap.i64(i64 <id>)
6211 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
6212 values with an even number of bytes (positive multiple of 16 bits). These
6213 are useful for performing operations on data that is not in the target's
6214 native byte order.</p>
6217 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
6218 and low byte of the input i16 swapped. Similarly,
6219 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
6220 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
6221 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
6222 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
6223 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
6224 more, respectively).</p>
6228 <!-- _______________________________________________________________________ -->
6229 <div class="doc_subsubsection">
6230 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
6233 <div class="doc_text">
6236 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6237 width. Not all targets support all bit widths however.</p>
6240 declare i8 @llvm.ctpop.i8(i8 <src>)
6241 declare i16 @llvm.ctpop.i16(i16 <src>)
6242 declare i32 @llvm.ctpop.i32(i32 <src>)
6243 declare i64 @llvm.ctpop.i64(i64 <src>)
6244 declare i256 @llvm.ctpop.i256(i256 <src>)
6248 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
6252 <p>The only argument is the value to be counted. The argument may be of any
6253 integer type. The return type must match the argument type.</p>
6256 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.</p>
6260 <!-- _______________________________________________________________________ -->
6261 <div class="doc_subsubsection">
6262 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
6265 <div class="doc_text">
6268 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
6269 integer bit width. Not all targets support all bit widths however.</p>
6272 declare i8 @llvm.ctlz.i8 (i8 <src>)
6273 declare i16 @llvm.ctlz.i16(i16 <src>)
6274 declare i32 @llvm.ctlz.i32(i32 <src>)
6275 declare i64 @llvm.ctlz.i64(i64 <src>)
6276 declare i256 @llvm.ctlz.i256(i256 <src>)
6280 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
6281 leading zeros in a variable.</p>
6284 <p>The only argument is the value to be counted. The argument may be of any
6285 integer type. The return type must match the argument type.</p>
6288 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
6289 zeros in a variable. If the src == 0 then the result is the size in bits of
6290 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
6294 <!-- _______________________________________________________________________ -->
6295 <div class="doc_subsubsection">
6296 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
6299 <div class="doc_text">
6302 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
6303 integer bit width. Not all targets support all bit widths however.</p>
6306 declare i8 @llvm.cttz.i8 (i8 <src>)
6307 declare i16 @llvm.cttz.i16(i16 <src>)
6308 declare i32 @llvm.cttz.i32(i32 <src>)
6309 declare i64 @llvm.cttz.i64(i64 <src>)
6310 declare i256 @llvm.cttz.i256(i256 <src>)
6314 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
6318 <p>The only argument is the value to be counted. The argument may be of any
6319 integer type. The return type must match the argument type.</p>
6322 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
6323 zeros in a variable. If the src == 0 then the result is the size in bits of
6324 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
6328 <!-- ======================================================================= -->
6329 <div class="doc_subsection">
6330 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
6333 <div class="doc_text">
6335 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
6339 <!-- _______________________________________________________________________ -->
6340 <div class="doc_subsubsection">
6341 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
6344 <div class="doc_text">
6347 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
6348 on any integer bit width.</p>
6351 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6352 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6353 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6357 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6358 a signed addition of the two arguments, and indicate whether an overflow
6359 occurred during the signed 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 signed addition.</p>
6369 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6370 a signed addition of the two variables. They return a structure — the
6371 first element of which is the signed summation, and the second element of
6372 which is a bit specifying if the signed summation resulted in an
6377 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6378 %sum = extractvalue {i32, i1} %res, 0
6379 %obit = extractvalue {i32, i1} %res, 1
6380 br i1 %obit, label %overflow, label %normal
6385 <!-- _______________________________________________________________________ -->
6386 <div class="doc_subsubsection">
6387 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
6390 <div class="doc_text">
6393 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
6394 on any integer bit width.</p>
6397 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6398 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6399 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6403 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6404 an unsigned addition of the two arguments, and indicate whether a carry
6405 occurred during the unsigned summation.</p>
6408 <p>The arguments (%a and %b) and the first element of the result structure may
6409 be of integer types of any bit width, but they must have the same bit
6410 width. The second element of the result structure must be of
6411 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6412 undergo unsigned addition.</p>
6415 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6416 an unsigned addition of the two arguments. They return a structure —
6417 the first element of which is the sum, and the second element of which is a
6418 bit specifying if the unsigned summation resulted in a carry.</p>
6422 %res = call {i32, i1} @llvm.uadd.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 %carry, label %normal
6430 <!-- _______________________________________________________________________ -->
6431 <div class="doc_subsubsection">
6432 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6435 <div class="doc_text">
6438 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6439 on any integer bit width.</p>
6442 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6443 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6444 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6448 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6449 a signed subtraction of the two arguments, and indicate whether an overflow
6450 occurred during the signed 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 signed subtraction.</p>
6460 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6461 a signed 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 signed subtraction resulted in an
6468 %res = call {i32, i1} @llvm.ssub.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_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6481 <div class="doc_text">
6484 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6485 on any integer bit width.</p>
6488 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6489 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6490 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6494 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6495 an unsigned subtraction of the two arguments, and indicate whether an
6496 overflow occurred during the unsigned subtraction.</p>
6499 <p>The arguments (%a and %b) and the first element of the result structure may
6500 be of integer types of any bit width, but they must have the same bit
6501 width. The second element of the result structure must be of
6502 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6503 undergo unsigned subtraction.</p>
6506 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6507 an unsigned subtraction of the two arguments. They return a structure —
6508 the first element of which is the subtraction, and the second element of
6509 which is a bit specifying if the unsigned subtraction resulted in an
6514 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6515 %sum = extractvalue {i32, i1} %res, 0
6516 %obit = extractvalue {i32, i1} %res, 1
6517 br i1 %obit, label %overflow, label %normal
6522 <!-- _______________________________________________________________________ -->
6523 <div class="doc_subsubsection">
6524 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6527 <div class="doc_text">
6530 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6531 on any integer bit width.</p>
6534 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6535 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6536 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6541 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6542 a signed multiplication of the two arguments, and indicate whether an
6543 overflow occurred during the signed 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 signed multiplication.</p>
6553 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6554 a signed multiplication of the two arguments. They return a structure —
6555 the first element of which is the multiplication, and the second element of
6556 which is a bit specifying if the signed multiplication resulted in an
6561 %res = call {i32, i1} @llvm.smul.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_subsubsection">
6571 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6574 <div class="doc_text">
6577 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6578 on any integer bit width.</p>
6581 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6582 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6583 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6587 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6588 a unsigned multiplication of the two arguments, and indicate whether an
6589 overflow occurred during the unsigned multiplication.</p>
6592 <p>The arguments (%a and %b) and the first element of the result structure may
6593 be of integer types of any bit width, but they must have the same bit
6594 width. The second element of the result structure must be of
6595 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6596 undergo unsigned multiplication.</p>
6599 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6600 an unsigned multiplication of the two arguments. They return a structure
6601 — the first element of which is the multiplication, and the second
6602 element of which is a bit specifying if the unsigned multiplication resulted
6607 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6608 %sum = extractvalue {i32, i1} %res, 0
6609 %obit = extractvalue {i32, i1} %res, 1
6610 br i1 %obit, label %overflow, label %normal
6615 <!-- ======================================================================= -->
6616 <div class="doc_subsection">
6617 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
6620 <div class="doc_text">
6622 <p>Half precision floating point is a storage-only format. This means that it is
6623 a dense encoding (in memory) but does not support computation in the
6626 <p>This means that code must first load the half-precision floating point
6627 value as an i16, then convert it to float with <a
6628 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
6629 Computation can then be performed on the float value (including extending to
6630 double etc). To store the value back to memory, it is first converted to
6631 float if needed, then converted to i16 with
6632 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
6633 storing as an i16 value.</p>
6636 <!-- _______________________________________________________________________ -->
6637 <div class="doc_subsubsection">
6638 <a name="int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a>
6641 <div class="doc_text">
6645 declare i16 @llvm.convert.to.fp16(f32 %a)
6649 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
6650 a conversion from single precision floating point format to half precision
6651 floating point format.</p>
6654 <p>The intrinsic function contains single argument - the value to be
6658 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
6659 a conversion from single precision floating point format to half precision
6660 floating point format. The return value is an <tt>i16</tt> which
6661 contains the converted number.</p>
6665 %res = call i16 @llvm.convert.to.fp16(f32 %a)
6666 store i16 %res, i16* @x, align 2
6671 <!-- _______________________________________________________________________ -->
6672 <div class="doc_subsubsection">
6673 <a name="int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a>
6676 <div class="doc_text">
6680 declare f32 @llvm.convert.from.fp16(i16 %a)
6684 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
6685 a conversion from half precision floating point format to single precision
6686 floating point format.</p>
6689 <p>The intrinsic function contains single argument - the value to be
6693 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
6694 conversion from half single precision floating point format to single
6695 precision floating point format. The input half-float value is represented by
6696 an <tt>i16</tt> value.</p>
6700 %a = load i16* @x, align 2
6701 %res = call f32 @llvm.convert.from.fp16(i16 %a)
6706 <!-- ======================================================================= -->
6707 <div class="doc_subsection">
6708 <a name="int_debugger">Debugger Intrinsics</a>
6711 <div class="doc_text">
6713 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
6714 prefix), are described in
6715 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
6716 Level Debugging</a> document.</p>
6720 <!-- ======================================================================= -->
6721 <div class="doc_subsection">
6722 <a name="int_eh">Exception Handling Intrinsics</a>
6725 <div class="doc_text">
6727 <p>The LLVM exception handling intrinsics (which all start with
6728 <tt>llvm.eh.</tt> prefix), are described in
6729 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6730 Handling</a> document.</p>
6734 <!-- ======================================================================= -->
6735 <div class="doc_subsection">
6736 <a name="int_trampoline">Trampoline Intrinsic</a>
6739 <div class="doc_text">
6741 <p>This intrinsic makes it possible to excise one parameter, marked with
6742 the <tt>nest</tt> attribute, from a function. The result is a callable
6743 function pointer lacking the nest parameter - the caller does not need to
6744 provide a value for it. Instead, the value to use is stored in advance in a
6745 "trampoline", a block of memory usually allocated on the stack, which also
6746 contains code to splice the nest value into the argument list. This is used
6747 to implement the GCC nested function address extension.</p>
6749 <p>For example, if the function is
6750 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6751 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
6754 <div class="doc_code">
6756 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6757 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6758 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6759 %fp = bitcast i8* %p to i32 (i32, i32)*
6763 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6764 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6768 <!-- _______________________________________________________________________ -->
6769 <div class="doc_subsubsection">
6770 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6773 <div class="doc_text">
6777 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6781 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
6782 function pointer suitable for executing it.</p>
6785 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6786 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
6787 sufficiently aligned block of memory; this memory is written to by the
6788 intrinsic. Note that the size and the alignment are target-specific - LLVM
6789 currently provides no portable way of determining them, so a front-end that
6790 generates this intrinsic needs to have some target-specific knowledge.
6791 The <tt>func</tt> argument must hold a function bitcast to
6792 an <tt>i8*</tt>.</p>
6795 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
6796 dependent code, turning it into a function. A pointer to this function is
6797 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
6798 function pointer type</a> before being called. The new function's signature
6799 is the same as that of <tt>func</tt> with any arguments marked with
6800 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
6801 is allowed, and it must be of pointer type. Calling the new function is
6802 equivalent to calling <tt>func</tt> with the same argument list, but
6803 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
6804 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
6805 by <tt>tramp</tt> is modified, then the effect of any later call to the
6806 returned function pointer is undefined.</p>
6810 <!-- ======================================================================= -->
6811 <div class="doc_subsection">
6812 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6815 <div class="doc_text">
6817 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
6818 hardware constructs for atomic operations and memory synchronization. This
6819 provides an interface to the hardware, not an interface to the programmer. It
6820 is aimed at a low enough level to allow any programming models or APIs
6821 (Application Programming Interfaces) which need atomic behaviors to map
6822 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6823 hardware provides a "universal IR" for source languages, it also provides a
6824 starting point for developing a "universal" atomic operation and
6825 synchronization IR.</p>
6827 <p>These do <em>not</em> form an API such as high-level threading libraries,
6828 software transaction memory systems, atomic primitives, and intrinsic
6829 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6830 application libraries. The hardware interface provided by LLVM should allow
6831 a clean implementation of all of these APIs and parallel programming models.
6832 No one model or paradigm should be selected above others unless the hardware
6833 itself ubiquitously does so.</p>
6837 <!-- _______________________________________________________________________ -->
6838 <div class="doc_subsubsection">
6839 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6841 <div class="doc_text">
6844 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device> )
6848 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6849 specific pairs of memory access types.</p>
6852 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6853 The first four arguments enables a specific barrier as listed below. The
6854 fifth argument specifies that the barrier applies to io or device or uncached
6858 <li><tt>ll</tt>: load-load barrier</li>
6859 <li><tt>ls</tt>: load-store barrier</li>
6860 <li><tt>sl</tt>: store-load barrier</li>
6861 <li><tt>ss</tt>: store-store barrier</li>
6862 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6866 <p>This intrinsic causes the system to enforce some ordering constraints upon
6867 the loads and stores of the program. This barrier does not
6868 indicate <em>when</em> any events will occur, it only enforces
6869 an <em>order</em> in which they occur. For any of the specified pairs of load
6870 and store operations (f.ex. load-load, or store-load), all of the first
6871 operations preceding the barrier will complete before any of the second
6872 operations succeeding the barrier begin. Specifically the semantics for each
6873 pairing is as follows:</p>
6876 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6877 after the barrier begins.</li>
6878 <li><tt>ls</tt>: All loads before the barrier must complete before any
6879 store after the barrier begins.</li>
6880 <li><tt>ss</tt>: All stores before the barrier must complete before any
6881 store after the barrier begins.</li>
6882 <li><tt>sl</tt>: All stores before the barrier must complete before any
6883 load after the barrier begins.</li>
6886 <p>These semantics are applied with a logical "and" behavior when more than one
6887 is enabled in a single memory barrier intrinsic.</p>
6889 <p>Backends may implement stronger barriers than those requested when they do
6890 not support as fine grained a barrier as requested. Some architectures do
6891 not need all types of barriers and on such architectures, these become
6896 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6897 %ptr = bitcast i8* %mallocP to i32*
6900 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6901 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6902 <i>; guarantee the above finishes</i>
6903 store i32 8, %ptr <i>; before this begins</i>
6908 <!-- _______________________________________________________________________ -->
6909 <div class="doc_subsubsection">
6910 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6913 <div class="doc_text">
6916 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6917 any integer bit width and for different address spaces. Not all targets
6918 support all bit widths however.</p>
6921 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6922 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6923 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6924 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6928 <p>This loads a value in memory and compares it to a given value. If they are
6929 equal, it stores a new value into the memory.</p>
6932 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
6933 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6934 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6935 this integer type. While any bit width integer may be used, targets may only
6936 lower representations they support in hardware.</p>
6939 <p>This entire intrinsic must be executed atomically. It first loads the value
6940 in memory pointed to by <tt>ptr</tt> and compares it with the
6941 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
6942 memory. The loaded value is yielded in all cases. This provides the
6943 equivalent of an atomic compare-and-swap operation within the SSA
6948 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6949 %ptr = bitcast i8* %mallocP to i32*
6952 %val1 = add i32 4, 4
6953 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6954 <i>; yields {i32}:result1 = 4</i>
6955 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6956 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6958 %val2 = add i32 1, 1
6959 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6960 <i>; yields {i32}:result2 = 8</i>
6961 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6963 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6968 <!-- _______________________________________________________________________ -->
6969 <div class="doc_subsubsection">
6970 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6972 <div class="doc_text">
6975 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6976 integer bit width. Not all targets support all bit widths however.</p>
6979 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6980 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6981 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6982 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6986 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6987 the value from memory. It then stores the value in <tt>val</tt> in the memory
6988 at <tt>ptr</tt>.</p>
6991 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
6992 the <tt>val</tt> argument and the result must be integers of the same bit
6993 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6994 integer type. The targets may only lower integer representations they
6998 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6999 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
7000 equivalent of an atomic swap operation within the SSA framework.</p>
7004 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7005 %ptr = bitcast i8* %mallocP to i32*
7008 %val1 = add i32 4, 4
7009 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
7010 <i>; yields {i32}:result1 = 4</i>
7011 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7012 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7014 %val2 = add i32 1, 1
7015 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
7016 <i>; yields {i32}:result2 = 8</i>
7018 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
7019 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
7024 <!-- _______________________________________________________________________ -->
7025 <div class="doc_subsubsection">
7026 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
7030 <div class="doc_text">
7033 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
7034 any integer bit width. Not all targets support all bit widths however.</p>
7037 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
7038 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
7039 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
7040 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
7044 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
7045 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7048 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7049 and the second an integer value. The result is also an integer value. These
7050 integer types can have any bit width, but they must all have the same bit
7051 width. The targets may only lower integer representations they support.</p>
7054 <p>This intrinsic does a series of operations atomically. It first loads the
7055 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
7056 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
7060 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7061 %ptr = bitcast i8* %mallocP to i32*
7063 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
7064 <i>; yields {i32}:result1 = 4</i>
7065 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
7066 <i>; yields {i32}:result2 = 8</i>
7067 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
7068 <i>; yields {i32}:result3 = 10</i>
7069 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
7074 <!-- _______________________________________________________________________ -->
7075 <div class="doc_subsubsection">
7076 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
7080 <div class="doc_text">
7083 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
7084 any integer bit width and for different address spaces. Not all targets
7085 support all bit widths however.</p>
7088 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
7089 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
7090 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
7091 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
7095 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
7096 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7099 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7100 and the second an integer value. The result is also an integer value. These
7101 integer types can have any bit width, but they must all have the same bit
7102 width. The targets may only lower integer representations they support.</p>
7105 <p>This intrinsic does a series of operations atomically. It first loads the
7106 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
7107 result to <tt>ptr</tt>. It yields the original value stored
7108 at <tt>ptr</tt>.</p>
7112 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7113 %ptr = bitcast i8* %mallocP to i32*
7115 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
7116 <i>; yields {i32}:result1 = 8</i>
7117 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
7118 <i>; yields {i32}:result2 = 4</i>
7119 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
7120 <i>; yields {i32}:result3 = 2</i>
7121 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
7126 <!-- _______________________________________________________________________ -->
7127 <div class="doc_subsubsection">
7128 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
7129 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
7130 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
7131 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
7134 <div class="doc_text">
7137 <p>These are overloaded intrinsics. You can
7138 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
7139 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
7140 bit width and for different address spaces. Not all targets support all bit
7144 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
7145 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
7146 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
7147 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
7151 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
7152 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
7153 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
7154 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
7158 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
7159 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
7160 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
7161 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
7165 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
7166 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
7167 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
7168 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
7172 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
7173 the value stored in memory at <tt>ptr</tt>. It yields the original value
7174 at <tt>ptr</tt>.</p>
7177 <p>These intrinsics take two arguments, the first a pointer to an integer value
7178 and the second an integer value. The result is also an integer value. These
7179 integer types can have any bit width, but they must all have the same bit
7180 width. The targets may only lower integer representations they support.</p>
7183 <p>These intrinsics does a series of operations atomically. They first load the
7184 value stored at <tt>ptr</tt>. They then do the bitwise
7185 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
7186 original value stored at <tt>ptr</tt>.</p>
7190 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7191 %ptr = bitcast i8* %mallocP to i32*
7192 store i32 0x0F0F, %ptr
7193 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
7194 <i>; yields {i32}:result0 = 0x0F0F</i>
7195 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
7196 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
7197 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
7198 <i>; yields {i32}:result2 = 0xF0</i>
7199 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
7200 <i>; yields {i32}:result3 = FF</i>
7201 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
7206 <!-- _______________________________________________________________________ -->
7207 <div class="doc_subsubsection">
7208 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
7209 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
7210 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
7211 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
7214 <div class="doc_text">
7217 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
7218 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
7219 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
7220 address spaces. Not all targets support all bit widths however.</p>
7223 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
7224 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
7225 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
7226 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
7230 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
7231 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
7232 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
7233 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
7237 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
7238 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
7239 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
7240 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
7244 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
7245 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
7246 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
7247 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
7251 <p>These intrinsics takes the signed or unsigned minimum or maximum of
7252 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
7253 original value at <tt>ptr</tt>.</p>
7256 <p>These intrinsics take two arguments, the first a pointer to an integer value
7257 and the second an integer value. The result is also an integer value. These
7258 integer types can have any bit width, but they must all have the same bit
7259 width. The targets may only lower integer representations they support.</p>
7262 <p>These intrinsics does a series of operations atomically. They first load the
7263 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
7264 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
7265 yield the original value stored at <tt>ptr</tt>.</p>
7269 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7270 %ptr = bitcast i8* %mallocP to i32*
7272 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
7273 <i>; yields {i32}:result0 = 7</i>
7274 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
7275 <i>; yields {i32}:result1 = -2</i>
7276 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
7277 <i>; yields {i32}:result2 = 8</i>
7278 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
7279 <i>; yields {i32}:result3 = 8</i>
7280 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
7286 <!-- ======================================================================= -->
7287 <div class="doc_subsection">
7288 <a name="int_memorymarkers">Memory Use Markers</a>
7291 <div class="doc_text">
7293 <p>This class of intrinsics exists to information about the lifetime of memory
7294 objects and ranges where variables are immutable.</p>
7298 <!-- _______________________________________________________________________ -->
7299 <div class="doc_subsubsection">
7300 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
7303 <div class="doc_text">
7307 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
7311 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
7312 object's lifetime.</p>
7315 <p>The first argument is a constant integer representing the size of the
7316 object, or -1 if it is variable sized. The second argument is a pointer to
7320 <p>This intrinsic indicates that before this point in the code, the value of the
7321 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7322 never be used and has an undefined value. A load from the pointer that
7323 precedes this intrinsic can be replaced with
7324 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
7328 <!-- _______________________________________________________________________ -->
7329 <div class="doc_subsubsection">
7330 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
7333 <div class="doc_text">
7337 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
7341 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
7342 object's lifetime.</p>
7345 <p>The first argument is a constant integer representing the size of the
7346 object, or -1 if it is variable sized. The second argument is a pointer to
7350 <p>This intrinsic indicates that after this point in the code, the value of the
7351 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7352 never be used and has an undefined value. Any stores into the memory object
7353 following this intrinsic may be removed as dead.
7357 <!-- _______________________________________________________________________ -->
7358 <div class="doc_subsubsection">
7359 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
7362 <div class="doc_text">
7366 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>) readonly
7370 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
7371 a memory object will not change.</p>
7374 <p>The first argument is a constant integer representing the size of the
7375 object, or -1 if it is variable sized. The second argument is a pointer to
7379 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
7380 the return value, the referenced memory location is constant and
7385 <!-- _______________________________________________________________________ -->
7386 <div class="doc_subsubsection">
7387 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
7390 <div class="doc_text">
7394 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7398 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
7399 a memory object are mutable.</p>
7402 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
7403 The second argument is a constant integer representing the size of the
7404 object, or -1 if it is variable sized and the third argument is a pointer
7408 <p>This intrinsic indicates that the memory is mutable again.</p>
7412 <!-- ======================================================================= -->
7413 <div class="doc_subsection">
7414 <a name="int_general">General Intrinsics</a>
7417 <div class="doc_text">
7419 <p>This class of intrinsics is designed to be generic and has no specific
7424 <!-- _______________________________________________________________________ -->
7425 <div class="doc_subsubsection">
7426 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
7429 <div class="doc_text">
7433 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
7437 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
7440 <p>The first argument is a pointer to a value, the second is a pointer to a
7441 global string, the third is a pointer to a global string which is the source
7442 file name, and the last argument is the line number.</p>
7445 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7446 This can be useful for special purpose optimizations that want to look for
7447 these annotations. These have no other defined use, they are ignored by code
7448 generation and optimization.</p>
7452 <!-- _______________________________________________________________________ -->
7453 <div class="doc_subsubsection">
7454 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
7457 <div class="doc_text">
7460 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
7461 any integer bit width.</p>
7464 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
7465 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
7466 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
7467 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
7468 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
7472 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
7475 <p>The first argument is an integer value (result of some expression), the
7476 second is a pointer to a global string, the third is a pointer to a global
7477 string which is the source file name, and the last argument is the line
7478 number. It returns the value of the first argument.</p>
7481 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7482 arbitrary strings. This can be useful for special purpose optimizations that
7483 want to look for these annotations. These have no other defined use, they
7484 are ignored by code generation and optimization.</p>
7488 <!-- _______________________________________________________________________ -->
7489 <div class="doc_subsubsection">
7490 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
7493 <div class="doc_text">
7497 declare void @llvm.trap()
7501 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
7507 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7508 target does not have a trap instruction, this intrinsic will be lowered to
7509 the call of the <tt>abort()</tt> function.</p>
7513 <!-- _______________________________________________________________________ -->
7514 <div class="doc_subsubsection">
7515 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
7518 <div class="doc_text">
7522 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
7526 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
7527 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
7528 ensure that it is placed on the stack before local variables.</p>
7531 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
7532 arguments. The first argument is the value loaded from the stack
7533 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
7534 that has enough space to hold the value of the guard.</p>
7537 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7538 the <tt>AllocaInst</tt> stack slot to be before local variables on the
7539 stack. This is to ensure that if a local variable on the stack is
7540 overwritten, it will destroy the value of the guard. When the function exits,
7541 the guard on the stack is checked against the original guard. If they're
7542 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
7547 <!-- _______________________________________________________________________ -->
7548 <div class="doc_subsubsection">
7549 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
7552 <div class="doc_text">
7556 declare i32 @llvm.objectsize.i32( i8* <object>, i1 <type> )
7557 declare i64 @llvm.objectsize.i64( i8* <object>, i1 <type> )
7561 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information
7562 to the optimizers to discover at compile time either a) when an
7563 operation like memcpy will either overflow a buffer that corresponds to
7564 an object, or b) to determine that a runtime check for overflow isn't
7565 necessary. An object in this context means an allocation of a
7566 specific class, structure, array, or other object.</p>
7569 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
7570 argument is a pointer to or into the <tt>object</tt>. The second argument
7571 is a boolean 0 or 1. This argument determines whether you want the
7572 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
7573 1, variables are not allowed.</p>
7576 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
7577 representing the size of the object concerned or <tt>i32/i64 -1 or 0</tt>
7578 (depending on the <tt>type</tt> argument if the size cannot be determined
7579 at compile time.</p>
7583 <!-- *********************************************************************** -->
7586 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
7587 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
7588 <a href="http://validator.w3.org/check/referer"><img
7589 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
7591 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
7592 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
7593 Last modified: $Date$