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="#trapvalues">Trap Values</a></li>
93 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
94 <li><a href="#constantexprs">Constant Expressions</a></li>
97 <li><a href="#othervalues">Other Values</a>
99 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
100 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a></li>
103 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
105 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
106 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
107 Global Variable</a></li>
108 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
109 Global Variable</a></li>
110 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
111 Global Variable</a></li>
114 <li><a href="#instref">Instruction Reference</a>
116 <li><a href="#terminators">Terminator Instructions</a>
118 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
119 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
120 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
121 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
122 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
123 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
124 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
127 <li><a href="#binaryops">Binary Operations</a>
129 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
130 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
131 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
132 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
133 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
134 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
135 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
136 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
137 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
138 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
139 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
140 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
143 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
145 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
146 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
147 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
148 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
149 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
150 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
153 <li><a href="#vectorops">Vector Operations</a>
155 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
156 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
157 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
160 <li><a href="#aggregateops">Aggregate Operations</a>
162 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
163 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
166 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
168 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
169 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
170 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
171 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
174 <li><a href="#convertops">Conversion Operations</a>
176 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
177 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
178 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
179 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
180 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
181 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
182 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
183 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
184 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
185 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
186 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
187 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
190 <li><a href="#otherops">Other Operations</a>
192 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
193 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
194 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
195 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
196 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
197 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
202 <li><a href="#intrinsics">Intrinsic Functions</a>
204 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
206 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
207 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
208 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
211 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
213 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
214 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
215 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
218 <li><a href="#int_codegen">Code Generator Intrinsics</a>
220 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
221 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
222 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
223 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
224 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
225 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
226 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
229 <li><a href="#int_libc">Standard C Library Intrinsics</a>
231 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
232 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
233 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
234 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
235 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
236 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
237 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
238 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
241 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
243 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
244 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
245 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
246 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
249 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
251 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
252 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
253 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
254 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
255 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
256 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
259 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
261 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
262 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
265 <li><a href="#int_debugger">Debugger intrinsics</a></li>
266 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
267 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
269 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
272 <li><a href="#int_atomics">Atomic intrinsics</a>
274 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
275 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
276 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
277 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
278 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
279 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
280 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
281 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
282 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
283 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
284 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
285 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
286 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
289 <li><a href="#int_memorymarkers">Memory Use Markers</a>
291 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
292 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
293 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
294 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
297 <li><a href="#int_general">General intrinsics</a>
299 <li><a href="#int_var_annotation">
300 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
301 <li><a href="#int_annotation">
302 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
303 <li><a href="#int_trap">
304 '<tt>llvm.trap</tt>' Intrinsic</a></li>
305 <li><a href="#int_stackprotector">
306 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
307 <li><a href="#int_objectsize">
308 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
315 <div class="doc_author">
316 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
317 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
320 <!-- *********************************************************************** -->
321 <div class="doc_section"> <a name="abstract">Abstract </a></div>
322 <!-- *********************************************************************** -->
324 <div class="doc_text">
326 <p>This document is a reference manual for the LLVM assembly language. LLVM is
327 a Static Single Assignment (SSA) based representation that provides type
328 safety, low-level operations, flexibility, and the capability of representing
329 'all' high-level languages cleanly. It is the common code representation
330 used throughout all phases of the LLVM compilation strategy.</p>
334 <!-- *********************************************************************** -->
335 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
336 <!-- *********************************************************************** -->
338 <div class="doc_text">
340 <p>The LLVM code representation is designed to be used in three different forms:
341 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
342 for fast loading by a Just-In-Time compiler), and as a human readable
343 assembly language representation. This allows LLVM to provide a powerful
344 intermediate representation for efficient compiler transformations and
345 analysis, while providing a natural means to debug and visualize the
346 transformations. The three different forms of LLVM are all equivalent. This
347 document describes the human readable representation and notation.</p>
349 <p>The LLVM representation aims to be light-weight and low-level while being
350 expressive, typed, and extensible at the same time. It aims to be a
351 "universal IR" of sorts, by being at a low enough level that high-level ideas
352 may be cleanly mapped to it (similar to how microprocessors are "universal
353 IR's", allowing many source languages to be mapped to them). By providing
354 type information, LLVM can be used as the target of optimizations: for
355 example, through pointer analysis, it can be proven that a C automatic
356 variable is never accessed outside of the current function, allowing it to
357 be promoted to a simple SSA value instead of a memory location.</p>
361 <!-- _______________________________________________________________________ -->
362 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
364 <div class="doc_text">
366 <p>It is important to note that this document describes 'well formed' LLVM
367 assembly language. There is a difference between what the parser accepts and
368 what is considered 'well formed'. For example, the following instruction is
369 syntactically okay, but not well formed:</p>
371 <div class="doc_code">
373 %x = <a href="#i_add">add</a> i32 1, %x
377 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
378 LLVM infrastructure provides a verification pass that may be used to verify
379 that an LLVM module is well formed. This pass is automatically run by the
380 parser after parsing input assembly and by the optimizer before it outputs
381 bitcode. The violations pointed out by the verifier pass indicate bugs in
382 transformation passes or input to the parser.</p>
386 <!-- Describe the typesetting conventions here. -->
388 <!-- *********************************************************************** -->
389 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
390 <!-- *********************************************************************** -->
392 <div class="doc_text">
394 <p>LLVM identifiers come in two basic types: global and local. Global
395 identifiers (functions, global variables) begin with the <tt>'@'</tt>
396 character. Local identifiers (register names, types) begin with
397 the <tt>'%'</tt> character. Additionally, there are three different formats
398 for identifiers, for different purposes:</p>
401 <li>Named values are represented as a string of characters with their prefix.
402 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
403 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
404 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
405 other characters in their names can be surrounded with quotes. Special
406 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
407 ASCII code for the character in hexadecimal. In this way, any character
408 can be used in a name value, even quotes themselves.</li>
410 <li>Unnamed values are represented as an unsigned numeric value with their
411 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
413 <li>Constants, which are described in a <a href="#constants">section about
414 constants</a>, below.</li>
417 <p>LLVM requires that values start with a prefix for two reasons: Compilers
418 don't need to worry about name clashes with reserved words, and the set of
419 reserved words may be expanded in the future without penalty. Additionally,
420 unnamed identifiers allow a compiler to quickly come up with a temporary
421 variable without having to avoid symbol table conflicts.</p>
423 <p>Reserved words in LLVM are very similar to reserved words in other
424 languages. There are keywords for different opcodes
425 ('<tt><a href="#i_add">add</a></tt>',
426 '<tt><a href="#i_bitcast">bitcast</a></tt>',
427 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
428 ('<tt><a href="#t_void">void</a></tt>',
429 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
430 reserved words cannot conflict with variable names, because none of them
431 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
433 <p>Here is an example of LLVM code to multiply the integer variable
434 '<tt>%X</tt>' by 8:</p>
438 <div class="doc_code">
440 %result = <a href="#i_mul">mul</a> i32 %X, 8
444 <p>After strength reduction:</p>
446 <div class="doc_code">
448 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
452 <p>And the hard way:</p>
454 <div class="doc_code">
456 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
457 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
458 %result = <a href="#i_add">add</a> i32 %1, %1
462 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
463 lexical features of LLVM:</p>
466 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
469 <li>Unnamed temporaries are created when the result of a computation is not
470 assigned to a named value.</li>
472 <li>Unnamed temporaries are numbered sequentially</li>
475 <p>It also shows a convention that we follow in this document. When
476 demonstrating instructions, we will follow an instruction with a comment that
477 defines the type and name of value produced. Comments are shown in italic
482 <!-- *********************************************************************** -->
483 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
484 <!-- *********************************************************************** -->
486 <!-- ======================================================================= -->
487 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
490 <div class="doc_text">
492 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
493 of the input programs. Each module consists of functions, global variables,
494 and symbol table entries. Modules may be combined together with the LLVM
495 linker, which merges function (and global variable) definitions, resolves
496 forward declarations, and merges symbol table entries. Here is an example of
497 the "hello world" module:</p>
499 <div class="doc_code">
501 <i>; Declare the string constant as a global constant.</i>
502 <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>
504 <i>; External declaration of the puts function</i>
505 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
507 <i>; Definition of main function</i>
508 define i32 @main() { <i>; i32()* </i>
509 <i>; Convert [13 x i8]* to i8 *...</i>
510 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
512 <i>; Call puts function to write out the string to stdout.</i>
513 <a href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
514 <a href="#i_ret">ret</a> i32 0<br>}
516 <i>; Named metadata</i>
517 !1 = metadata !{i32 41}
522 <p>This example is made up of a <a href="#globalvars">global variable</a> named
523 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
524 a <a href="#functionstructure">function definition</a> for
525 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
528 <p>In general, a module is made up of a list of global values, where both
529 functions and global variables are global values. Global values are
530 represented by a pointer to a memory location (in this case, a pointer to an
531 array of char, and a pointer to a function), and have one of the
532 following <a href="#linkage">linkage types</a>.</p>
536 <!-- ======================================================================= -->
537 <div class="doc_subsection">
538 <a name="linkage">Linkage Types</a>
541 <div class="doc_text">
543 <p>All Global Variables and Functions have one of the following types of
547 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
548 <dd>Global values with private linkage are only directly accessible by objects
549 in the current module. In particular, linking code into a module with an
550 private global value may cause the private to be renamed as necessary to
551 avoid collisions. Because the symbol is private to the module, all
552 references can be updated. This doesn't show up in any symbol table in the
555 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
556 <dd>Similar to private, but the symbol is passed through the assembler and
557 removed by the linker after evaluation. Note that (unlike private
558 symbols) linker_private symbols are subject to coalescing by the linker:
559 weak symbols get merged and redefinitions are rejected. However, unlike
560 normal strong symbols, they are removed by the linker from the final
561 linked image (executable or dynamic library).</dd>
563 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
564 <dd>Similar to private, but the value shows as a local symbol
565 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
566 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
568 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
569 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
570 into the object file corresponding to the LLVM module. They exist to
571 allow inlining and other optimizations to take place given knowledge of
572 the definition of the global, which is known to be somewhere outside the
573 module. Globals with <tt>available_externally</tt> linkage are allowed to
574 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
575 This linkage type is only allowed on definitions, not declarations.</dd>
577 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
578 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
579 the same name when linkage occurs. This can be used to implement
580 some forms of inline functions, templates, or other code which must be
581 generated in each translation unit that uses it, but where the body may
582 be overridden with a more definitive definition later. Unreferenced
583 <tt>linkonce</tt> globals are allowed to be discarded. Note that
584 <tt>linkonce</tt> linkage does not actually allow the optimizer to
585 inline the body of this function into callers because it doesn't know if
586 this definition of the function is the definitive definition within the
587 program or whether it will be overridden by a stronger definition.
588 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
591 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
592 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
593 <tt>linkonce</tt> linkage, except that unreferenced globals with
594 <tt>weak</tt> linkage may not be discarded. This is used for globals that
595 are declared "weak" in C source code.</dd>
597 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
598 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
599 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
601 Symbols with "<tt>common</tt>" linkage are merged in the same way as
602 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
603 <tt>common</tt> symbols may not have an explicit section,
604 must have a zero initializer, and may not be marked '<a
605 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
606 have common linkage.</dd>
609 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
610 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
611 pointer to array type. When two global variables with appending linkage
612 are linked together, the two global arrays are appended together. This is
613 the LLVM, typesafe, equivalent of having the system linker append together
614 "sections" with identical names when .o files are linked.</dd>
616 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
617 <dd>The semantics of this linkage follow the ELF object file model: the symbol
618 is weak until linked, if not linked, the symbol becomes null instead of
619 being an undefined reference.</dd>
621 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
622 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
623 <dd>Some languages allow differing globals to be merged, such as two functions
624 with different semantics. Other languages, such as <tt>C++</tt>, ensure
625 that only equivalent globals are ever merged (the "one definition rule" -
626 "ODR"). Such languages can use the <tt>linkonce_odr</tt>
627 and <tt>weak_odr</tt> linkage types to indicate that the global will only
628 be merged with equivalent globals. These linkage types are otherwise the
629 same as their non-<tt>odr</tt> versions.</dd>
631 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
632 <dd>If none of the above identifiers are used, the global is externally
633 visible, meaning that it participates in linkage and can be used to
634 resolve external symbol references.</dd>
637 <p>The next two types of linkage are targeted for Microsoft Windows platform
638 only. They are designed to support importing (exporting) symbols from (to)
639 DLLs (Dynamic Link Libraries).</p>
642 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
643 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
644 or variable via a global pointer to a pointer that is set up by the DLL
645 exporting the symbol. On Microsoft Windows targets, the pointer name is
646 formed by combining <code>__imp_</code> and the function or variable
649 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
650 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
651 pointer to a pointer in a DLL, so that it can be referenced with the
652 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
653 name is formed by combining <code>__imp_</code> and the function or
657 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
658 another module defined a "<tt>.LC0</tt>" variable and was linked with this
659 one, one of the two would be renamed, preventing a collision. Since
660 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
661 declarations), they are accessible outside of the current module.</p>
663 <p>It is illegal for a function <i>declaration</i> to have any linkage type
664 other than "externally visible", <tt>dllimport</tt>
665 or <tt>extern_weak</tt>.</p>
667 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
668 or <tt>weak_odr</tt> linkages.</p>
672 <!-- ======================================================================= -->
673 <div class="doc_subsection">
674 <a name="callingconv">Calling Conventions</a>
677 <div class="doc_text">
679 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
680 and <a href="#i_invoke">invokes</a> can all have an optional calling
681 convention specified for the call. The calling convention of any pair of
682 dynamic caller/callee must match, or the behavior of the program is
683 undefined. The following calling conventions are supported by LLVM, and more
684 may be added in the future:</p>
687 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
688 <dd>This calling convention (the default if no other calling convention is
689 specified) matches the target C calling conventions. This calling
690 convention supports varargs function calls and tolerates some mismatch in
691 the declared prototype and implemented declaration of the function (as
694 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
695 <dd>This calling convention attempts to make calls as fast as possible
696 (e.g. by passing things in registers). This calling convention allows the
697 target to use whatever tricks it wants to produce fast code for the
698 target, without having to conform to an externally specified ABI
699 (Application Binary Interface).
700 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
701 when this or the GHC convention is used.</a> This calling convention
702 does not support varargs and requires the prototype of all callees to
703 exactly match the prototype of the function definition.</dd>
705 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
706 <dd>This calling convention attempts to make code in the caller as efficient
707 as possible under the assumption that the call is not commonly executed.
708 As such, these calls often preserve all registers so that the call does
709 not break any live ranges in the caller side. This calling convention
710 does not support varargs and requires the prototype of all callees to
711 exactly match the prototype of the function definition.</dd>
713 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
714 <dd>This calling convention has been implemented specifically for use by the
715 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
716 It passes everything in registers, going to extremes to achieve this by
717 disabling callee save registers. This calling convention should not be
718 used lightly but only for specific situations such as an alternative to
719 the <em>register pinning</em> performance technique often used when
720 implementing functional programming languages.At the moment only X86
721 supports this convention and it has the following limitations:
723 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
724 floating point types are supported.</li>
725 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
726 6 floating point parameters.</li>
728 This calling convention supports
729 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
730 requires both the caller and callee are using it.
733 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
734 <dd>Any calling convention may be specified by number, allowing
735 target-specific calling conventions to be used. Target specific calling
736 conventions start at 64.</dd>
739 <p>More calling conventions can be added/defined on an as-needed basis, to
740 support Pascal conventions or any other well-known target-independent
745 <!-- ======================================================================= -->
746 <div class="doc_subsection">
747 <a name="visibility">Visibility Styles</a>
750 <div class="doc_text">
752 <p>All Global Variables and Functions have one of the following visibility
756 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
757 <dd>On targets that use the ELF object file format, default visibility means
758 that the declaration is visible to other modules and, in shared libraries,
759 means that the declared entity may be overridden. On Darwin, default
760 visibility means that the declaration is visible to other modules. Default
761 visibility corresponds to "external linkage" in the language.</dd>
763 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
764 <dd>Two declarations of an object with hidden visibility refer to the same
765 object if they are in the same shared object. Usually, hidden visibility
766 indicates that the symbol will not be placed into the dynamic symbol
767 table, so no other module (executable or shared library) can reference it
770 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
771 <dd>On ELF, protected visibility indicates that the symbol will be placed in
772 the dynamic symbol table, but that references within the defining module
773 will bind to the local symbol. That is, the symbol cannot be overridden by
779 <!-- ======================================================================= -->
780 <div class="doc_subsection">
781 <a name="namedtypes">Named Types</a>
784 <div class="doc_text">
786 <p>LLVM IR allows you to specify name aliases for certain types. This can make
787 it easier to read the IR and make the IR more condensed (particularly when
788 recursive types are involved). An example of a name specification is:</p>
790 <div class="doc_code">
792 %mytype = type { %mytype*, i32 }
796 <p>You may give a name to any <a href="#typesystem">type</a> except
797 "<a href="t_void">void</a>". Type name aliases may be used anywhere a type
798 is expected with the syntax "%mytype".</p>
800 <p>Note that type names are aliases for the structural type that they indicate,
801 and that you can therefore specify multiple names for the same type. This
802 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
803 uses structural typing, the name is not part of the type. When printing out
804 LLVM IR, the printer will pick <em>one name</em> to render all types of a
805 particular shape. This means that if you have code where two different
806 source types end up having the same LLVM type, that the dumper will sometimes
807 print the "wrong" or unexpected type. This is an important design point and
808 isn't going to change.</p>
812 <!-- ======================================================================= -->
813 <div class="doc_subsection">
814 <a name="globalvars">Global Variables</a>
817 <div class="doc_text">
819 <p>Global variables define regions of memory allocated at compilation time
820 instead of run-time. Global variables may optionally be initialized, may
821 have an explicit section to be placed in, and may have an optional explicit
822 alignment specified. A variable may be defined as "thread_local", which
823 means that it will not be shared by threads (each thread will have a
824 separated copy of the variable). A variable may be defined as a global
825 "constant," which indicates that the contents of the variable
826 will <b>never</b> be modified (enabling better optimization, allowing the
827 global data to be placed in the read-only section of an executable, etc).
828 Note that variables that need runtime initialization cannot be marked
829 "constant" as there is a store to the variable.</p>
831 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
832 constant, even if the final definition of the global is not. This capability
833 can be used to enable slightly better optimization of the program, but
834 requires the language definition to guarantee that optimizations based on the
835 'constantness' are valid for the translation units that do not include the
838 <p>As SSA values, global variables define pointer values that are in scope
839 (i.e. they dominate) all basic blocks in the program. Global variables
840 always define a pointer to their "content" type because they describe a
841 region of memory, and all memory objects in LLVM are accessed through
844 <p>A global variable may be declared to reside in a target-specific numbered
845 address space. For targets that support them, address spaces may affect how
846 optimizations are performed and/or what target instructions are used to
847 access the variable. The default address space is zero. The address space
848 qualifier must precede any other attributes.</p>
850 <p>LLVM allows an explicit section to be specified for globals. If the target
851 supports it, it will emit globals to the section specified.</p>
853 <p>An explicit alignment may be specified for a global. If not present, or if
854 the alignment is set to zero, the alignment of the global is set by the
855 target to whatever it feels convenient. If an explicit alignment is
856 specified, the global is forced to have at least that much alignment. All
857 alignments must be a power of 2.</p>
859 <p>For example, the following defines a global in a numbered address space with
860 an initializer, section, and alignment:</p>
862 <div class="doc_code">
864 @G = addrspace(5) constant float 1.0, section "foo", align 4
871 <!-- ======================================================================= -->
872 <div class="doc_subsection">
873 <a name="functionstructure">Functions</a>
876 <div class="doc_text">
878 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
879 optional <a href="#linkage">linkage type</a>, an optional
880 <a href="#visibility">visibility style</a>, an optional
881 <a href="#callingconv">calling convention</a>, a return type, an optional
882 <a href="#paramattrs">parameter attribute</a> for the return type, a function
883 name, a (possibly empty) argument list (each with optional
884 <a href="#paramattrs">parameter attributes</a>), optional
885 <a href="#fnattrs">function attributes</a>, an optional section, an optional
886 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
887 curly brace, a list of basic blocks, and a closing curly brace.</p>
889 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
890 optional <a href="#linkage">linkage type</a>, an optional
891 <a href="#visibility">visibility style</a>, an optional
892 <a href="#callingconv">calling convention</a>, a return type, an optional
893 <a href="#paramattrs">parameter attribute</a> for the return type, a function
894 name, a possibly empty list of arguments, an optional alignment, and an
895 optional <a href="#gc">garbage collector name</a>.</p>
897 <p>A function definition contains a list of basic blocks, forming the CFG
898 (Control Flow Graph) for the function. Each basic block may optionally start
899 with a label (giving the basic block a symbol table entry), contains a list
900 of instructions, and ends with a <a href="#terminators">terminator</a>
901 instruction (such as a branch or function return).</p>
903 <p>The first basic block in a function is special in two ways: it is immediately
904 executed on entrance to the function, and it is not allowed to have
905 predecessor basic blocks (i.e. there can not be any branches to the entry
906 block of a function). Because the block can have no predecessors, it also
907 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
909 <p>LLVM allows an explicit section to be specified for functions. If the target
910 supports it, it will emit functions to the section specified.</p>
912 <p>An explicit alignment may be specified for a function. If not present, or if
913 the alignment is set to zero, the alignment of the function is set by the
914 target to whatever it feels convenient. If an explicit alignment is
915 specified, the function is forced to have at least that much alignment. All
916 alignments must be a power of 2.</p>
919 <div class="doc_code">
921 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
922 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
923 <ResultType> @<FunctionName> ([argument list])
924 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
925 [<a href="#gc">gc</a>] { ... }
931 <!-- ======================================================================= -->
932 <div class="doc_subsection">
933 <a name="aliasstructure">Aliases</a>
936 <div class="doc_text">
938 <p>Aliases act as "second name" for the aliasee value (which can be either
939 function, global variable, another alias or bitcast of global value). Aliases
940 may have an optional <a href="#linkage">linkage type</a>, and an
941 optional <a href="#visibility">visibility style</a>.</p>
944 <div class="doc_code">
946 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
952 <!-- ======================================================================= -->
953 <div class="doc_subsection">
954 <a name="namedmetadatastructure">Named Metadata</a>
957 <div class="doc_text">
959 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
960 nodes</a> (but not metadata strings) and null are the only valid operands for
961 a named metadata.</p>
964 <div class="doc_code">
966 !1 = metadata !{metadata !"one"}
973 <!-- ======================================================================= -->
974 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
976 <div class="doc_text">
978 <p>The return type and each parameter of a function type may have a set of
979 <i>parameter attributes</i> associated with them. Parameter attributes are
980 used to communicate additional information about the result or parameters of
981 a function. Parameter attributes are considered to be part of the function,
982 not of the function type, so functions with different parameter attributes
983 can have the same function type.</p>
985 <p>Parameter attributes are simple keywords that follow the type specified. If
986 multiple parameter attributes are needed, they are space separated. For
989 <div class="doc_code">
991 declare i32 @printf(i8* noalias nocapture, ...)
992 declare i32 @atoi(i8 zeroext)
993 declare signext i8 @returns_signed_char()
997 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
998 <tt>readonly</tt>) come immediately after the argument list.</p>
1000 <p>Currently, only the following parameter attributes are defined:</p>
1003 <dt><tt><b>zeroext</b></tt></dt>
1004 <dd>This indicates to the code generator that the parameter or return value
1005 should be zero-extended to a 32-bit value by the caller (for a parameter)
1006 or the callee (for a return value).</dd>
1008 <dt><tt><b>signext</b></tt></dt>
1009 <dd>This indicates to the code generator that the parameter or return value
1010 should be sign-extended to a 32-bit value by the caller (for a parameter)
1011 or the callee (for a return value).</dd>
1013 <dt><tt><b>inreg</b></tt></dt>
1014 <dd>This indicates that this parameter or return value should be treated in a
1015 special target-dependent fashion during while emitting code for a function
1016 call or return (usually, by putting it in a register as opposed to memory,
1017 though some targets use it to distinguish between two different kinds of
1018 registers). Use of this attribute is target-specific.</dd>
1020 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1021 <dd>This indicates that the pointer parameter should really be passed by value
1022 to the function. The attribute implies that a hidden copy of the pointee
1023 is made between the caller and the callee, so the callee is unable to
1024 modify the value in the callee. This attribute is only valid on LLVM
1025 pointer arguments. It is generally used to pass structs and arrays by
1026 value, but is also valid on pointers to scalars. The copy is considered
1027 to belong to the caller not the callee (for example,
1028 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1029 <tt>byval</tt> parameters). This is not a valid attribute for return
1030 values. The byval attribute also supports specifying an alignment with
1031 the align attribute. This has a target-specific effect on the code
1032 generator that usually indicates a desired alignment for the synthesized
1035 <dt><tt><b>sret</b></tt></dt>
1036 <dd>This indicates that the pointer parameter specifies the address of a
1037 structure that is the return value of the function in the source program.
1038 This pointer must be guaranteed by the caller to be valid: loads and
1039 stores to the structure may be assumed by the callee to not to trap. This
1040 may only be applied to the first parameter. This is not a valid attribute
1041 for return values. </dd>
1043 <dt><tt><b>noalias</b></tt></dt>
1044 <dd>This indicates that the pointer does not alias any global or any other
1045 parameter. The caller is responsible for ensuring that this is the
1046 case. On a function return value, <tt>noalias</tt> additionally indicates
1047 that the pointer does not alias any other pointers visible to the
1048 caller. For further details, please see the discussion of the NoAlias
1050 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
1053 <dt><tt><b>nocapture</b></tt></dt>
1054 <dd>This indicates that the callee does not make any copies of the pointer
1055 that outlive the callee itself. This is not a valid attribute for return
1058 <dt><tt><b>nest</b></tt></dt>
1059 <dd>This indicates that the pointer parameter can be excised using the
1060 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1061 attribute for return values.</dd>
1066 <!-- ======================================================================= -->
1067 <div class="doc_subsection">
1068 <a name="gc">Garbage Collector Names</a>
1071 <div class="doc_text">
1073 <p>Each function may specify a garbage collector name, which is simply a
1076 <div class="doc_code">
1078 define void @f() gc "name" { ... }
1082 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1083 collector which will cause the compiler to alter its output in order to
1084 support the named garbage collection algorithm.</p>
1088 <!-- ======================================================================= -->
1089 <div class="doc_subsection">
1090 <a name="fnattrs">Function Attributes</a>
1093 <div class="doc_text">
1095 <p>Function attributes are set to communicate additional information about a
1096 function. Function attributes are considered to be part of the function, not
1097 of the function type, so functions with different parameter attributes can
1098 have the same function type.</p>
1100 <p>Function attributes are simple keywords that follow the type specified. If
1101 multiple attributes are needed, they are space separated. For example:</p>
1103 <div class="doc_code">
1105 define void @f() noinline { ... }
1106 define void @f() alwaysinline { ... }
1107 define void @f() alwaysinline optsize { ... }
1108 define void @f() optsize { ... }
1113 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1114 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1115 the backend should forcibly align the stack pointer. Specify the
1116 desired alignment, which must be a power of two, in parentheses.
1118 <dt><tt><b>alwaysinline</b></tt></dt>
1119 <dd>This attribute indicates that the inliner should attempt to inline this
1120 function into callers whenever possible, ignoring any active inlining size
1121 threshold for this caller.</dd>
1123 <dt><tt><b>inlinehint</b></tt></dt>
1124 <dd>This attribute indicates that the source code contained a hint that inlining
1125 this function is desirable (such as the "inline" keyword in C/C++). It
1126 is just a hint; it imposes no requirements on the inliner.</dd>
1128 <dt><tt><b>noinline</b></tt></dt>
1129 <dd>This attribute indicates that the inliner should never inline this
1130 function in any situation. This attribute may not be used together with
1131 the <tt>alwaysinline</tt> attribute.</dd>
1133 <dt><tt><b>optsize</b></tt></dt>
1134 <dd>This attribute suggests that optimization passes and code generator passes
1135 make choices that keep the code size of this function low, and otherwise
1136 do optimizations specifically to reduce code size.</dd>
1138 <dt><tt><b>noreturn</b></tt></dt>
1139 <dd>This function attribute indicates that the function never returns
1140 normally. This produces undefined behavior at runtime if the function
1141 ever does dynamically return.</dd>
1143 <dt><tt><b>nounwind</b></tt></dt>
1144 <dd>This function attribute indicates that the function never returns with an
1145 unwind or exceptional control flow. If the function does unwind, its
1146 runtime behavior is undefined.</dd>
1148 <dt><tt><b>readnone</b></tt></dt>
1149 <dd>This attribute indicates that the function computes its result (or decides
1150 to unwind an exception) based strictly on its arguments, without
1151 dereferencing any pointer arguments or otherwise accessing any mutable
1152 state (e.g. memory, control registers, etc) visible to caller functions.
1153 It does not write through any pointer arguments
1154 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1155 changes any state visible to callers. This means that it cannot unwind
1156 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1157 could use the <tt>unwind</tt> instruction.</dd>
1159 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1160 <dd>This attribute indicates that the function does not write through any
1161 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1162 arguments) or otherwise modify any state (e.g. memory, control registers,
1163 etc) visible to caller functions. It may dereference pointer arguments
1164 and read state that may be set in the caller. A readonly function always
1165 returns the same value (or unwinds an exception identically) when called
1166 with the same set of arguments and global state. It cannot unwind an
1167 exception by calling the <tt>C++</tt> exception throwing methods, but may
1168 use the <tt>unwind</tt> instruction.</dd>
1170 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1171 <dd>This attribute indicates that the function should emit a stack smashing
1172 protector. It is in the form of a "canary"—a random value placed on
1173 the stack before the local variables that's checked upon return from the
1174 function to see if it has been overwritten. A heuristic is used to
1175 determine if a function needs stack protectors or not.<br>
1177 If a function that has an <tt>ssp</tt> attribute is inlined into a
1178 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1179 function will have an <tt>ssp</tt> attribute.</dd>
1181 <dt><tt><b>sspreq</b></tt></dt>
1182 <dd>This attribute indicates that the function should <em>always</em> emit a
1183 stack smashing protector. This overrides
1184 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1186 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1187 function that doesn't have an <tt>sspreq</tt> attribute or which has
1188 an <tt>ssp</tt> attribute, then the resulting function will have
1189 an <tt>sspreq</tt> attribute.</dd>
1191 <dt><tt><b>noredzone</b></tt></dt>
1192 <dd>This attribute indicates that the code generator should not use a red
1193 zone, even if the target-specific ABI normally permits it.</dd>
1195 <dt><tt><b>noimplicitfloat</b></tt></dt>
1196 <dd>This attributes disables implicit floating point instructions.</dd>
1198 <dt><tt><b>naked</b></tt></dt>
1199 <dd>This attribute disables prologue / epilogue emission for the function.
1200 This can have very system-specific consequences.</dd>
1205 <!-- ======================================================================= -->
1206 <div class="doc_subsection">
1207 <a name="moduleasm">Module-Level Inline Assembly</a>
1210 <div class="doc_text">
1212 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1213 the GCC "file scope inline asm" blocks. These blocks are internally
1214 concatenated by LLVM and treated as a single unit, but may be separated in
1215 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1217 <div class="doc_code">
1219 module asm "inline asm code goes here"
1220 module asm "more can go here"
1224 <p>The strings can contain any character by escaping non-printable characters.
1225 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1228 <p>The inline asm code is simply printed to the machine code .s file when
1229 assembly code is generated.</p>
1233 <!-- ======================================================================= -->
1234 <div class="doc_subsection">
1235 <a name="datalayout">Data Layout</a>
1238 <div class="doc_text">
1240 <p>A module may specify a target specific data layout string that specifies how
1241 data is to be laid out in memory. The syntax for the data layout is
1244 <div class="doc_code">
1246 target datalayout = "<i>layout specification</i>"
1250 <p>The <i>layout specification</i> consists of a list of specifications
1251 separated by the minus sign character ('-'). Each specification starts with
1252 a letter and may include other information after the letter to define some
1253 aspect of the data layout. The specifications accepted are as follows:</p>
1257 <dd>Specifies that the target lays out data in big-endian form. That is, the
1258 bits with the most significance have the lowest address location.</dd>
1261 <dd>Specifies that the target lays out data in little-endian form. That is,
1262 the bits with the least significance have the lowest address
1265 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1266 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1267 <i>preferred</i> alignments. All sizes are in bits. Specifying
1268 the <i>pref</i> alignment is optional. If omitted, the
1269 preceding <tt>:</tt> should be omitted too.</dd>
1271 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1272 <dd>This specifies the alignment for an integer type of a given bit
1273 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1275 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1276 <dd>This specifies the alignment for a vector type of a given bit
1279 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1280 <dd>This specifies the alignment for a floating point type of a given bit
1281 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1284 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1285 <dd>This specifies the alignment for an aggregate type of a given bit
1288 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1289 <dd>This specifies the alignment for a stack object of a given bit
1292 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1293 <dd>This specifies a set of native integer widths for the target CPU
1294 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1295 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1296 this set are considered to support most general arithmetic
1297 operations efficiently.</dd>
1300 <p>When constructing the data layout for a given target, LLVM starts with a
1301 default set of specifications which are then (possibly) overriden by the
1302 specifications in the <tt>datalayout</tt> keyword. The default specifications
1303 are given in this list:</p>
1306 <li><tt>E</tt> - big endian</li>
1307 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1308 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1309 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1310 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1311 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1312 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1313 alignment of 64-bits</li>
1314 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1315 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1316 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1317 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1318 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1319 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1322 <p>When LLVM is determining the alignment for a given type, it uses the
1323 following rules:</p>
1326 <li>If the type sought is an exact match for one of the specifications, that
1327 specification is used.</li>
1329 <li>If no match is found, and the type sought is an integer type, then the
1330 smallest integer type that is larger than the bitwidth of the sought type
1331 is used. If none of the specifications are larger than the bitwidth then
1332 the the largest integer type is used. For example, given the default
1333 specifications above, the i7 type will use the alignment of i8 (next
1334 largest) while both i65 and i256 will use the alignment of i64 (largest
1337 <li>If no match is found, and the type sought is a vector type, then the
1338 largest vector type that is smaller than the sought vector type will be
1339 used as a fall back. This happens because <128 x double> can be
1340 implemented in terms of 64 <2 x double>, for example.</li>
1345 <!-- ======================================================================= -->
1346 <div class="doc_subsection">
1347 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1350 <div class="doc_text">
1352 <p>Any memory access must be done through a pointer value associated
1353 with an address range of the memory access, otherwise the behavior
1354 is undefined. Pointer values are associated with address ranges
1355 according to the following rules:</p>
1358 <li>A pointer value formed from a
1359 <tt><a href="#i_getelementptr">getelementptr</a></tt> instruction
1360 is associated with the addresses associated with the first operand
1361 of the <tt>getelementptr</tt>.</li>
1362 <li>An address of a global variable is associated with the address
1363 range of the variable's storage.</li>
1364 <li>The result value of an allocation instruction is associated with
1365 the address range of the allocated storage.</li>
1366 <li>A null pointer in the default address-space is associated with
1368 <li>A pointer value formed by an
1369 <tt><a href="#i_inttoptr">inttoptr</a></tt> is associated with all
1370 address ranges of all pointer values that contribute (directly or
1371 indirectly) to the computation of the pointer's value.</li>
1372 <li>The result value of a
1373 <tt><a href="#i_bitcast">bitcast</a></tt> is associated with all
1374 addresses associated with the operand of the <tt>bitcast</tt>.</li>
1375 <li>An integer constant other than zero or a pointer value returned
1376 from a function not defined within LLVM may be associated with address
1377 ranges allocated through mechanisms other than those provided by
1378 LLVM. Such ranges shall not overlap with any ranges of addresses
1379 allocated by mechanisms provided by LLVM.</li>
1382 <p>LLVM IR does not associate types with memory. The result type of a
1383 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1384 alignment of the memory from which to load, as well as the
1385 interpretation of the value. The first operand of a
1386 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1387 and alignment of the store.</p>
1389 <p>Consequently, type-based alias analysis, aka TBAA, aka
1390 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1391 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1392 additional information which specialized optimization passes may use
1393 to implement type-based alias analysis.</p>
1397 <!-- *********************************************************************** -->
1398 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1399 <!-- *********************************************************************** -->
1401 <div class="doc_text">
1403 <p>The LLVM type system is one of the most important features of the
1404 intermediate representation. Being typed enables a number of optimizations
1405 to be performed on the intermediate representation directly, without having
1406 to do extra analyses on the side before the transformation. A strong type
1407 system makes it easier to read the generated code and enables novel analyses
1408 and transformations that are not feasible to perform on normal three address
1409 code representations.</p>
1413 <!-- ======================================================================= -->
1414 <div class="doc_subsection"> <a name="t_classifications">Type
1415 Classifications</a> </div>
1417 <div class="doc_text">
1419 <p>The types fall into a few useful classifications:</p>
1421 <table border="1" cellspacing="0" cellpadding="4">
1423 <tr><th>Classification</th><th>Types</th></tr>
1425 <td><a href="#t_integer">integer</a></td>
1426 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1429 <td><a href="#t_floating">floating point</a></td>
1430 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1433 <td><a name="t_firstclass">first class</a></td>
1434 <td><a href="#t_integer">integer</a>,
1435 <a href="#t_floating">floating point</a>,
1436 <a href="#t_pointer">pointer</a>,
1437 <a href="#t_vector">vector</a>,
1438 <a href="#t_struct">structure</a>,
1439 <a href="#t_union">union</a>,
1440 <a href="#t_array">array</a>,
1441 <a href="#t_label">label</a>,
1442 <a href="#t_metadata">metadata</a>.
1446 <td><a href="#t_primitive">primitive</a></td>
1447 <td><a href="#t_label">label</a>,
1448 <a href="#t_void">void</a>,
1449 <a href="#t_floating">floating point</a>,
1450 <a href="#t_metadata">metadata</a>.</td>
1453 <td><a href="#t_derived">derived</a></td>
1454 <td><a href="#t_array">array</a>,
1455 <a href="#t_function">function</a>,
1456 <a href="#t_pointer">pointer</a>,
1457 <a href="#t_struct">structure</a>,
1458 <a href="#t_pstruct">packed structure</a>,
1459 <a href="#t_union">union</a>,
1460 <a href="#t_vector">vector</a>,
1461 <a href="#t_opaque">opaque</a>.
1467 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1468 important. Values of these types are the only ones which can be produced by
1473 <!-- ======================================================================= -->
1474 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1476 <div class="doc_text">
1478 <p>The primitive types are the fundamental building blocks of the LLVM
1483 <!-- _______________________________________________________________________ -->
1484 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1486 <div class="doc_text">
1489 <p>The integer type is a very simple type that simply specifies an arbitrary
1490 bit width for the integer type desired. Any bit width from 1 bit to
1491 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1498 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1502 <table class="layout">
1504 <td class="left"><tt>i1</tt></td>
1505 <td class="left">a single-bit integer.</td>
1508 <td class="left"><tt>i32</tt></td>
1509 <td class="left">a 32-bit integer.</td>
1512 <td class="left"><tt>i1942652</tt></td>
1513 <td class="left">a really big integer of over 1 million bits.</td>
1519 <!-- _______________________________________________________________________ -->
1520 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1522 <div class="doc_text">
1526 <tr><th>Type</th><th>Description</th></tr>
1527 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1528 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1529 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1530 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1531 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1537 <!-- _______________________________________________________________________ -->
1538 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1540 <div class="doc_text">
1543 <p>The void type does not represent any value and has no size.</p>
1552 <!-- _______________________________________________________________________ -->
1553 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1555 <div class="doc_text">
1558 <p>The label type represents code labels.</p>
1567 <!-- _______________________________________________________________________ -->
1568 <div class="doc_subsubsection"> <a name="t_metadata">Metadata Type</a> </div>
1570 <div class="doc_text">
1573 <p>The metadata type represents embedded metadata. No derived types may be
1574 created from metadata except for <a href="#t_function">function</a>
1585 <!-- ======================================================================= -->
1586 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1588 <div class="doc_text">
1590 <p>The real power in LLVM comes from the derived types in the system. This is
1591 what allows a programmer to represent arrays, functions, pointers, and other
1592 useful types. Each of these types contain one or more element types which
1593 may be a primitive type, or another derived type. For example, it is
1594 possible to have a two dimensional array, using an array as the element type
1595 of another array.</p>
1600 <!-- _______________________________________________________________________ -->
1601 <div class="doc_subsubsection"> <a name="t_aggregate">Aggregate Types</a> </div>
1603 <div class="doc_text">
1605 <p>Aggregate Types are a subset of derived types that can contain multiple
1606 member types. <a href="#t_array">Arrays</a>,
1607 <a href="#t_struct">structs</a>, <a href="#t_vector">vectors</a> and
1608 <a href="#t_union">unions</a> are aggregate types.</p>
1614 <!-- _______________________________________________________________________ -->
1615 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1617 <div class="doc_text">
1620 <p>The array type is a very simple derived type that arranges elements
1621 sequentially in memory. The array type requires a size (number of elements)
1622 and an underlying data type.</p>
1626 [<# elements> x <elementtype>]
1629 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1630 be any type with a size.</p>
1633 <table class="layout">
1635 <td class="left"><tt>[40 x i32]</tt></td>
1636 <td class="left">Array of 40 32-bit integer values.</td>
1639 <td class="left"><tt>[41 x i32]</tt></td>
1640 <td class="left">Array of 41 32-bit integer values.</td>
1643 <td class="left"><tt>[4 x i8]</tt></td>
1644 <td class="left">Array of 4 8-bit integer values.</td>
1647 <p>Here are some examples of multidimensional arrays:</p>
1648 <table class="layout">
1650 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1651 <td class="left">3x4 array of 32-bit integer values.</td>
1654 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1655 <td class="left">12x10 array of single precision floating point values.</td>
1658 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1659 <td class="left">2x3x4 array of 16-bit integer values.</td>
1663 <p>There is no restriction on indexing beyond the end of the array implied by
1664 a static type (though there are restrictions on indexing beyond the bounds
1665 of an allocated object in some cases). This means that single-dimension
1666 'variable sized array' addressing can be implemented in LLVM with a zero
1667 length array type. An implementation of 'pascal style arrays' in LLVM could
1668 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1672 <!-- _______________________________________________________________________ -->
1673 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1675 <div class="doc_text">
1678 <p>The function type can be thought of as a function signature. It consists of
1679 a return type and a list of formal parameter types. The return type of a
1680 function type is a scalar type, a void type, a struct type, or a union
1681 type. If the return type is a struct type then all struct elements must be
1682 of first class types, and the struct must have at least one element.</p>
1686 <returntype> (<parameter list>)
1689 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1690 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1691 which indicates that the function takes a variable number of arguments.
1692 Variable argument functions can access their arguments with
1693 the <a href="#int_varargs">variable argument handling intrinsic</a>
1694 functions. '<tt><returntype></tt>' is any type except
1695 <a href="#t_label">label</a>.</p>
1698 <table class="layout">
1700 <td class="left"><tt>i32 (i32)</tt></td>
1701 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1703 </tr><tr class="layout">
1704 <td class="left"><tt>float (i16, i32 *) *
1706 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1707 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
1708 returning <tt>float</tt>.
1710 </tr><tr class="layout">
1711 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1712 <td class="left">A vararg function that takes at least one
1713 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1714 which returns an integer. This is the signature for <tt>printf</tt> in
1717 </tr><tr class="layout">
1718 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1719 <td class="left">A function taking an <tt>i32</tt>, returning a
1720 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
1727 <!-- _______________________________________________________________________ -->
1728 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1730 <div class="doc_text">
1733 <p>The structure type is used to represent a collection of data members together
1734 in memory. The packing of the field types is defined to match the ABI of the
1735 underlying processor. The elements of a structure may be any type that has a
1738 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
1739 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
1740 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
1741 Structures in registers are accessed using the
1742 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
1743 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
1746 { <type list> }
1750 <table class="layout">
1752 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1753 <td class="left">A triple of three <tt>i32</tt> values</td>
1754 </tr><tr class="layout">
1755 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1756 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1757 second element is a <a href="#t_pointer">pointer</a> to a
1758 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1759 an <tt>i32</tt>.</td>
1765 <!-- _______________________________________________________________________ -->
1766 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1769 <div class="doc_text">
1772 <p>The packed structure type is used to represent a collection of data members
1773 together in memory. There is no padding between fields. Further, the
1774 alignment of a packed structure is 1 byte. The elements of a packed
1775 structure may be any type that has a size.</p>
1777 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1778 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1779 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1783 < { <type list> } >
1787 <table class="layout">
1789 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1790 <td class="left">A triple of three <tt>i32</tt> values</td>
1791 </tr><tr class="layout">
1793 <tt>< { float, i32 (i32)* } ></tt></td>
1794 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1795 second element is a <a href="#t_pointer">pointer</a> to a
1796 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1797 an <tt>i32</tt>.</td>
1803 <!-- _______________________________________________________________________ -->
1804 <div class="doc_subsubsection"> <a name="t_union">Union Type</a> </div>
1806 <div class="doc_text">
1809 <p>A union type describes an object with size and alignment suitable for
1810 an object of any one of a given set of types (also known as an "untagged"
1811 union). It is similar in concept and usage to a
1812 <a href="#t_struct">struct</a>, except that all members of the union
1813 have an offset of zero. The elements of a union may be any type that has a
1814 size. Unions must have at least one member - empty unions are not allowed.
1817 <p>The size of the union as a whole will be the size of its largest member,
1818 and the alignment requirements of the union as a whole will be the largest
1819 alignment requirement of any member.</p>
1821 <p>Union members are accessed using '<tt><a href="#i_load">load</a></tt> and
1822 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1823 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
1824 Since all members are at offset zero, the getelementptr instruction does
1825 not affect the address, only the type of the resulting pointer.</p>
1829 union { <type list> }
1833 <table class="layout">
1835 <td class="left"><tt>union { i32, i32*, float }</tt></td>
1836 <td class="left">A union of three types: an <tt>i32</tt>, a pointer to
1837 an <tt>i32</tt>, and a <tt>float</tt>.</td>
1838 </tr><tr class="layout">
1840 <tt>union { float, i32 (i32) * }</tt></td>
1841 <td class="left">A union, where the first element is a <tt>float</tt> and the
1842 second element is a <a href="#t_pointer">pointer</a> to a
1843 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1844 an <tt>i32</tt>.</td>
1850 <!-- _______________________________________________________________________ -->
1851 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1853 <div class="doc_text">
1856 <p>The pointer type is used to specify memory locations.
1857 Pointers are commonly used to reference objects in memory.</p>
1859 <p>Pointer types may have an optional address space attribute defining the
1860 numbered address space where the pointed-to object resides. The default
1861 address space is number zero. The semantics of non-zero address
1862 spaces are target-specific.</p>
1864 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
1865 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1873 <table class="layout">
1875 <td class="left"><tt>[4 x i32]*</tt></td>
1876 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1877 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1880 <td class="left"><tt>i32 (i32 *) *</tt></td>
1881 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1882 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1886 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1887 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1888 that resides in address space #5.</td>
1894 <!-- _______________________________________________________________________ -->
1895 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1897 <div class="doc_text">
1900 <p>A vector type is a simple derived type that represents a vector of elements.
1901 Vector types are used when multiple primitive data are operated in parallel
1902 using a single instruction (SIMD). A vector type requires a size (number of
1903 elements) and an underlying primitive data type. Vector types are considered
1904 <a href="#t_firstclass">first class</a>.</p>
1908 < <# elements> x <elementtype> >
1911 <p>The number of elements is a constant integer value; elementtype may be any
1912 integer or floating point type.</p>
1915 <table class="layout">
1917 <td class="left"><tt><4 x i32></tt></td>
1918 <td class="left">Vector of 4 32-bit integer values.</td>
1921 <td class="left"><tt><8 x float></tt></td>
1922 <td class="left">Vector of 8 32-bit floating-point values.</td>
1925 <td class="left"><tt><2 x i64></tt></td>
1926 <td class="left">Vector of 2 64-bit integer values.</td>
1932 <!-- _______________________________________________________________________ -->
1933 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1934 <div class="doc_text">
1937 <p>Opaque types are used to represent unknown types in the system. This
1938 corresponds (for example) to the C notion of a forward declared structure
1939 type. In LLVM, opaque types can eventually be resolved to any type (not just
1940 a structure type).</p>
1948 <table class="layout">
1950 <td class="left"><tt>opaque</tt></td>
1951 <td class="left">An opaque type.</td>
1957 <!-- ======================================================================= -->
1958 <div class="doc_subsection">
1959 <a name="t_uprefs">Type Up-references</a>
1962 <div class="doc_text">
1965 <p>An "up reference" allows you to refer to a lexically enclosing type without
1966 requiring it to have a name. For instance, a structure declaration may
1967 contain a pointer to any of the types it is lexically a member of. Example
1968 of up references (with their equivalent as named type declarations)
1972 { \2 * } %x = type { %x* }
1973 { \2 }* %y = type { %y }*
1977 <p>An up reference is needed by the asmprinter for printing out cyclic types
1978 when there is no declared name for a type in the cycle. Because the
1979 asmprinter does not want to print out an infinite type string, it needs a
1980 syntax to handle recursive types that have no names (all names are optional
1988 <p>The level is the count of the lexical type that is being referred to.</p>
1991 <table class="layout">
1993 <td class="left"><tt>\1*</tt></td>
1994 <td class="left">Self-referential pointer.</td>
1997 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1998 <td class="left">Recursive structure where the upref refers to the out-most
2005 <!-- *********************************************************************** -->
2006 <div class="doc_section"> <a name="constants">Constants</a> </div>
2007 <!-- *********************************************************************** -->
2009 <div class="doc_text">
2011 <p>LLVM has several different basic types of constants. This section describes
2012 them all and their syntax.</p>
2016 <!-- ======================================================================= -->
2017 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
2019 <div class="doc_text">
2022 <dt><b>Boolean constants</b></dt>
2023 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2024 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2026 <dt><b>Integer constants</b></dt>
2027 <dd>Standard integers (such as '4') are constants of
2028 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2029 with integer types.</dd>
2031 <dt><b>Floating point constants</b></dt>
2032 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2033 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2034 notation (see below). The assembler requires the exact decimal value of a
2035 floating-point constant. For example, the assembler accepts 1.25 but
2036 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2037 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2039 <dt><b>Null pointer constants</b></dt>
2040 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2041 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2044 <p>The one non-intuitive notation for constants is the hexadecimal form of
2045 floating point constants. For example, the form '<tt>double
2046 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2047 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2048 constants are required (and the only time that they are generated by the
2049 disassembler) is when a floating point constant must be emitted but it cannot
2050 be represented as a decimal floating point number in a reasonable number of
2051 digits. For example, NaN's, infinities, and other special values are
2052 represented in their IEEE hexadecimal format so that assembly and disassembly
2053 do not cause any bits to change in the constants.</p>
2055 <p>When using the hexadecimal form, constants of types float and double are
2056 represented using the 16-digit form shown above (which matches the IEEE754
2057 representation for double); float values must, however, be exactly
2058 representable as IEE754 single precision. Hexadecimal format is always used
2059 for long double, and there are three forms of long double. The 80-bit format
2060 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2061 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2062 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2063 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2064 currently supported target uses this format. Long doubles will only work if
2065 they match the long double format on your target. All hexadecimal formats
2066 are big-endian (sign bit at the left).</p>
2070 <!-- ======================================================================= -->
2071 <div class="doc_subsection">
2072 <a name="aggregateconstants"></a> <!-- old anchor -->
2073 <a name="complexconstants">Complex Constants</a>
2076 <div class="doc_text">
2078 <p>Complex constants are a (potentially recursive) combination of simple
2079 constants and smaller complex constants.</p>
2082 <dt><b>Structure constants</b></dt>
2083 <dd>Structure constants are represented with notation similar to structure
2084 type definitions (a comma separated list of elements, surrounded by braces
2085 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2086 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2087 Structure constants must have <a href="#t_struct">structure type</a>, and
2088 the number and types of elements must match those specified by the
2091 <dt><b>Union constants</b></dt>
2092 <dd>Union constants are represented with notation similar to a structure with
2093 a single element - that is, a single typed element surrounded
2094 by braces (<tt>{}</tt>)). For example: "<tt>{ i32 4 }</tt>". The
2095 <a href="#t_union">union type</a> can be initialized with a single-element
2096 struct as long as the type of the struct element matches the type of
2097 one of the union members.</dd>
2099 <dt><b>Array constants</b></dt>
2100 <dd>Array constants are represented with notation similar to array type
2101 definitions (a comma separated list of elements, surrounded by square
2102 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2103 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2104 the number and types of elements must match those specified by the
2107 <dt><b>Vector constants</b></dt>
2108 <dd>Vector constants are represented with notation similar to vector type
2109 definitions (a comma separated list of elements, surrounded by
2110 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2111 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2112 have <a href="#t_vector">vector type</a>, and the number and types of
2113 elements must match those specified by the type.</dd>
2115 <dt><b>Zero initialization</b></dt>
2116 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2117 value to zero of <em>any</em> type, including scalar and
2118 <a href="#t_aggregate">aggregate</a> types.
2119 This is often used to avoid having to print large zero initializers
2120 (e.g. for large arrays) and is always exactly equivalent to using explicit
2121 zero initializers.</dd>
2123 <dt><b>Metadata node</b></dt>
2124 <dd>A metadata node is a structure-like constant with
2125 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2126 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2127 be interpreted as part of the instruction stream, metadata is a place to
2128 attach additional information such as debug info.</dd>
2133 <!-- ======================================================================= -->
2134 <div class="doc_subsection">
2135 <a name="globalconstants">Global Variable and Function Addresses</a>
2138 <div class="doc_text">
2140 <p>The addresses of <a href="#globalvars">global variables</a>
2141 and <a href="#functionstructure">functions</a> are always implicitly valid
2142 (link-time) constants. These constants are explicitly referenced when
2143 the <a href="#identifiers">identifier for the global</a> is used and always
2144 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2145 legal LLVM file:</p>
2147 <div class="doc_code">
2151 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2157 <!-- ======================================================================= -->
2158 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
2159 <div class="doc_text">
2161 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2162 indicates that the user of the value may receive an unspecified bit-pattern.
2163 Undefined values may be of any type (other than label or void) and be used
2164 anywhere a constant is permitted.</p>
2166 <p>Undefined values are useful because they indicate to the compiler that the
2167 program is well defined no matter what value is used. This gives the
2168 compiler more freedom to optimize. Here are some examples of (potentially
2169 surprising) transformations that are valid (in pseudo IR):</p>
2172 <div class="doc_code">
2184 <p>This is safe because all of the output bits are affected by the undef bits.
2185 Any output bit can have a zero or one depending on the input bits.</p>
2187 <div class="doc_code">
2200 <p>These logical operations have bits that are not always affected by the input.
2201 For example, if "%X" has a zero bit, then the output of the 'and' operation will
2202 always be a zero, no matter what the corresponding bit from the undef is. As
2203 such, it is unsafe to optimize or assume that the result of the and is undef.
2204 However, it is safe to assume that all bits of the undef could be 0, and
2205 optimize the and to 0. Likewise, it is safe to assume that all the bits of
2206 the undef operand to the or could be set, allowing the or to be folded to
2209 <div class="doc_code">
2211 %A = select undef, %X, %Y
2212 %B = select undef, 42, %Y
2213 %C = select %X, %Y, undef
2225 <p>This set of examples show that undefined select (and conditional branch)
2226 conditions can go "either way" but they have to come from one of the two
2227 operands. In the %A example, if %X and %Y were both known to have a clear low
2228 bit, then %A would have to have a cleared low bit. However, in the %C example,
2229 the optimizer is allowed to assume that the undef operand could be the same as
2230 %Y, allowing the whole select to be eliminated.</p>
2233 <div class="doc_code">
2235 %A = xor undef, undef
2254 <p>This example points out that two undef operands are not necessarily the same.
2255 This can be surprising to people (and also matches C semantics) where they
2256 assume that "X^X" is always zero, even if X is undef. This isn't true for a
2257 number of reasons, but the short answer is that an undef "variable" can
2258 arbitrarily change its value over its "live range". This is true because the
2259 "variable" doesn't actually <em>have a live range</em>. Instead, the value is
2260 logically read from arbitrary registers that happen to be around when needed,
2261 so the value is not necessarily consistent over time. In fact, %A and %C need
2262 to have the same semantics or the core LLVM "replace all uses with" concept
2265 <div class="doc_code">
2275 <p>These examples show the crucial difference between an <em>undefined
2276 value</em> and <em>undefined behavior</em>. An undefined value (like undef) is
2277 allowed to have an arbitrary bit-pattern. This means that the %A operation
2278 can be constant folded to undef because the undef could be an SNaN, and fdiv is
2279 not (currently) defined on SNaN's. However, in the second example, we can make
2280 a more aggressive assumption: because the undef is allowed to be an arbitrary
2281 value, we are allowed to assume that it could be zero. Since a divide by zero
2282 has <em>undefined behavior</em>, we are allowed to assume that the operation
2283 does not execute at all. This allows us to delete the divide and all code after
2284 it: since the undefined operation "can't happen", the optimizer can assume that
2285 it occurs in dead code.
2288 <div class="doc_code">
2290 a: store undef -> %X
2291 b: store %X -> undef
2298 <p>These examples reiterate the fdiv example: a store "of" an undefined value
2299 can be assumed to not have any effect: we can assume that the value is
2300 overwritten with bits that happen to match what was already there. However, a
2301 store "to" an undefined location could clobber arbitrary memory, therefore, it
2302 has undefined behavior.</p>
2306 <!-- ======================================================================= -->
2307 <div class="doc_subsection"><a name="trapvalues">Trap Values</a></div>
2308 <div class="doc_text">
2310 <p>Trap values are similar to <a href="undefvalues">undef values</a>, however
2311 instead of representing an unspecified bit pattern, they represent the
2312 fact that an instruction or constant expression which cannot evoke side
2313 effects has nevertheless detected a condition which results in undefined
2316 <p>Any non-void instruction or constant expression other than non-intrinsic
2317 calls or invokes with a trap operand has trap as its result value.
2318 Any instruction with a trap operand which may have side effects emits
2319 those side effects as if it had an undef operand instead.</p>
2321 <p>For example, an <a href="#i_and"><tt>and</tt></a> of a trap value with
2322 zero still has a trap value result. Using that value as an index in a
2323 <a href="#i_getelementptr"><tt>getelementptr</tt></a> yields a trap
2324 result. Using that result as the address of a
2325 <a href="#i_store"><tt>store</tt></a> produces undefined behavior.</p>
2327 <p>There is currently no way of representing a trap constant in the IR; they
2328 only exist when produced by certain instructions, such as an
2329 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag
2330 set, when overflow occurs.</p>
2334 <!-- ======================================================================= -->
2335 <div class="doc_subsection"><a name="blockaddress">Addresses of Basic
2337 <div class="doc_text">
2339 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2341 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2342 basic block in the specified function, and always has an i8* type. Taking
2343 the address of the entry block is illegal.</p>
2345 <p>This value only has defined behavior when used as an operand to the
2346 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction or for comparisons
2347 against null. Pointer equality tests between labels addresses is undefined
2348 behavior - though, again, comparison against null is ok, and no label is
2349 equal to the null pointer. This may also be passed around as an opaque
2350 pointer sized value as long as the bits are not inspected. This allows
2351 <tt>ptrtoint</tt> and arithmetic to be performed on these values so long as
2352 the original value is reconstituted before the <tt>indirectbr</tt>.</p>
2354 <p>Finally, some targets may provide defined semantics when
2355 using the value as the operand to an inline assembly, but that is target
2362 <!-- ======================================================================= -->
2363 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
2366 <div class="doc_text">
2368 <p>Constant expressions are used to allow expressions involving other constants
2369 to be used as constants. Constant expressions may be of
2370 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2371 operation that does not have side effects (e.g. load and call are not
2372 supported). The following is the syntax for constant expressions:</p>
2375 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
2376 <dd>Truncate a constant to another type. The bit size of CST must be larger
2377 than the bit size of TYPE. Both types must be integers.</dd>
2379 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
2380 <dd>Zero extend a constant to another type. The bit size of CST must be
2381 smaller or equal to the bit size of TYPE. Both types must be
2384 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
2385 <dd>Sign extend a constant to another type. The bit size of CST must be
2386 smaller or equal to the bit size of TYPE. Both types must be
2389 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
2390 <dd>Truncate a floating point constant to another floating point type. The
2391 size of CST must be larger than the size of TYPE. Both types must be
2392 floating point.</dd>
2394 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
2395 <dd>Floating point extend a constant to another type. The size of CST must be
2396 smaller or equal to the size of TYPE. Both types must be floating
2399 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
2400 <dd>Convert a floating point constant to the corresponding unsigned integer
2401 constant. TYPE must be a scalar or vector integer type. CST must be of
2402 scalar or vector floating point type. Both CST and TYPE must be scalars,
2403 or vectors of the same number of elements. If the value won't fit in the
2404 integer type, the results are undefined.</dd>
2406 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
2407 <dd>Convert a floating point constant to the corresponding signed integer
2408 constant. TYPE must be a scalar or vector integer type. CST must be of
2409 scalar or vector floating point type. Both CST and TYPE must be scalars,
2410 or vectors of the same number of elements. If the value won't fit in the
2411 integer type, the results are undefined.</dd>
2413 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
2414 <dd>Convert an unsigned integer constant to the corresponding floating point
2415 constant. TYPE must be a scalar or vector floating point type. CST must be
2416 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2417 vectors of the same number of elements. If the value won't fit in the
2418 floating point type, the results are undefined.</dd>
2420 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
2421 <dd>Convert a signed integer constant to the corresponding floating point
2422 constant. TYPE must be a scalar or vector floating point type. CST must be
2423 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2424 vectors of the same number of elements. If the value won't fit in the
2425 floating point type, the results are undefined.</dd>
2427 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
2428 <dd>Convert a pointer typed constant to the corresponding integer constant
2429 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2430 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2431 make it fit in <tt>TYPE</tt>.</dd>
2433 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
2434 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2435 type. CST must be of integer type. The CST value is zero extended,
2436 truncated, or unchanged to make it fit in a pointer size. This one is
2437 <i>really</i> dangerous!</dd>
2439 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
2440 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2441 are the same as those for the <a href="#i_bitcast">bitcast
2442 instruction</a>.</dd>
2444 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2445 <dt><b><tt>getelementptr inbounds ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2446 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2447 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2448 instruction, the index list may have zero or more indexes, which are
2449 required to make sense for the type of "CSTPTR".</dd>
2451 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
2452 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2454 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
2455 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2457 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
2458 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2460 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2461 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2464 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2465 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2468 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2469 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2472 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2473 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2474 be any of the <a href="#binaryops">binary</a>
2475 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2476 on operands are the same as those for the corresponding instruction
2477 (e.g. no bitwise operations on floating point values are allowed).</dd>
2482 <!-- *********************************************************************** -->
2483 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2484 <!-- *********************************************************************** -->
2486 <!-- ======================================================================= -->
2487 <div class="doc_subsection">
2488 <a name="inlineasm">Inline Assembler Expressions</a>
2491 <div class="doc_text">
2493 <p>LLVM supports inline assembler expressions (as opposed
2494 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2495 a special value. This value represents the inline assembler as a string
2496 (containing the instructions to emit), a list of operand constraints (stored
2497 as a string), a flag that indicates whether or not the inline asm
2498 expression has side effects, and a flag indicating whether the function
2499 containing the asm needs to align its stack conservatively. An example
2500 inline assembler expression is:</p>
2502 <div class="doc_code">
2504 i32 (i32) asm "bswap $0", "=r,r"
2508 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2509 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2512 <div class="doc_code">
2514 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2518 <p>Inline asms with side effects not visible in the constraint list must be
2519 marked as having side effects. This is done through the use of the
2520 '<tt>sideeffect</tt>' keyword, like so:</p>
2522 <div class="doc_code">
2524 call void asm sideeffect "eieio", ""()
2528 <p>In some cases inline asms will contain code that will not work unless the
2529 stack is aligned in some way, such as calls or SSE instructions on x86,
2530 yet will not contain code that does that alignment within the asm.
2531 The compiler should make conservative assumptions about what the asm might
2532 contain and should generate its usual stack alignment code in the prologue
2533 if the '<tt>alignstack</tt>' keyword is present:</p>
2535 <div class="doc_code">
2537 call void asm alignstack "eieio", ""()
2541 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2544 <p>TODO: The format of the asm and constraints string still need to be
2545 documented here. Constraints on what can be done (e.g. duplication, moving,
2546 etc need to be documented). This is probably best done by reference to
2547 another document that covers inline asm from a holistic perspective.</p>
2550 <div class="doc_subsubsection">
2551 <a name="inlineasm_md">Inline Asm Metadata</a>
2554 <div class="doc_text">
2556 <p>The call instructions that wrap inline asm nodes may have a "!srcloc" MDNode
2557 attached to it that contains a constant integer. If present, the code
2558 generator will use the integer as the location cookie value when report
2559 errors through the LLVMContext error reporting mechanisms. This allows a
2560 front-end to corrolate backend errors that occur with inline asm back to the
2561 source code that produced it. For example:</p>
2563 <div class="doc_code">
2565 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2567 !42 = !{ i32 1234567 }
2571 <p>It is up to the front-end to make sense of the magic numbers it places in the
2576 <!-- ======================================================================= -->
2577 <div class="doc_subsection"><a name="metadata">Metadata Nodes and Metadata
2581 <div class="doc_text">
2583 <p>LLVM IR allows metadata to be attached to instructions in the program that
2584 can convey extra information about the code to the optimizers and code
2585 generator. One example application of metadata is source-level debug
2586 information. There are two metadata primitives: strings and nodes. All
2587 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2588 preceding exclamation point ('<tt>!</tt>').</p>
2590 <p>A metadata string is a string surrounded by double quotes. It can contain
2591 any character by escaping non-printable characters with "\xx" where "xx" is
2592 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2594 <p>Metadata nodes are represented with notation similar to structure constants
2595 (a comma separated list of elements, surrounded by braces and preceded by an
2596 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2597 10}</tt>". Metadata nodes can have any values as their operand.</p>
2599 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2600 metadata nodes, which can be looked up in the module symbol table. For
2601 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2603 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2604 function is using two metadata arguments.
2606 <div class="doc_code">
2608 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2612 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2613 attached with <tt>add</tt> instruction using <tt>!dbg</tt> identifier.
2615 <div class="doc_code">
2617 %indvar.next = add i64 %indvar, 1, !dbg !21
2623 <!-- *********************************************************************** -->
2624 <div class="doc_section">
2625 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2627 <!-- *********************************************************************** -->
2629 <p>LLVM has a number of "magic" global variables that contain data that affect
2630 code generation or other IR semantics. These are documented here. All globals
2631 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2632 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2635 <!-- ======================================================================= -->
2636 <div class="doc_subsection">
2637 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2640 <div class="doc_text">
2642 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2643 href="#linkage_appending">appending linkage</a>. This array contains a list of
2644 pointers to global variables and functions which may optionally have a pointer
2645 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2651 @llvm.used = appending global [2 x i8*] [
2653 i8* bitcast (i32* @Y to i8*)
2654 ], section "llvm.metadata"
2657 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2658 compiler, assembler, and linker are required to treat the symbol as if there is
2659 a reference to the global that it cannot see. For example, if a variable has
2660 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2661 list, it cannot be deleted. This is commonly used to represent references from
2662 inline asms and other things the compiler cannot "see", and corresponds to
2663 "attribute((used))" in GNU C.</p>
2665 <p>On some targets, the code generator must emit a directive to the assembler or
2666 object file to prevent the assembler and linker from molesting the symbol.</p>
2670 <!-- ======================================================================= -->
2671 <div class="doc_subsection">
2672 <a name="intg_compiler_used">The '<tt>llvm.compiler.used</tt>' Global Variable</a>
2675 <div class="doc_text">
2677 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2678 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2679 touching the symbol. On targets that support it, this allows an intelligent
2680 linker to optimize references to the symbol without being impeded as it would be
2681 by <tt>@llvm.used</tt>.</p>
2683 <p>This is a rare construct that should only be used in rare circumstances, and
2684 should not be exposed to source languages.</p>
2688 <!-- ======================================================================= -->
2689 <div class="doc_subsection">
2690 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2693 <div class="doc_text">
2695 <p>TODO: Describe this.</p>
2699 <!-- ======================================================================= -->
2700 <div class="doc_subsection">
2701 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2704 <div class="doc_text">
2706 <p>TODO: Describe this.</p>
2711 <!-- *********************************************************************** -->
2712 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2713 <!-- *********************************************************************** -->
2715 <div class="doc_text">
2717 <p>The LLVM instruction set consists of several different classifications of
2718 instructions: <a href="#terminators">terminator
2719 instructions</a>, <a href="#binaryops">binary instructions</a>,
2720 <a href="#bitwiseops">bitwise binary instructions</a>,
2721 <a href="#memoryops">memory instructions</a>, and
2722 <a href="#otherops">other instructions</a>.</p>
2726 <!-- ======================================================================= -->
2727 <div class="doc_subsection"> <a name="terminators">Terminator
2728 Instructions</a> </div>
2730 <div class="doc_text">
2732 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
2733 in a program ends with a "Terminator" instruction, which indicates which
2734 block should be executed after the current block is finished. These
2735 terminator instructions typically yield a '<tt>void</tt>' value: they produce
2736 control flow, not values (the one exception being the
2737 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2739 <p>There are seven different terminator instructions: the
2740 '<a href="#i_ret"><tt>ret</tt></a>' instruction, the
2741 '<a href="#i_br"><tt>br</tt></a>' instruction, the
2742 '<a href="#i_switch"><tt>switch</tt></a>' instruction, the
2743 '<a href="#i_indirectbr">'<tt>indirectbr</tt></a>' Instruction, the
2744 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the
2745 '<a href="#i_unwind"><tt>unwind</tt></a>' instruction, and the
2746 '<a href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2750 <!-- _______________________________________________________________________ -->
2751 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2752 Instruction</a> </div>
2754 <div class="doc_text">
2758 ret <type> <value> <i>; Return a value from a non-void function</i>
2759 ret void <i>; Return from void function</i>
2763 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
2764 a value) from a function back to the caller.</p>
2766 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
2767 value and then causes control flow, and one that just causes control flow to
2771 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
2772 return value. The type of the return value must be a
2773 '<a href="#t_firstclass">first class</a>' type.</p>
2775 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
2776 non-void return type and contains a '<tt>ret</tt>' instruction with no return
2777 value or a return value with a type that does not match its type, or if it
2778 has a void return type and contains a '<tt>ret</tt>' instruction with a
2782 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
2783 the calling function's context. If the caller is a
2784 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
2785 instruction after the call. If the caller was an
2786 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
2787 the beginning of the "normal" destination block. If the instruction returns
2788 a value, that value shall set the call or invoke instruction's return
2793 ret i32 5 <i>; Return an integer value of 5</i>
2794 ret void <i>; Return from a void function</i>
2795 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2799 <!-- _______________________________________________________________________ -->
2800 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2802 <div class="doc_text">
2806 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2810 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
2811 different basic block in the current function. There are two forms of this
2812 instruction, corresponding to a conditional branch and an unconditional
2816 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
2817 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
2818 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
2822 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2823 argument is evaluated. If the value is <tt>true</tt>, control flows to the
2824 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2825 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2830 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
2831 br i1 %cond, label %IfEqual, label %IfUnequal
2833 <a href="#i_ret">ret</a> i32 1
2835 <a href="#i_ret">ret</a> i32 0
2840 <!-- _______________________________________________________________________ -->
2841 <div class="doc_subsubsection">
2842 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2845 <div class="doc_text">
2849 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2853 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2854 several different places. It is a generalization of the '<tt>br</tt>'
2855 instruction, allowing a branch to occur to one of many possible
2859 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2860 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
2861 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
2862 The table is not allowed to contain duplicate constant entries.</p>
2865 <p>The <tt>switch</tt> instruction specifies a table of values and
2866 destinations. When the '<tt>switch</tt>' instruction is executed, this table
2867 is searched for the given value. If the value is found, control flow is
2868 transferred to the corresponding destination; otherwise, control flow is
2869 transferred to the default destination.</p>
2871 <h5>Implementation:</h5>
2872 <p>Depending on properties of the target machine and the particular
2873 <tt>switch</tt> instruction, this instruction may be code generated in
2874 different ways. For example, it could be generated as a series of chained
2875 conditional branches or with a lookup table.</p>
2879 <i>; Emulate a conditional br instruction</i>
2880 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2881 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2883 <i>; Emulate an unconditional br instruction</i>
2884 switch i32 0, label %dest [ ]
2886 <i>; Implement a jump table:</i>
2887 switch i32 %val, label %otherwise [ i32 0, label %onzero
2889 i32 2, label %ontwo ]
2895 <!-- _______________________________________________________________________ -->
2896 <div class="doc_subsubsection">
2897 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
2900 <div class="doc_text">
2904 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
2909 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
2910 within the current function, whose address is specified by
2911 "<tt>address</tt>". Address must be derived from a <a
2912 href="#blockaddress">blockaddress</a> constant.</p>
2916 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
2917 rest of the arguments indicate the full set of possible destinations that the
2918 address may point to. Blocks are allowed to occur multiple times in the
2919 destination list, though this isn't particularly useful.</p>
2921 <p>This destination list is required so that dataflow analysis has an accurate
2922 understanding of the CFG.</p>
2926 <p>Control transfers to the block specified in the address argument. All
2927 possible destination blocks must be listed in the label list, otherwise this
2928 instruction has undefined behavior. This implies that jumps to labels
2929 defined in other functions have undefined behavior as well.</p>
2931 <h5>Implementation:</h5>
2933 <p>This is typically implemented with a jump through a register.</p>
2937 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
2943 <!-- _______________________________________________________________________ -->
2944 <div class="doc_subsubsection">
2945 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2948 <div class="doc_text">
2952 <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>]
2953 to label <normal label> unwind label <exception label>
2957 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2958 function, with the possibility of control flow transfer to either the
2959 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
2960 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
2961 control flow will return to the "normal" label. If the callee (or any
2962 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
2963 instruction, control is interrupted and continued at the dynamically nearest
2964 "exception" label.</p>
2967 <p>This instruction requires several arguments:</p>
2970 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
2971 convention</a> the call should use. If none is specified, the call
2972 defaults to using C calling conventions.</li>
2974 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2975 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
2976 '<tt>inreg</tt>' attributes are valid here.</li>
2978 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2979 function value being invoked. In most cases, this is a direct function
2980 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
2981 off an arbitrary pointer to function value.</li>
2983 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2984 function to be invoked. </li>
2986 <li>'<tt>function args</tt>': argument list whose types match the function
2987 signature argument types and parameter attributes. All arguments must be
2988 of <a href="#t_firstclass">first class</a> type. If the function
2989 signature indicates the function accepts a variable number of arguments,
2990 the extra arguments can be specified.</li>
2992 <li>'<tt>normal label</tt>': the label reached when the called function
2993 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2995 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2996 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2998 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2999 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3000 '<tt>readnone</tt>' attributes are valid here.</li>
3004 <p>This instruction is designed to operate as a standard
3005 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3006 primary difference is that it establishes an association with a label, which
3007 is used by the runtime library to unwind the stack.</p>
3009 <p>This instruction is used in languages with destructors to ensure that proper
3010 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3011 exception. Additionally, this is important for implementation of
3012 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3014 <p>For the purposes of the SSA form, the definition of the value returned by the
3015 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3016 block to the "normal" label. If the callee unwinds then no return value is
3019 <p>Note that the code generator does not yet completely support unwind, and
3020 that the invoke/unwind semantics are likely to change in future versions.</p>
3024 %retval = invoke i32 @Test(i32 15) to label %Continue
3025 unwind label %TestCleanup <i>; {i32}:retval set</i>
3026 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3027 unwind label %TestCleanup <i>; {i32}:retval set</i>
3032 <!-- _______________________________________________________________________ -->
3034 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
3035 Instruction</a> </div>
3037 <div class="doc_text">
3045 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
3046 at the first callee in the dynamic call stack which used
3047 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
3048 This is primarily used to implement exception handling.</p>
3051 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
3052 immediately halt. The dynamic call stack is then searched for the
3053 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
3054 Once found, execution continues at the "exceptional" destination block
3055 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
3056 instruction in the dynamic call chain, undefined behavior results.</p>
3058 <p>Note that the code generator does not yet completely support unwind, and
3059 that the invoke/unwind semantics are likely to change in future versions.</p>
3063 <!-- _______________________________________________________________________ -->
3065 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
3066 Instruction</a> </div>
3068 <div class="doc_text">
3076 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3077 instruction is used to inform the optimizer that a particular portion of the
3078 code is not reachable. This can be used to indicate that the code after a
3079 no-return function cannot be reached, and other facts.</p>
3082 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3086 <!-- ======================================================================= -->
3087 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
3089 <div class="doc_text">
3091 <p>Binary operators are used to do most of the computation in a program. They
3092 require two operands of the same type, execute an operation on them, and
3093 produce a single value. The operands might represent multiple data, as is
3094 the case with the <a href="#t_vector">vector</a> data type. The result value
3095 has the same type as its operands.</p>
3097 <p>There are several different binary operators:</p>
3101 <!-- _______________________________________________________________________ -->
3102 <div class="doc_subsubsection">
3103 <a name="i_add">'<tt>add</tt>' Instruction</a>
3106 <div class="doc_text">
3110 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3111 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3112 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3113 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3117 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3120 <p>The two arguments to the '<tt>add</tt>' instruction must
3121 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3122 integer values. Both arguments must have identical types.</p>
3125 <p>The value produced is the integer sum of the two operands.</p>
3127 <p>If the sum has unsigned overflow, the result returned is the mathematical
3128 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3130 <p>Because LLVM integers use a two's complement representation, this instruction
3131 is appropriate for both signed and unsigned integers.</p>
3133 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3134 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3135 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3136 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3137 respectively, occurs.</p>
3141 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3146 <!-- _______________________________________________________________________ -->
3147 <div class="doc_subsubsection">
3148 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3151 <div class="doc_text">
3155 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3159 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3162 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3163 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3164 floating point values. Both arguments must have identical types.</p>
3167 <p>The value produced is the floating point sum of the two operands.</p>
3171 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3176 <!-- _______________________________________________________________________ -->
3177 <div class="doc_subsubsection">
3178 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3181 <div class="doc_text">
3185 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3186 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3187 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3188 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3192 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3195 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3196 '<tt>neg</tt>' instruction present in most other intermediate
3197 representations.</p>
3200 <p>The two arguments to the '<tt>sub</tt>' instruction must
3201 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3202 integer values. Both arguments must have identical types.</p>
3205 <p>The value produced is the integer difference of the two operands.</p>
3207 <p>If the difference has unsigned overflow, the result returned is the
3208 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3211 <p>Because LLVM integers use a two's complement representation, this instruction
3212 is appropriate for both signed and unsigned integers.</p>
3214 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3215 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3216 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3217 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3218 respectively, occurs.</p>
3222 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3223 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3228 <!-- _______________________________________________________________________ -->
3229 <div class="doc_subsubsection">
3230 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3233 <div class="doc_text">
3237 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3241 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3244 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3245 '<tt>fneg</tt>' instruction present in most other intermediate
3246 representations.</p>
3249 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3250 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3251 floating point values. Both arguments must have identical types.</p>
3254 <p>The value produced is the floating point difference of the two operands.</p>
3258 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3259 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3264 <!-- _______________________________________________________________________ -->
3265 <div class="doc_subsubsection">
3266 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3269 <div class="doc_text">
3273 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3274 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3275 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3276 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3280 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3283 <p>The two arguments to the '<tt>mul</tt>' instruction must
3284 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3285 integer values. Both arguments must have identical types.</p>
3288 <p>The value produced is the integer product of the two operands.</p>
3290 <p>If the result of the multiplication has unsigned overflow, the result
3291 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3292 width of the result.</p>
3294 <p>Because LLVM integers use a two's complement representation, and the result
3295 is the same width as the operands, this instruction returns the correct
3296 result for both signed and unsigned integers. If a full product
3297 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3298 be sign-extended or zero-extended as appropriate to the width of the full
3301 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3302 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3303 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3304 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3305 respectively, occurs.</p>
3309 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3314 <!-- _______________________________________________________________________ -->
3315 <div class="doc_subsubsection">
3316 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3319 <div class="doc_text">
3323 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3327 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3330 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3331 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3332 floating point values. Both arguments must have identical types.</p>
3335 <p>The value produced is the floating point product of the two operands.</p>
3339 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3344 <!-- _______________________________________________________________________ -->
3345 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
3348 <div class="doc_text">
3352 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3356 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3359 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3360 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3361 values. Both arguments must have identical types.</p>
3364 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3366 <p>Note that unsigned integer division and signed integer division are distinct
3367 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3369 <p>Division by zero leads to undefined behavior.</p>
3373 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3378 <!-- _______________________________________________________________________ -->
3379 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
3382 <div class="doc_text">
3386 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3387 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3391 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3394 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3395 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3396 values. Both arguments must have identical types.</p>
3399 <p>The value produced is the signed integer quotient of the two operands rounded
3402 <p>Note that signed integer division and unsigned integer division are distinct
3403 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3405 <p>Division by zero leads to undefined behavior. Overflow also leads to
3406 undefined behavior; this is a rare case, but can occur, for example, by doing
3407 a 32-bit division of -2147483648 by -1.</p>
3409 <p>If the <tt>exact</tt> keyword is present, the result value of the
3410 <tt>sdiv</tt> is a <a href="#trapvalues">trap value</a> if the result would
3411 be rounded or if overflow would occur.</p>
3415 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3420 <!-- _______________________________________________________________________ -->
3421 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
3422 Instruction</a> </div>
3424 <div class="doc_text">
3428 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3432 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3435 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3436 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3437 floating point values. Both arguments must have identical types.</p>
3440 <p>The value produced is the floating point quotient of the two operands.</p>
3444 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3449 <!-- _______________________________________________________________________ -->
3450 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3453 <div class="doc_text">
3457 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3461 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3462 division of its two arguments.</p>
3465 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3466 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3467 values. Both arguments must have identical types.</p>
3470 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3471 This instruction always performs an unsigned division to get the
3474 <p>Note that unsigned integer remainder and signed integer remainder are
3475 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3477 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3481 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3486 <!-- _______________________________________________________________________ -->
3487 <div class="doc_subsubsection">
3488 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3491 <div class="doc_text">
3495 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3499 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3500 division of its two operands. This instruction can also take
3501 <a href="#t_vector">vector</a> versions of the values in which case the
3502 elements must be integers.</p>
3505 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3506 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3507 values. Both arguments must have identical types.</p>
3510 <p>This instruction returns the <i>remainder</i> of a division (where the result
3511 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
3512 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
3513 a value. For more information about the difference,
3514 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3515 Math Forum</a>. For a table of how this is implemented in various languages,
3516 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3517 Wikipedia: modulo operation</a>.</p>
3519 <p>Note that signed integer remainder and unsigned integer remainder are
3520 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3522 <p>Taking the remainder of a division by zero leads to undefined behavior.
3523 Overflow also leads to undefined behavior; this is a rare case, but can
3524 occur, for example, by taking the remainder of a 32-bit division of
3525 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3526 lets srem be implemented using instructions that return both the result of
3527 the division and the remainder.)</p>
3531 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3536 <!-- _______________________________________________________________________ -->
3537 <div class="doc_subsubsection">
3538 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
3540 <div class="doc_text">
3544 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3548 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3549 its two operands.</p>
3552 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3553 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3554 floating point values. Both arguments must have identical types.</p>
3557 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3558 has the same sign as the dividend.</p>
3562 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3567 <!-- ======================================================================= -->
3568 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
3569 Operations</a> </div>
3571 <div class="doc_text">
3573 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3574 program. They are generally very efficient instructions and can commonly be
3575 strength reduced from other instructions. They require two operands of the
3576 same type, execute an operation on them, and produce a single value. The
3577 resulting value is the same type as its operands.</p>
3581 <!-- _______________________________________________________________________ -->
3582 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
3583 Instruction</a> </div>
3585 <div class="doc_text">
3589 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3593 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3594 a specified number of bits.</p>
3597 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3598 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3599 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3602 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3603 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3604 is (statically or dynamically) negative or equal to or larger than the number
3605 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3606 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3607 shift amount in <tt>op2</tt>.</p>
3611 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3612 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3613 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3614 <result> = shl i32 1, 32 <i>; undefined</i>
3615 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3620 <!-- _______________________________________________________________________ -->
3621 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
3622 Instruction</a> </div>
3624 <div class="doc_text">
3628 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3632 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
3633 operand shifted to the right a specified number of bits with zero fill.</p>
3636 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
3637 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3638 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3641 <p>This instruction always performs a logical shift right operation. The most
3642 significant bits of the result will be filled with zero bits after the shift.
3643 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
3644 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3645 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3646 shift amount in <tt>op2</tt>.</p>
3650 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
3651 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
3652 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
3653 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
3654 <result> = lshr i32 1, 32 <i>; undefined</i>
3655 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3660 <!-- _______________________________________________________________________ -->
3661 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
3662 Instruction</a> </div>
3663 <div class="doc_text">
3667 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3671 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
3672 operand shifted to the right a specified number of bits with sign
3676 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
3677 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3678 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3681 <p>This instruction always performs an arithmetic shift right operation, The
3682 most significant bits of the result will be filled with the sign bit
3683 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
3684 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
3685 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
3686 the corresponding shift amount in <tt>op2</tt>.</p>
3690 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
3691 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
3692 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
3693 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
3694 <result> = ashr i32 1, 32 <i>; undefined</i>
3695 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3700 <!-- _______________________________________________________________________ -->
3701 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
3702 Instruction</a> </div>
3704 <div class="doc_text">
3708 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3712 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
3716 <p>The two arguments to the '<tt>and</tt>' instruction must be
3717 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3718 values. Both arguments must have identical types.</p>
3721 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
3723 <table border="1" cellspacing="0" cellpadding="4">
3755 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
3756 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
3757 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
3760 <!-- _______________________________________________________________________ -->
3761 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
3763 <div class="doc_text">
3767 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3771 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
3775 <p>The two arguments to the '<tt>or</tt>' instruction must be
3776 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3777 values. Both arguments must have identical types.</p>
3780 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
3782 <table border="1" cellspacing="0" cellpadding="4">
3814 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3815 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3816 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3821 <!-- _______________________________________________________________________ -->
3822 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3823 Instruction</a> </div>
3825 <div class="doc_text">
3829 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3833 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
3834 its two operands. The <tt>xor</tt> is used to implement the "one's
3835 complement" operation, which is the "~" operator in C.</p>
3838 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3839 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3840 values. Both arguments must have identical types.</p>
3843 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3845 <table border="1" cellspacing="0" cellpadding="4">
3877 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3878 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3879 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3880 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3885 <!-- ======================================================================= -->
3886 <div class="doc_subsection">
3887 <a name="vectorops">Vector Operations</a>
3890 <div class="doc_text">
3892 <p>LLVM supports several instructions to represent vector operations in a
3893 target-independent manner. These instructions cover the element-access and
3894 vector-specific operations needed to process vectors effectively. While LLVM
3895 does directly support these vector operations, many sophisticated algorithms
3896 will want to use target-specific intrinsics to take full advantage of a
3897 specific target.</p>
3901 <!-- _______________________________________________________________________ -->
3902 <div class="doc_subsubsection">
3903 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3906 <div class="doc_text">
3910 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3914 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
3915 from a vector at a specified index.</p>
3919 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
3920 of <a href="#t_vector">vector</a> type. The second operand is an index
3921 indicating the position from which to extract the element. The index may be
3925 <p>The result is a scalar of the same type as the element type of
3926 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3927 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3928 results are undefined.</p>
3932 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3937 <!-- _______________________________________________________________________ -->
3938 <div class="doc_subsubsection">
3939 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3942 <div class="doc_text">
3946 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3950 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
3951 vector at a specified index.</p>
3954 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
3955 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
3956 whose type must equal the element type of the first operand. The third
3957 operand is an index indicating the position at which to insert the value.
3958 The index may be a variable.</p>
3961 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
3962 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
3963 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3964 results are undefined.</p>
3968 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3973 <!-- _______________________________________________________________________ -->
3974 <div class="doc_subsubsection">
3975 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3978 <div class="doc_text">
3982 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3986 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3987 from two input vectors, returning a vector with the same element type as the
3988 input and length that is the same as the shuffle mask.</p>
3991 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3992 with types that match each other. The third argument is a shuffle mask whose
3993 element type is always 'i32'. The result of the instruction is a vector
3994 whose length is the same as the shuffle mask and whose element type is the
3995 same as the element type of the first two operands.</p>
3997 <p>The shuffle mask operand is required to be a constant vector with either
3998 constant integer or undef values.</p>
4001 <p>The elements of the two input vectors are numbered from left to right across
4002 both of the vectors. The shuffle mask operand specifies, for each element of
4003 the result vector, which element of the two input vectors the result element
4004 gets. The element selector may be undef (meaning "don't care") and the
4005 second operand may be undef if performing a shuffle from only one vector.</p>
4009 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4010 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4011 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4012 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4013 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4014 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4015 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4016 <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>
4021 <!-- ======================================================================= -->
4022 <div class="doc_subsection">
4023 <a name="aggregateops">Aggregate Operations</a>
4026 <div class="doc_text">
4028 <p>LLVM supports several instructions for working with
4029 <a href="#t_aggregate">aggregate</a> values.</p>
4033 <!-- _______________________________________________________________________ -->
4034 <div class="doc_subsubsection">
4035 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4038 <div class="doc_text">
4042 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4046 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4047 from an <a href="#t_aggregate">aggregate</a> value.</p>
4050 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4051 of <a href="#t_struct">struct</a>, <a href="#t_union">union</a> or
4052 <a href="#t_array">array</a> type. The operands are constant indices to
4053 specify which value to extract in a similar manner as indices in a
4054 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4057 <p>The result is the value at the position in the aggregate specified by the
4062 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4067 <!-- _______________________________________________________________________ -->
4068 <div class="doc_subsubsection">
4069 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4072 <div class="doc_text">
4076 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx> <i>; yields <aggregate type></i>
4080 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4081 in an <a href="#t_aggregate">aggregate</a> value.</p>
4084 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4085 of <a href="#t_struct">struct</a>, <a href="#t_union">union</a> or
4086 <a href="#t_array">array</a> type. The second operand is a first-class
4087 value to insert. The following operands are constant indices indicating
4088 the position at which to insert the value in a similar manner as indices in a
4089 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction. The
4090 value to insert must have the same type as the value identified by the
4094 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4095 that of <tt>val</tt> except that the value at the position specified by the
4096 indices is that of <tt>elt</tt>.</p>
4100 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4101 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4107 <!-- ======================================================================= -->
4108 <div class="doc_subsection">
4109 <a name="memoryops">Memory Access and Addressing Operations</a>
4112 <div class="doc_text">
4114 <p>A key design point of an SSA-based representation is how it represents
4115 memory. In LLVM, no memory locations are in SSA form, which makes things
4116 very simple. This section describes how to read, write, and allocate
4121 <!-- _______________________________________________________________________ -->
4122 <div class="doc_subsubsection">
4123 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4126 <div class="doc_text">
4130 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4134 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4135 currently executing function, to be automatically released when this function
4136 returns to its caller. The object is always allocated in the generic address
4137 space (address space zero).</p>
4140 <p>The '<tt>alloca</tt>' instruction
4141 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4142 runtime stack, returning a pointer of the appropriate type to the program.
4143 If "NumElements" is specified, it is the number of elements allocated,
4144 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4145 specified, the value result of the allocation is guaranteed to be aligned to
4146 at least that boundary. If not specified, or if zero, the target can choose
4147 to align the allocation on any convenient boundary compatible with the
4150 <p>'<tt>type</tt>' may be any sized type.</p>
4153 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4154 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4155 memory is automatically released when the function returns. The
4156 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4157 variables that must have an address available. When the function returns
4158 (either with the <tt><a href="#i_ret">ret</a></tt>
4159 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
4160 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4164 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4165 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4166 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4167 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4172 <!-- _______________________________________________________________________ -->
4173 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
4174 Instruction</a> </div>
4176 <div class="doc_text">
4180 <result> = load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4181 <result> = volatile load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4182 !<index> = !{ i32 1 }
4186 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4189 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4190 from which to load. The pointer must point to
4191 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4192 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4193 number or order of execution of this <tt>load</tt> with other
4194 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
4197 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4198 operation (that is, the alignment of the memory address). A value of 0 or an
4199 omitted <tt>align</tt> argument means that the operation has the preferential
4200 alignment for the target. It is the responsibility of the code emitter to
4201 ensure that the alignment information is correct. Overestimating the
4202 alignment results in undefined behavior. Underestimating the alignment may
4203 produce less efficient code. An alignment of 1 is always safe.</p>
4205 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4206 metatadata name <index> corresponding to a metadata node with
4207 one <tt>i32</tt> entry of value 1. The existence of
4208 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4209 and code generator that this load is not expected to be reused in the cache.
4210 The code generator may select special instructions to save cache bandwidth,
4211 such as the <tt>MOVNT</tt> instruction on x86.</p>
4214 <p>The location of memory pointed to is loaded. If the value being loaded is of
4215 scalar type then the number of bytes read does not exceed the minimum number
4216 of bytes needed to hold all bits of the type. For example, loading an
4217 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4218 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4219 is undefined if the value was not originally written using a store of the
4224 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4225 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4226 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4231 <!-- _______________________________________________________________________ -->
4232 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
4233 Instruction</a> </div>
4235 <div class="doc_text">
4239 store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4240 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4244 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4247 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4248 and an address at which to store it. The type of the
4249 '<tt><pointer></tt>' operand must be a pointer to
4250 the <a href="#t_firstclass">first class</a> type of the
4251 '<tt><value></tt>' operand. If the <tt>store</tt> is marked
4252 as <tt>volatile</tt>, then the optimizer is not allowed to modify the number
4253 or order of execution of this <tt>store</tt> with other
4254 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
4257 <p>The optional constant "align" argument specifies the alignment of the
4258 operation (that is, the alignment of the memory address). A value of 0 or an
4259 omitted "align" argument means that the operation has the preferential
4260 alignment for the target. It is the responsibility of the code emitter to
4261 ensure that the alignment information is correct. Overestimating the
4262 alignment results in an undefined behavior. Underestimating the alignment may
4263 produce less efficient code. An alignment of 1 is always safe.</p>
4265 <p>The optional !nontemporal metadata must reference a single metatadata
4266 name <index> corresponding to a metadata node with one i32 entry of
4267 value 1. The existence of the !nontemporal metatadata on the
4268 instruction tells the optimizer and code generator that this load is
4269 not expected to be reused in the cache. The code generator may
4270 select special instructions to save cache bandwidth, such as the
4271 MOVNT instruction on x86.</p>
4275 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4276 location specified by the '<tt><pointer></tt>' operand. If
4277 '<tt><value></tt>' is of scalar type then the number of bytes written
4278 does not exceed the minimum number of bytes needed to hold all bits of the
4279 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4280 writing a value of a type like <tt>i20</tt> with a size that is not an
4281 integral number of bytes, it is unspecified what happens to the extra bits
4282 that do not belong to the type, but they will typically be overwritten.</p>
4286 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4287 store i32 3, i32* %ptr <i>; yields {void}</i>
4288 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4293 <!-- _______________________________________________________________________ -->
4294 <div class="doc_subsubsection">
4295 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
4298 <div class="doc_text">
4302 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4303 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4307 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4308 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
4309 It performs address calculation only and does not access memory.</p>
4312 <p>The first argument is always a pointer, and forms the basis of the
4313 calculation. The remaining arguments are indices that indicate which of the
4314 elements of the aggregate object are indexed. The interpretation of each
4315 index is dependent on the type being indexed into. The first index always
4316 indexes the pointer value given as the first argument, the second index
4317 indexes a value of the type pointed to (not necessarily the value directly
4318 pointed to, since the first index can be non-zero), etc. The first type
4319 indexed into must be a pointer value, subsequent types can be arrays,
4320 vectors, structs and unions. Note that subsequent types being indexed into
4321 can never be pointers, since that would require loading the pointer before
4322 continuing calculation.</p>
4324 <p>The type of each index argument depends on the type it is indexing into.
4325 When indexing into a (optionally packed) structure or union, only <tt>i32</tt>
4326 integer <b>constants</b> are allowed. When indexing into an array, pointer
4327 or vector, integers of any width are allowed, and they are not required to be
4330 <p>For example, let's consider a C code fragment and how it gets compiled to
4333 <div class="doc_code">
4346 int *foo(struct ST *s) {
4347 return &s[1].Z.B[5][13];
4352 <p>The LLVM code generated by the GCC frontend is:</p>
4354 <div class="doc_code">
4356 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4357 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
4359 define i32* @foo(%ST* %s) {
4361 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
4368 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
4369 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
4370 }</tt>' type, a structure. The second index indexes into the third element
4371 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
4372 i8 }</tt>' type, another structure. The third index indexes into the second
4373 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
4374 array. The two dimensions of the array are subscripted into, yielding an
4375 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
4376 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
4378 <p>Note that it is perfectly legal to index partially through a structure,
4379 returning a pointer to an inner element. Because of this, the LLVM code for
4380 the given testcase is equivalent to:</p>
4383 define i32* @foo(%ST* %s) {
4384 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
4385 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
4386 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
4387 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
4388 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
4393 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
4394 <tt>getelementptr</tt> is a <a href="#trapvalues">trap value</a> if the
4395 base pointer is not an <i>in bounds</i> address of an allocated object,
4396 or if any of the addresses that would be formed by successive addition of
4397 the offsets implied by the indices to the base address with infinitely
4398 precise arithmetic are not an <i>in bounds</i> address of that allocated
4399 object. The <i>in bounds</i> addresses for an allocated object are all
4400 the addresses that point into the object, plus the address one byte past
4403 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
4404 the base address with silently-wrapping two's complement arithmetic, and
4405 the result value of the <tt>getelementptr</tt> may be outside the object
4406 pointed to by the base pointer. The result value may not necessarily be
4407 used to access memory though, even if it happens to point into allocated
4408 storage. See the <a href="#pointeraliasing">Pointer Aliasing Rules</a>
4409 section for more information.</p>
4411 <p>The getelementptr instruction is often confusing. For some more insight into
4412 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
4416 <i>; yields [12 x i8]*:aptr</i>
4417 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4418 <i>; yields i8*:vptr</i>
4419 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4420 <i>; yields i8*:eptr</i>
4421 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4422 <i>; yields i32*:iptr</i>
4423 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4428 <!-- ======================================================================= -->
4429 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
4432 <div class="doc_text">
4434 <p>The instructions in this category are the conversion instructions (casting)
4435 which all take a single operand and a type. They perform various bit
4436 conversions on the operand.</p>
4440 <!-- _______________________________________________________________________ -->
4441 <div class="doc_subsubsection">
4442 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
4444 <div class="doc_text">
4448 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
4452 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
4453 type <tt>ty2</tt>.</p>
4456 <p>The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
4457 be an <a href="#t_integer">integer</a> type, and a type that specifies the
4458 size and type of the result, which must be
4459 an <a href="#t_integer">integer</a> type. The bit size of <tt>value</tt> must
4460 be larger than the bit size of <tt>ty2</tt>. Equal sized types are not
4464 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
4465 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
4466 source size must be larger than the destination size, <tt>trunc</tt> cannot
4467 be a <i>no-op cast</i>. It will always truncate bits.</p>
4471 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
4472 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
4473 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
4478 <!-- _______________________________________________________________________ -->
4479 <div class="doc_subsubsection">
4480 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
4482 <div class="doc_text">
4486 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
4490 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
4495 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
4496 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4497 also be of <a href="#t_integer">integer</a> type. The bit size of the
4498 <tt>value</tt> must be smaller than the bit size of the destination type,
4502 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
4503 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
4505 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
4509 %X = zext i32 257 to i64 <i>; yields i64:257</i>
4510 %Y = zext i1 true to i32 <i>; yields i32:1</i>
4515 <!-- _______________________________________________________________________ -->
4516 <div class="doc_subsubsection">
4517 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
4519 <div class="doc_text">
4523 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
4527 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
4530 <p>The '<tt>sext</tt>' instruction takes a value to cast, which must be of
4531 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4532 also be of <a href="#t_integer">integer</a> type. The bit size of the
4533 <tt>value</tt> must be smaller than the bit size of the destination type,
4537 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
4538 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
4539 of the type <tt>ty2</tt>.</p>
4541 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
4545 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
4546 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
4551 <!-- _______________________________________________________________________ -->
4552 <div class="doc_subsubsection">
4553 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
4556 <div class="doc_text">
4560 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
4564 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
4568 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
4569 point</a> value to cast and a <a href="#t_floating">floating point</a> type
4570 to cast it to. The size of <tt>value</tt> must be larger than the size of
4571 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
4572 <i>no-op cast</i>.</p>
4575 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
4576 <a href="#t_floating">floating point</a> type to a smaller
4577 <a href="#t_floating">floating point</a> type. If the value cannot fit
4578 within the destination type, <tt>ty2</tt>, then the results are
4583 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
4584 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
4589 <!-- _______________________________________________________________________ -->
4590 <div class="doc_subsubsection">
4591 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
4593 <div class="doc_text">
4597 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
4601 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
4602 floating point value.</p>
4605 <p>The '<tt>fpext</tt>' instruction takes a
4606 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
4607 a <a href="#t_floating">floating point</a> type to cast it to. The source
4608 type must be smaller than the destination type.</p>
4611 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
4612 <a href="#t_floating">floating point</a> type to a larger
4613 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
4614 used to make a <i>no-op cast</i> because it always changes bits. Use
4615 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
4619 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
4620 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
4625 <!-- _______________________________________________________________________ -->
4626 <div class="doc_subsubsection">
4627 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
4629 <div class="doc_text">
4633 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
4637 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
4638 unsigned integer equivalent of type <tt>ty2</tt>.</p>
4641 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
4642 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4643 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4644 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4645 vector integer type with the same number of elements as <tt>ty</tt></p>
4648 <p>The '<tt>fptoui</tt>' instruction converts its
4649 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4650 towards zero) unsigned integer value. If the value cannot fit
4651 in <tt>ty2</tt>, the results are undefined.</p>
4655 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
4656 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
4657 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
4662 <!-- _______________________________________________________________________ -->
4663 <div class="doc_subsubsection">
4664 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
4666 <div class="doc_text">
4670 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
4674 <p>The '<tt>fptosi</tt>' instruction converts
4675 <a href="#t_floating">floating point</a> <tt>value</tt> to
4676 type <tt>ty2</tt>.</p>
4679 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
4680 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4681 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4682 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4683 vector integer type with the same number of elements as <tt>ty</tt></p>
4686 <p>The '<tt>fptosi</tt>' instruction converts its
4687 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4688 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
4689 the results are undefined.</p>
4693 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
4694 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
4695 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
4700 <!-- _______________________________________________________________________ -->
4701 <div class="doc_subsubsection">
4702 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
4704 <div class="doc_text">
4708 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4712 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4713 integer and converts that value to the <tt>ty2</tt> type.</p>
4716 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4717 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4718 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4719 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4720 floating point type with the same number of elements as <tt>ty</tt></p>
4723 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4724 integer quantity and converts it to the corresponding floating point
4725 value. If the value cannot fit in the floating point value, the results are
4730 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4731 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4736 <!-- _______________________________________________________________________ -->
4737 <div class="doc_subsubsection">
4738 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4740 <div class="doc_text">
4744 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4748 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
4749 and converts that value to the <tt>ty2</tt> type.</p>
4752 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4753 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4754 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4755 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4756 floating point type with the same number of elements as <tt>ty</tt></p>
4759 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
4760 quantity and converts it to the corresponding floating point value. If the
4761 value cannot fit in the floating point value, the results are undefined.</p>
4765 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4766 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4771 <!-- _______________________________________________________________________ -->
4772 <div class="doc_subsubsection">
4773 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4775 <div class="doc_text">
4779 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4783 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4784 the integer type <tt>ty2</tt>.</p>
4787 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4788 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4789 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4792 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4793 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4794 truncating or zero extending that value to the size of the integer type. If
4795 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4796 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4797 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4802 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4803 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4808 <!-- _______________________________________________________________________ -->
4809 <div class="doc_subsubsection">
4810 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4812 <div class="doc_text">
4816 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4820 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
4821 pointer type, <tt>ty2</tt>.</p>
4824 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4825 value to cast, and a type to cast it to, which must be a
4826 <a href="#t_pointer">pointer</a> type.</p>
4829 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4830 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4831 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4832 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
4833 than the size of a pointer then a zero extension is done. If they are the
4834 same size, nothing is done (<i>no-op cast</i>).</p>
4838 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4839 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4840 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4845 <!-- _______________________________________________________________________ -->
4846 <div class="doc_subsubsection">
4847 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4849 <div class="doc_text">
4853 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4857 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4858 <tt>ty2</tt> without changing any bits.</p>
4861 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
4862 non-aggregate first class value, and a type to cast it to, which must also be
4863 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
4864 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
4865 identical. If the source type is a pointer, the destination type must also be
4866 a pointer. This instruction supports bitwise conversion of vectors to
4867 integers and to vectors of other types (as long as they have the same
4871 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4872 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4873 this conversion. The conversion is done as if the <tt>value</tt> had been
4874 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
4875 be converted to other pointer types with this instruction. To convert
4876 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
4877 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4881 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4882 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4883 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4888 <!-- ======================================================================= -->
4889 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4891 <div class="doc_text">
4893 <p>The instructions in this category are the "miscellaneous" instructions, which
4894 defy better classification.</p>
4898 <!-- _______________________________________________________________________ -->
4899 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4902 <div class="doc_text">
4906 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4910 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
4911 boolean values based on comparison of its two integer, integer vector, or
4912 pointer operands.</p>
4915 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4916 the condition code indicating the kind of comparison to perform. It is not a
4917 value, just a keyword. The possible condition code are:</p>
4920 <li><tt>eq</tt>: equal</li>
4921 <li><tt>ne</tt>: not equal </li>
4922 <li><tt>ugt</tt>: unsigned greater than</li>
4923 <li><tt>uge</tt>: unsigned greater or equal</li>
4924 <li><tt>ult</tt>: unsigned less than</li>
4925 <li><tt>ule</tt>: unsigned less or equal</li>
4926 <li><tt>sgt</tt>: signed greater than</li>
4927 <li><tt>sge</tt>: signed greater or equal</li>
4928 <li><tt>slt</tt>: signed less than</li>
4929 <li><tt>sle</tt>: signed less or equal</li>
4932 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4933 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
4934 typed. They must also be identical types.</p>
4937 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
4938 condition code given as <tt>cond</tt>. The comparison performed always yields
4939 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
4940 result, as follows:</p>
4943 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4944 <tt>false</tt> otherwise. No sign interpretation is necessary or
4947 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4948 <tt>false</tt> otherwise. No sign interpretation is necessary or
4951 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4952 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4954 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4955 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4956 to <tt>op2</tt>.</li>
4958 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4959 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4961 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4962 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4964 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4965 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4967 <li><tt>sge</tt>: interprets the operands as signed values and yields
4968 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4969 to <tt>op2</tt>.</li>
4971 <li><tt>slt</tt>: interprets the operands as signed values and yields
4972 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4974 <li><tt>sle</tt>: interprets the operands as signed values and yields
4975 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4978 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4979 values are compared as if they were integers.</p>
4981 <p>If the operands are integer vectors, then they are compared element by
4982 element. The result is an <tt>i1</tt> vector with the same number of elements
4983 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
4987 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4988 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4989 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4990 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4991 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4992 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4995 <p>Note that the code generator does not yet support vector types with
4996 the <tt>icmp</tt> instruction.</p>
5000 <!-- _______________________________________________________________________ -->
5001 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
5004 <div class="doc_text">
5008 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5012 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
5013 values based on comparison of its operands.</p>
5015 <p>If the operands are floating point scalars, then the result type is a boolean
5016 (<a href="#t_integer"><tt>i1</tt></a>).</p>
5018 <p>If the operands are floating point vectors, then the result type is a vector
5019 of boolean with the same number of elements as the operands being
5023 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
5024 the condition code indicating the kind of comparison to perform. It is not a
5025 value, just a keyword. The possible condition code are:</p>
5028 <li><tt>false</tt>: no comparison, always returns false</li>
5029 <li><tt>oeq</tt>: ordered and equal</li>
5030 <li><tt>ogt</tt>: ordered and greater than </li>
5031 <li><tt>oge</tt>: ordered and greater than or equal</li>
5032 <li><tt>olt</tt>: ordered and less than </li>
5033 <li><tt>ole</tt>: ordered and less than or equal</li>
5034 <li><tt>one</tt>: ordered and not equal</li>
5035 <li><tt>ord</tt>: ordered (no nans)</li>
5036 <li><tt>ueq</tt>: unordered or equal</li>
5037 <li><tt>ugt</tt>: unordered or greater than </li>
5038 <li><tt>uge</tt>: unordered or greater than or equal</li>
5039 <li><tt>ult</tt>: unordered or less than </li>
5040 <li><tt>ule</tt>: unordered or less than or equal</li>
5041 <li><tt>une</tt>: unordered or not equal</li>
5042 <li><tt>uno</tt>: unordered (either nans)</li>
5043 <li><tt>true</tt>: no comparison, always returns true</li>
5046 <p><i>Ordered</i> means that neither operand is a QNAN while
5047 <i>unordered</i> means that either operand may be a QNAN.</p>
5049 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
5050 a <a href="#t_floating">floating point</a> type or
5051 a <a href="#t_vector">vector</a> of floating point type. They must have
5052 identical types.</p>
5055 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
5056 according to the condition code given as <tt>cond</tt>. If the operands are
5057 vectors, then the vectors are compared element by element. Each comparison
5058 performed always yields an <a href="#t_integer">i1</a> result, as
5062 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
5064 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5065 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5067 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5068 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5070 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5071 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5073 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5074 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5076 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5077 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5079 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5080 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5082 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
5084 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
5085 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5087 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
5088 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5090 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
5091 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5093 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
5094 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5096 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
5097 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5099 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
5100 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5102 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
5104 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5109 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
5110 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5111 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5112 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
5115 <p>Note that the code generator does not yet support vector types with
5116 the <tt>fcmp</tt> instruction.</p>
5120 <!-- _______________________________________________________________________ -->
5121 <div class="doc_subsubsection">
5122 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
5125 <div class="doc_text">
5129 <result> = phi <ty> [ <val0>, <label0>], ...
5133 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
5134 SSA graph representing the function.</p>
5137 <p>The type of the incoming values is specified with the first type field. After
5138 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5139 one pair for each predecessor basic block of the current block. Only values
5140 of <a href="#t_firstclass">first class</a> type may be used as the value
5141 arguments to the PHI node. Only labels may be used as the label
5144 <p>There must be no non-phi instructions between the start of a basic block and
5145 the PHI instructions: i.e. PHI instructions must be first in a basic
5148 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5149 occur on the edge from the corresponding predecessor block to the current
5150 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5151 value on the same edge).</p>
5154 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
5155 specified by the pair corresponding to the predecessor basic block that
5156 executed just prior to the current block.</p>
5160 Loop: ; Infinite loop that counts from 0 on up...
5161 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5162 %nextindvar = add i32 %indvar, 1
5168 <!-- _______________________________________________________________________ -->
5169 <div class="doc_subsubsection">
5170 <a name="i_select">'<tt>select</tt>' Instruction</a>
5173 <div class="doc_text">
5177 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5179 <i>selty</i> is either i1 or {<N x i1>}
5183 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
5184 condition, without branching.</p>
5188 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5189 values indicating the condition, and two values of the
5190 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
5191 vectors and the condition is a scalar, then entire vectors are selected, not
5192 individual elements.</p>
5195 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
5196 first value argument; otherwise, it returns the second value argument.</p>
5198 <p>If the condition is a vector of i1, then the value arguments must be vectors
5199 of the same size, and the selection is done element by element.</p>
5203 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
5206 <p>Note that the code generator does not yet support conditions
5207 with vector type.</p>
5211 <!-- _______________________________________________________________________ -->
5212 <div class="doc_subsubsection">
5213 <a name="i_call">'<tt>call</tt>' Instruction</a>
5216 <div class="doc_text">
5220 <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>]
5224 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
5227 <p>This instruction requires several arguments:</p>
5230 <li>The optional "tail" marker indicates that the callee function does not
5231 access any allocas or varargs in the caller. Note that calls may be
5232 marked "tail" even if they do not occur before
5233 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
5234 present, the function call is eligible for tail call optimization,
5235 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
5236 optimized into a jump</a>. The code generator may optimize calls marked
5237 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
5238 sibling call optimization</a> when the caller and callee have
5239 matching signatures, or 2) forced tail call optimization when the
5240 following extra requirements are met:
5242 <li>Caller and callee both have the calling
5243 convention <tt>fastcc</tt>.</li>
5244 <li>The call is in tail position (ret immediately follows call and ret
5245 uses value of call or is void).</li>
5246 <li>Option <tt>-tailcallopt</tt> is enabled,
5247 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
5248 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
5249 constraints are met.</a></li>
5253 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5254 convention</a> the call should use. If none is specified, the call
5255 defaults to using C calling conventions. The calling convention of the
5256 call must match the calling convention of the target function, or else the
5257 behavior is undefined.</li>
5259 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5260 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5261 '<tt>inreg</tt>' attributes are valid here.</li>
5263 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5264 type of the return value. Functions that return no value are marked
5265 <tt><a href="#t_void">void</a></tt>.</li>
5267 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5268 being invoked. The argument types must match the types implied by this
5269 signature. This type can be omitted if the function is not varargs and if
5270 the function type does not return a pointer to a function.</li>
5272 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5273 be invoked. In most cases, this is a direct function invocation, but
5274 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5275 to function value.</li>
5277 <li>'<tt>function args</tt>': argument list whose types match the function
5278 signature argument types and parameter attributes. All arguments must be
5279 of <a href="#t_firstclass">first class</a> type. If the function
5280 signature indicates the function accepts a variable number of arguments,
5281 the extra arguments can be specified.</li>
5283 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5284 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5285 '<tt>readnone</tt>' attributes are valid here.</li>
5289 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5290 a specified function, with its incoming arguments bound to the specified
5291 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5292 function, control flow continues with the instruction after the function
5293 call, and the return value of the function is bound to the result
5298 %retval = call i32 @test(i32 %argc)
5299 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
5300 %X = tail call i32 @foo() <i>; yields i32</i>
5301 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5302 call void %foo(i8 97 signext)
5304 %struct.A = type { i32, i8 }
5305 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
5306 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5307 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
5308 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
5309 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
5312 <p>llvm treats calls to some functions with names and arguments that match the
5313 standard C99 library as being the C99 library functions, and may perform
5314 optimizations or generate code for them under that assumption. This is
5315 something we'd like to change in the future to provide better support for
5316 freestanding environments and non-C-based languages.</p>
5320 <!-- _______________________________________________________________________ -->
5321 <div class="doc_subsubsection">
5322 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5325 <div class="doc_text">
5329 <resultval> = va_arg <va_list*> <arglist>, <argty>
5333 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
5334 the "variable argument" area of a function call. It is used to implement the
5335 <tt>va_arg</tt> macro in C.</p>
5338 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
5339 argument. It returns a value of the specified argument type and increments
5340 the <tt>va_list</tt> to point to the next argument. The actual type
5341 of <tt>va_list</tt> is target specific.</p>
5344 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
5345 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
5346 to the next argument. For more information, see the variable argument
5347 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
5349 <p>It is legal for this instruction to be called in a function which does not
5350 take a variable number of arguments, for example, the <tt>vfprintf</tt>
5353 <p><tt>va_arg</tt> is an LLVM instruction instead of
5354 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
5358 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
5360 <p>Note that the code generator does not yet fully support va_arg on many
5361 targets. Also, it does not currently support va_arg with aggregate types on
5366 <!-- *********************************************************************** -->
5367 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
5368 <!-- *********************************************************************** -->
5370 <div class="doc_text">
5372 <p>LLVM supports the notion of an "intrinsic function". These functions have
5373 well known names and semantics and are required to follow certain
5374 restrictions. Overall, these intrinsics represent an extension mechanism for
5375 the LLVM language that does not require changing all of the transformations
5376 in LLVM when adding to the language (or the bitcode reader/writer, the
5377 parser, etc...).</p>
5379 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
5380 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5381 begin with this prefix. Intrinsic functions must always be external
5382 functions: you cannot define the body of intrinsic functions. Intrinsic
5383 functions may only be used in call or invoke instructions: it is illegal to
5384 take the address of an intrinsic function. Additionally, because intrinsic
5385 functions are part of the LLVM language, it is required if any are added that
5386 they be documented here.</p>
5388 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5389 family of functions that perform the same operation but on different data
5390 types. Because LLVM can represent over 8 million different integer types,
5391 overloading is used commonly to allow an intrinsic function to operate on any
5392 integer type. One or more of the argument types or the result type can be
5393 overloaded to accept any integer type. Argument types may also be defined as
5394 exactly matching a previous argument's type or the result type. This allows
5395 an intrinsic function which accepts multiple arguments, but needs all of them
5396 to be of the same type, to only be overloaded with respect to a single
5397 argument or the result.</p>
5399 <p>Overloaded intrinsics will have the names of its overloaded argument types
5400 encoded into its function name, each preceded by a period. Only those types
5401 which are overloaded result in a name suffix. Arguments whose type is matched
5402 against another type do not. For example, the <tt>llvm.ctpop</tt> function
5403 can take an integer of any width and returns an integer of exactly the same
5404 integer width. This leads to a family of functions such as
5405 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
5406 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5407 suffix is required. Because the argument's type is matched against the return
5408 type, it does not require its own name suffix.</p>
5410 <p>To learn how to add an intrinsic function, please see the
5411 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
5415 <!-- ======================================================================= -->
5416 <div class="doc_subsection">
5417 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
5420 <div class="doc_text">
5422 <p>Variable argument support is defined in LLVM with
5423 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
5424 intrinsic functions. These functions are related to the similarly named
5425 macros defined in the <tt><stdarg.h></tt> header file.</p>
5427 <p>All of these functions operate on arguments that use a target-specific value
5428 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5429 not define what this type is, so all transformations should be prepared to
5430 handle these functions regardless of the type used.</p>
5432 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
5433 instruction and the variable argument handling intrinsic functions are
5436 <div class="doc_code">
5438 define i32 @test(i32 %X, ...) {
5439 ; Initialize variable argument processing
5441 %ap2 = bitcast i8** %ap to i8*
5442 call void @llvm.va_start(i8* %ap2)
5444 ; Read a single integer argument
5445 %tmp = va_arg i8** %ap, i32
5447 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5449 %aq2 = bitcast i8** %aq to i8*
5450 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5451 call void @llvm.va_end(i8* %aq2)
5453 ; Stop processing of arguments.
5454 call void @llvm.va_end(i8* %ap2)
5458 declare void @llvm.va_start(i8*)
5459 declare void @llvm.va_copy(i8*, i8*)
5460 declare void @llvm.va_end(i8*)
5466 <!-- _______________________________________________________________________ -->
5467 <div class="doc_subsubsection">
5468 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
5472 <div class="doc_text">
5476 declare void %llvm.va_start(i8* <arglist>)
5480 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
5481 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
5484 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
5487 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
5488 macro available in C. In a target-dependent way, it initializes
5489 the <tt>va_list</tt> element to which the argument points, so that the next
5490 call to <tt>va_arg</tt> will produce the first variable argument passed to
5491 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
5492 need to know the last argument of the function as the compiler can figure
5497 <!-- _______________________________________________________________________ -->
5498 <div class="doc_subsubsection">
5499 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
5502 <div class="doc_text">
5506 declare void @llvm.va_end(i8* <arglist>)
5510 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
5511 which has been initialized previously
5512 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
5513 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
5516 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
5519 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
5520 macro available in C. In a target-dependent way, it destroys
5521 the <tt>va_list</tt> element to which the argument points. Calls
5522 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
5523 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
5524 with calls to <tt>llvm.va_end</tt>.</p>
5528 <!-- _______________________________________________________________________ -->
5529 <div class="doc_subsubsection">
5530 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
5533 <div class="doc_text">
5537 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
5541 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
5542 from the source argument list to the destination argument list.</p>
5545 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
5546 The second argument is a pointer to a <tt>va_list</tt> element to copy
5550 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
5551 macro available in C. In a target-dependent way, it copies the
5552 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
5553 element. This intrinsic is necessary because
5554 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
5555 arbitrarily complex and require, for example, memory allocation.</p>
5559 <!-- ======================================================================= -->
5560 <div class="doc_subsection">
5561 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
5564 <div class="doc_text">
5566 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
5567 Collection</a> (GC) requires the implementation and generation of these
5568 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
5569 roots on the stack</a>, as well as garbage collector implementations that
5570 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
5571 barriers. Front-ends for type-safe garbage collected languages should generate
5572 these intrinsics to make use of the LLVM garbage collectors. For more details,
5573 see <a href="GarbageCollection.html">Accurate Garbage Collection with
5576 <p>The garbage collection intrinsics only operate on objects in the generic
5577 address space (address space zero).</p>
5581 <!-- _______________________________________________________________________ -->
5582 <div class="doc_subsubsection">
5583 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
5586 <div class="doc_text">
5590 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5594 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
5595 the code generator, and allows some metadata to be associated with it.</p>
5598 <p>The first argument specifies the address of a stack object that contains the
5599 root pointer. The second pointer (which must be either a constant or a
5600 global value address) contains the meta-data to be associated with the
5604 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
5605 location. At compile-time, the code generator generates information to allow
5606 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
5607 intrinsic may only be used in a function which <a href="#gc">specifies a GC
5612 <!-- _______________________________________________________________________ -->
5613 <div class="doc_subsubsection">
5614 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
5617 <div class="doc_text">
5621 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5625 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
5626 locations, allowing garbage collector implementations that require read
5630 <p>The second argument is the address to read from, which should be an address
5631 allocated from the garbage collector. The first object is a pointer to the
5632 start of the referenced object, if needed by the language runtime (otherwise
5636 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
5637 instruction, but may be replaced with substantially more complex code by the
5638 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
5639 may only be used in a function which <a href="#gc">specifies a GC
5644 <!-- _______________________________________________________________________ -->
5645 <div class="doc_subsubsection">
5646 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5649 <div class="doc_text">
5653 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5657 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5658 locations, allowing garbage collector implementations that require write
5659 barriers (such as generational or reference counting collectors).</p>
5662 <p>The first argument is the reference to store, the second is the start of the
5663 object to store it to, and the third is the address of the field of Obj to
5664 store to. If the runtime does not require a pointer to the object, Obj may
5668 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5669 instruction, but may be replaced with substantially more complex code by the
5670 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5671 may only be used in a function which <a href="#gc">specifies a GC
5676 <!-- ======================================================================= -->
5677 <div class="doc_subsection">
5678 <a name="int_codegen">Code Generator Intrinsics</a>
5681 <div class="doc_text">
5683 <p>These intrinsics are provided by LLVM to expose special features that may
5684 only be implemented with code generator support.</p>
5688 <!-- _______________________________________________________________________ -->
5689 <div class="doc_subsubsection">
5690 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5693 <div class="doc_text">
5697 declare i8 *@llvm.returnaddress(i32 <level>)
5701 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5702 target-specific value indicating the return address of the current function
5703 or one of its callers.</p>
5706 <p>The argument to this intrinsic indicates which function to return the address
5707 for. Zero indicates the calling function, one indicates its caller, etc.
5708 The argument is <b>required</b> to be a constant integer value.</p>
5711 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
5712 indicating the return address of the specified call frame, or zero if it
5713 cannot be identified. The value returned by this intrinsic is likely to be
5714 incorrect or 0 for arguments other than zero, so it should only be used for
5715 debugging purposes.</p>
5717 <p>Note that calling this intrinsic does not prevent function inlining or other
5718 aggressive transformations, so the value returned may not be that of the
5719 obvious source-language caller.</p>
5723 <!-- _______________________________________________________________________ -->
5724 <div class="doc_subsubsection">
5725 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5728 <div class="doc_text">
5732 declare i8 *@llvm.frameaddress(i32 <level>)
5736 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5737 target-specific frame pointer value for the specified stack frame.</p>
5740 <p>The argument to this intrinsic indicates which function to return the frame
5741 pointer for. Zero indicates the calling function, one indicates its caller,
5742 etc. The argument is <b>required</b> to be a constant integer value.</p>
5745 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
5746 indicating the frame address of the specified call frame, or zero if it
5747 cannot be identified. The value returned by this intrinsic is likely to be
5748 incorrect or 0 for arguments other than zero, so it should only be used for
5749 debugging purposes.</p>
5751 <p>Note that calling this intrinsic does not prevent function inlining or other
5752 aggressive transformations, so the value returned may not be that of the
5753 obvious source-language caller.</p>
5757 <!-- _______________________________________________________________________ -->
5758 <div class="doc_subsubsection">
5759 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5762 <div class="doc_text">
5766 declare i8 *@llvm.stacksave()
5770 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
5771 of the function stack, for use
5772 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
5773 useful for implementing language features like scoped automatic variable
5774 sized arrays in C99.</p>
5777 <p>This intrinsic returns a opaque pointer value that can be passed
5778 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
5779 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
5780 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
5781 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
5782 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
5783 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
5787 <!-- _______________________________________________________________________ -->
5788 <div class="doc_subsubsection">
5789 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5792 <div class="doc_text">
5796 declare void @llvm.stackrestore(i8 * %ptr)
5800 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5801 the function stack to the state it was in when the
5802 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
5803 executed. This is useful for implementing language features like scoped
5804 automatic variable sized arrays in C99.</p>
5807 <p>See the description
5808 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
5812 <!-- _______________________________________________________________________ -->
5813 <div class="doc_subsubsection">
5814 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5817 <div class="doc_text">
5821 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5825 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
5826 insert a prefetch instruction if supported; otherwise, it is a noop.
5827 Prefetches have no effect on the behavior of the program but can change its
5828 performance characteristics.</p>
5831 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
5832 specifier determining if the fetch should be for a read (0) or write (1),
5833 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5834 locality, to (3) - extremely local keep in cache. The <tt>rw</tt>
5835 and <tt>locality</tt> arguments must be constant integers.</p>
5838 <p>This intrinsic does not modify the behavior of the program. In particular,
5839 prefetches cannot trap and do not produce a value. On targets that support
5840 this intrinsic, the prefetch can provide hints to the processor cache for
5841 better performance.</p>
5845 <!-- _______________________________________________________________________ -->
5846 <div class="doc_subsubsection">
5847 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5850 <div class="doc_text">
5854 declare void @llvm.pcmarker(i32 <id>)
5858 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
5859 Counter (PC) in a region of code to simulators and other tools. The method
5860 is target specific, but it is expected that the marker will use exported
5861 symbols to transmit the PC of the marker. The marker makes no guarantees
5862 that it will remain with any specific instruction after optimizations. It is
5863 possible that the presence of a marker will inhibit optimizations. The
5864 intended use is to be inserted after optimizations to allow correlations of
5865 simulation runs.</p>
5868 <p><tt>id</tt> is a numerical id identifying the marker.</p>
5871 <p>This intrinsic does not modify the behavior of the program. Backends that do
5872 not support this intrinsic may ignore it.</p>
5876 <!-- _______________________________________________________________________ -->
5877 <div class="doc_subsubsection">
5878 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5881 <div class="doc_text">
5885 declare i64 @llvm.readcyclecounter( )
5889 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5890 counter register (or similar low latency, high accuracy clocks) on those
5891 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5892 should map to RPCC. As the backing counters overflow quickly (on the order
5893 of 9 seconds on alpha), this should only be used for small timings.</p>
5896 <p>When directly supported, reading the cycle counter should not modify any
5897 memory. Implementations are allowed to either return a application specific
5898 value or a system wide value. On backends without support, this is lowered
5899 to a constant 0.</p>
5903 <!-- ======================================================================= -->
5904 <div class="doc_subsection">
5905 <a name="int_libc">Standard C Library Intrinsics</a>
5908 <div class="doc_text">
5910 <p>LLVM provides intrinsics for a few important standard C library functions.
5911 These intrinsics allow source-language front-ends to pass information about
5912 the alignment of the pointer arguments to the code generator, providing
5913 opportunity for more efficient code generation.</p>
5917 <!-- _______________________________________________________________________ -->
5918 <div class="doc_subsubsection">
5919 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5922 <div class="doc_text">
5925 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
5926 integer bit width and for different address spaces. Not all targets support
5927 all bit widths however.</p>
5930 declare void @llvm.memcpy.p0i8.p0i8.i32(i8 * <dest>, i8 * <src>,
5931 i32 <len>, i32 <align>, i1 <isvolatile>)
5932 declare void @llvm.memcpy.p0i8.p0i8.i64(i8 * <dest>, i8 * <src>,
5933 i64 <len>, i32 <align>, i1 <isvolatile>)
5937 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5938 source location to the destination location.</p>
5940 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5941 intrinsics do not return a value, takes extra alignment/isvolatile arguments
5942 and the pointers can be in specified address spaces.</p>
5946 <p>The first argument is a pointer to the destination, the second is a pointer
5947 to the source. The third argument is an integer argument specifying the
5948 number of bytes to copy, the fourth argument is the alignment of the
5949 source and destination locations, and the fifth is a boolean indicating a
5950 volatile access.</p>
5952 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
5953 then the caller guarantees that both the source and destination pointers are
5954 aligned to that boundary.</p>
5956 <p>Volatile accesses should not be deleted if dead, but the access behavior is
5957 not very cleanly specified and it is unwise to depend on it.</p>
5961 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5962 source location to the destination location, which are not allowed to
5963 overlap. It copies "len" bytes of memory over. If the argument is known to
5964 be aligned to some boundary, this can be specified as the fourth argument,
5965 otherwise it should be set to 0 or 1.</p>
5969 <!-- _______________________________________________________________________ -->
5970 <div class="doc_subsubsection">
5971 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5974 <div class="doc_text">
5977 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5978 width and for different address space. Not all targets support all bit
5982 declare void @llvm.memmove.p0i8.p0i8.i32(i8 * <dest>, i8 * <src>,
5983 i32 <len>, i32 <align>, i1 <isvolatile>)
5984 declare void @llvm.memmove.p0i8.p0i8.i64(i8 * <dest>, i8 * <src>,
5985 i64 <len>, i32 <align>, i1 <isvolatile>)
5989 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
5990 source location to the destination location. It is similar to the
5991 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
5994 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5995 intrinsics do not return a value, takes extra alignment/isvolatile arguments
5996 and the pointers can be in specified address spaces.</p>
6000 <p>The first argument is a pointer to the destination, the second is a pointer
6001 to the source. The third argument is an integer argument specifying the
6002 number of bytes to copy, the fourth argument is the alignment of the
6003 source and destination locations, and the fifth is a boolean indicating a
6004 volatile access.</p>
6006 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6007 then the caller guarantees that the source and destination pointers are
6008 aligned to that boundary.</p>
6010 <p>Volatile accesses should not be deleted if dead, but the access behavior is
6011 not very cleanly specified and it is unwise to depend on it.</p>
6015 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
6016 source location to the destination location, which may overlap. It copies
6017 "len" bytes of memory over. If the argument is known to be aligned to some
6018 boundary, this can be specified as the fourth argument, otherwise it should
6019 be set to 0 or 1.</p>
6023 <!-- _______________________________________________________________________ -->
6024 <div class="doc_subsubsection">
6025 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
6028 <div class="doc_text">
6031 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6032 width and for different address spaces. Not all targets support all bit
6036 declare void @llvm.memset.p0i8.i32(i8 * <dest>, i8 <val>,
6037 i32 <len>, i32 <align>, i1 <isvolatile>)
6038 declare void @llvm.memset.p0i8.i64(i8 * <dest>, i8 <val>,
6039 i64 <len>, i32 <align>, i1 <isvolatile>)
6043 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
6044 particular byte value.</p>
6046 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
6047 intrinsic does not return a value, takes extra alignment/volatile arguments,
6048 and the destination can be in an arbitrary address space.</p>
6051 <p>The first argument is a pointer to the destination to fill, the second is the
6052 byte value to fill it with, the third argument is an integer argument
6053 specifying the number of bytes to fill, and the fourth argument is the known
6054 alignment of destination location.</p>
6056 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6057 then the caller guarantees that the destination pointer is aligned to that
6060 <p>Volatile accesses should not be deleted if dead, but the access behavior is
6061 not very cleanly specified and it is unwise to depend on it.</p>
6064 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
6065 at the destination location. If the argument is known to be aligned to some
6066 boundary, this can be specified as the fourth argument, otherwise it should
6067 be set to 0 or 1.</p>
6071 <!-- _______________________________________________________________________ -->
6072 <div class="doc_subsubsection">
6073 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
6076 <div class="doc_text">
6079 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
6080 floating point or vector of floating point type. Not all targets support all
6084 declare float @llvm.sqrt.f32(float %Val)
6085 declare double @llvm.sqrt.f64(double %Val)
6086 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6087 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6088 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6092 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
6093 returning the same value as the libm '<tt>sqrt</tt>' functions would.
6094 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
6095 behavior for negative numbers other than -0.0 (which allows for better
6096 optimization, because there is no need to worry about errno being
6097 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
6100 <p>The argument and return value are floating point numbers of the same
6104 <p>This function returns the sqrt of the specified operand if it is a
6105 nonnegative floating point number.</p>
6109 <!-- _______________________________________________________________________ -->
6110 <div class="doc_subsubsection">
6111 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
6114 <div class="doc_text">
6117 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
6118 floating point or vector of floating point type. Not all targets support all
6122 declare float @llvm.powi.f32(float %Val, i32 %power)
6123 declare double @llvm.powi.f64(double %Val, i32 %power)
6124 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6125 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6126 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6130 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
6131 specified (positive or negative) power. The order of evaluation of
6132 multiplications is not defined. When a vector of floating point type is
6133 used, the second argument remains a scalar integer value.</p>
6136 <p>The second argument is an integer power, and the first is a value to raise to
6140 <p>This function returns the first value raised to the second power with an
6141 unspecified sequence of rounding operations.</p>
6145 <!-- _______________________________________________________________________ -->
6146 <div class="doc_subsubsection">
6147 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
6150 <div class="doc_text">
6153 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
6154 floating point or vector of floating point type. Not all targets support all
6158 declare float @llvm.sin.f32(float %Val)
6159 declare double @llvm.sin.f64(double %Val)
6160 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6161 declare fp128 @llvm.sin.f128(fp128 %Val)
6162 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6166 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
6169 <p>The argument and return value are floating point numbers of the same
6173 <p>This function returns the sine of the specified operand, returning the same
6174 values as the libm <tt>sin</tt> functions would, and handles error conditions
6175 in the same way.</p>
6179 <!-- _______________________________________________________________________ -->
6180 <div class="doc_subsubsection">
6181 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
6184 <div class="doc_text">
6187 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
6188 floating point or vector of floating point type. Not all targets support all
6192 declare float @llvm.cos.f32(float %Val)
6193 declare double @llvm.cos.f64(double %Val)
6194 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6195 declare fp128 @llvm.cos.f128(fp128 %Val)
6196 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6200 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
6203 <p>The argument and return value are floating point numbers of the same
6207 <p>This function returns the cosine of the specified operand, returning the same
6208 values as the libm <tt>cos</tt> functions would, and handles error conditions
6209 in the same way.</p>
6213 <!-- _______________________________________________________________________ -->
6214 <div class="doc_subsubsection">
6215 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
6218 <div class="doc_text">
6221 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
6222 floating point or vector of floating point type. Not all targets support all
6226 declare float @llvm.pow.f32(float %Val, float %Power)
6227 declare double @llvm.pow.f64(double %Val, double %Power)
6228 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6229 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6230 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6234 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
6235 specified (positive or negative) power.</p>
6238 <p>The second argument is a floating point power, and the first is a value to
6239 raise to that power.</p>
6242 <p>This function returns the first value raised to the second power, returning
6243 the same values as the libm <tt>pow</tt> functions would, and handles error
6244 conditions in the same way.</p>
6248 <!-- ======================================================================= -->
6249 <div class="doc_subsection">
6250 <a name="int_manip">Bit Manipulation Intrinsics</a>
6253 <div class="doc_text">
6255 <p>LLVM provides intrinsics for a few important bit manipulation operations.
6256 These allow efficient code generation for some algorithms.</p>
6260 <!-- _______________________________________________________________________ -->
6261 <div class="doc_subsubsection">
6262 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
6265 <div class="doc_text">
6268 <p>This is an overloaded intrinsic function. You can use bswap on any integer
6269 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
6272 declare i16 @llvm.bswap.i16(i16 <id>)
6273 declare i32 @llvm.bswap.i32(i32 <id>)
6274 declare i64 @llvm.bswap.i64(i64 <id>)
6278 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
6279 values with an even number of bytes (positive multiple of 16 bits). These
6280 are useful for performing operations on data that is not in the target's
6281 native byte order.</p>
6284 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
6285 and low byte of the input i16 swapped. Similarly,
6286 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
6287 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
6288 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
6289 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
6290 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
6291 more, respectively).</p>
6295 <!-- _______________________________________________________________________ -->
6296 <div class="doc_subsubsection">
6297 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
6300 <div class="doc_text">
6303 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6304 width. Not all targets support all bit widths however.</p>
6307 declare i8 @llvm.ctpop.i8(i8 <src>)
6308 declare i16 @llvm.ctpop.i16(i16 <src>)
6309 declare i32 @llvm.ctpop.i32(i32 <src>)
6310 declare i64 @llvm.ctpop.i64(i64 <src>)
6311 declare i256 @llvm.ctpop.i256(i256 <src>)
6315 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
6319 <p>The only argument is the value to be counted. The argument may be of any
6320 integer type. The return type must match the argument type.</p>
6323 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.</p>
6327 <!-- _______________________________________________________________________ -->
6328 <div class="doc_subsubsection">
6329 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
6332 <div class="doc_text">
6335 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
6336 integer bit width. Not all targets support all bit widths however.</p>
6339 declare i8 @llvm.ctlz.i8 (i8 <src>)
6340 declare i16 @llvm.ctlz.i16(i16 <src>)
6341 declare i32 @llvm.ctlz.i32(i32 <src>)
6342 declare i64 @llvm.ctlz.i64(i64 <src>)
6343 declare i256 @llvm.ctlz.i256(i256 <src>)
6347 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
6348 leading zeros in a variable.</p>
6351 <p>The only argument is the value to be counted. The argument may be of any
6352 integer type. The return type must match the argument type.</p>
6355 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
6356 zeros in a variable. If the src == 0 then the result is the size in bits of
6357 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
6361 <!-- _______________________________________________________________________ -->
6362 <div class="doc_subsubsection">
6363 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
6366 <div class="doc_text">
6369 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
6370 integer bit width. Not all targets support all bit widths however.</p>
6373 declare i8 @llvm.cttz.i8 (i8 <src>)
6374 declare i16 @llvm.cttz.i16(i16 <src>)
6375 declare i32 @llvm.cttz.i32(i32 <src>)
6376 declare i64 @llvm.cttz.i64(i64 <src>)
6377 declare i256 @llvm.cttz.i256(i256 <src>)
6381 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
6385 <p>The only argument is the value to be counted. The argument may be of any
6386 integer type. The return type must match the argument type.</p>
6389 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
6390 zeros in a variable. If the src == 0 then the result is the size in bits of
6391 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
6395 <!-- ======================================================================= -->
6396 <div class="doc_subsection">
6397 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
6400 <div class="doc_text">
6402 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
6406 <!-- _______________________________________________________________________ -->
6407 <div class="doc_subsubsection">
6408 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
6411 <div class="doc_text">
6414 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
6415 on any integer bit width.</p>
6418 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6419 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6420 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6424 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6425 a signed addition of the two arguments, and indicate whether an overflow
6426 occurred during the signed summation.</p>
6429 <p>The arguments (%a and %b) and the first element of the result structure may
6430 be of integer types of any bit width, but they must have the same bit
6431 width. The second element of the result structure must be of
6432 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6433 undergo signed addition.</p>
6436 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6437 a signed addition of the two variables. They return a structure — the
6438 first element of which is the signed summation, and the second element of
6439 which is a bit specifying if the signed summation resulted in an
6444 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6445 %sum = extractvalue {i32, i1} %res, 0
6446 %obit = extractvalue {i32, i1} %res, 1
6447 br i1 %obit, label %overflow, label %normal
6452 <!-- _______________________________________________________________________ -->
6453 <div class="doc_subsubsection">
6454 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
6457 <div class="doc_text">
6460 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
6461 on any integer bit width.</p>
6464 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6465 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6466 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6470 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6471 an unsigned addition of the two arguments, and indicate whether a carry
6472 occurred during the unsigned summation.</p>
6475 <p>The arguments (%a and %b) and the first element of the result structure may
6476 be of integer types of any bit width, but they must have the same bit
6477 width. The second element of the result structure must be of
6478 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6479 undergo unsigned addition.</p>
6482 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6483 an unsigned addition of the two arguments. They return a structure —
6484 the first element of which is the sum, and the second element of which is a
6485 bit specifying if the unsigned summation resulted in a carry.</p>
6489 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6490 %sum = extractvalue {i32, i1} %res, 0
6491 %obit = extractvalue {i32, i1} %res, 1
6492 br i1 %obit, label %carry, label %normal
6497 <!-- _______________________________________________________________________ -->
6498 <div class="doc_subsubsection">
6499 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6502 <div class="doc_text">
6505 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6506 on any integer bit width.</p>
6509 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6510 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6511 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6515 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6516 a signed subtraction of the two arguments, and indicate whether an overflow
6517 occurred during the signed subtraction.</p>
6520 <p>The arguments (%a and %b) and the first element of the result structure may
6521 be of integer types of any bit width, but they must have the same bit
6522 width. The second element of the result structure must be of
6523 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6524 undergo signed subtraction.</p>
6527 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6528 a signed subtraction of the two arguments. They return a structure —
6529 the first element of which is the subtraction, and the second element of
6530 which is a bit specifying if the signed subtraction resulted in an
6535 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6536 %sum = extractvalue {i32, i1} %res, 0
6537 %obit = extractvalue {i32, i1} %res, 1
6538 br i1 %obit, label %overflow, label %normal
6543 <!-- _______________________________________________________________________ -->
6544 <div class="doc_subsubsection">
6545 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6548 <div class="doc_text">
6551 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6552 on any integer bit width.</p>
6555 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6556 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6557 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6561 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6562 an unsigned subtraction of the two arguments, and indicate whether an
6563 overflow occurred during the unsigned subtraction.</p>
6566 <p>The arguments (%a and %b) and the first element of the result structure may
6567 be of integer types of any bit width, but they must have the same bit
6568 width. The second element of the result structure must be of
6569 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6570 undergo unsigned subtraction.</p>
6573 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6574 an unsigned subtraction of the two arguments. They return a structure —
6575 the first element of which is the subtraction, and the second element of
6576 which is a bit specifying if the unsigned subtraction resulted in an
6581 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6582 %sum = extractvalue {i32, i1} %res, 0
6583 %obit = extractvalue {i32, i1} %res, 1
6584 br i1 %obit, label %overflow, label %normal
6589 <!-- _______________________________________________________________________ -->
6590 <div class="doc_subsubsection">
6591 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6594 <div class="doc_text">
6597 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6598 on any integer bit width.</p>
6601 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6602 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6603 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6608 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6609 a signed multiplication of the two arguments, and indicate whether an
6610 overflow occurred during the signed multiplication.</p>
6613 <p>The arguments (%a and %b) and the first element of the result structure may
6614 be of integer types of any bit width, but they must have the same bit
6615 width. The second element of the result structure must be of
6616 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6617 undergo signed multiplication.</p>
6620 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6621 a signed multiplication of the two arguments. They return a structure —
6622 the first element of which is the multiplication, and the second element of
6623 which is a bit specifying if the signed multiplication resulted in an
6628 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6629 %sum = extractvalue {i32, i1} %res, 0
6630 %obit = extractvalue {i32, i1} %res, 1
6631 br i1 %obit, label %overflow, label %normal
6636 <!-- _______________________________________________________________________ -->
6637 <div class="doc_subsubsection">
6638 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6641 <div class="doc_text">
6644 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6645 on any integer bit width.</p>
6648 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6649 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6650 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6654 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6655 a unsigned multiplication of the two arguments, and indicate whether an
6656 overflow occurred during the unsigned multiplication.</p>
6659 <p>The arguments (%a and %b) and the first element of the result structure may
6660 be of integer types of any bit width, but they must have the same bit
6661 width. The second element of the result structure must be of
6662 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6663 undergo unsigned multiplication.</p>
6666 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6667 an unsigned multiplication of the two arguments. They return a structure
6668 — the first element of which is the multiplication, and the second
6669 element of which is a bit specifying if the unsigned multiplication resulted
6674 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6675 %sum = extractvalue {i32, i1} %res, 0
6676 %obit = extractvalue {i32, i1} %res, 1
6677 br i1 %obit, label %overflow, label %normal
6682 <!-- ======================================================================= -->
6683 <div class="doc_subsection">
6684 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
6687 <div class="doc_text">
6689 <p>Half precision floating point is a storage-only format. This means that it is
6690 a dense encoding (in memory) but does not support computation in the
6693 <p>This means that code must first load the half-precision floating point
6694 value as an i16, then convert it to float with <a
6695 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
6696 Computation can then be performed on the float value (including extending to
6697 double etc). To store the value back to memory, it is first converted to
6698 float if needed, then converted to i16 with
6699 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
6700 storing as an i16 value.</p>
6703 <!-- _______________________________________________________________________ -->
6704 <div class="doc_subsubsection">
6705 <a name="int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a>
6708 <div class="doc_text">
6712 declare i16 @llvm.convert.to.fp16(f32 %a)
6716 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
6717 a conversion from single precision floating point format to half precision
6718 floating point format.</p>
6721 <p>The intrinsic function contains single argument - the value to be
6725 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
6726 a conversion from single precision floating point format to half precision
6727 floating point format. The return value is an <tt>i16</tt> which
6728 contains the converted number.</p>
6732 %res = call i16 @llvm.convert.to.fp16(f32 %a)
6733 store i16 %res, i16* @x, align 2
6738 <!-- _______________________________________________________________________ -->
6739 <div class="doc_subsubsection">
6740 <a name="int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a>
6743 <div class="doc_text">
6747 declare f32 @llvm.convert.from.fp16(i16 %a)
6751 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
6752 a conversion from half precision floating point format to single precision
6753 floating point format.</p>
6756 <p>The intrinsic function contains single argument - the value to be
6760 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
6761 conversion from half single precision floating point format to single
6762 precision floating point format. The input half-float value is represented by
6763 an <tt>i16</tt> value.</p>
6767 %a = load i16* @x, align 2
6768 %res = call f32 @llvm.convert.from.fp16(i16 %a)
6773 <!-- ======================================================================= -->
6774 <div class="doc_subsection">
6775 <a name="int_debugger">Debugger Intrinsics</a>
6778 <div class="doc_text">
6780 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
6781 prefix), are described in
6782 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
6783 Level Debugging</a> document.</p>
6787 <!-- ======================================================================= -->
6788 <div class="doc_subsection">
6789 <a name="int_eh">Exception Handling Intrinsics</a>
6792 <div class="doc_text">
6794 <p>The LLVM exception handling intrinsics (which all start with
6795 <tt>llvm.eh.</tt> prefix), are described in
6796 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6797 Handling</a> document.</p>
6801 <!-- ======================================================================= -->
6802 <div class="doc_subsection">
6803 <a name="int_trampoline">Trampoline Intrinsic</a>
6806 <div class="doc_text">
6808 <p>This intrinsic makes it possible to excise one parameter, marked with
6809 the <tt>nest</tt> attribute, from a function. The result is a callable
6810 function pointer lacking the nest parameter - the caller does not need to
6811 provide a value for it. Instead, the value to use is stored in advance in a
6812 "trampoline", a block of memory usually allocated on the stack, which also
6813 contains code to splice the nest value into the argument list. This is used
6814 to implement the GCC nested function address extension.</p>
6816 <p>For example, if the function is
6817 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6818 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
6821 <div class="doc_code">
6823 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6824 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6825 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6826 %fp = bitcast i8* %p to i32 (i32, i32)*
6830 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6831 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6835 <!-- _______________________________________________________________________ -->
6836 <div class="doc_subsubsection">
6837 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6840 <div class="doc_text">
6844 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6848 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
6849 function pointer suitable for executing it.</p>
6852 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6853 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
6854 sufficiently aligned block of memory; this memory is written to by the
6855 intrinsic. Note that the size and the alignment are target-specific - LLVM
6856 currently provides no portable way of determining them, so a front-end that
6857 generates this intrinsic needs to have some target-specific knowledge.
6858 The <tt>func</tt> argument must hold a function bitcast to
6859 an <tt>i8*</tt>.</p>
6862 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
6863 dependent code, turning it into a function. A pointer to this function is
6864 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
6865 function pointer type</a> before being called. The new function's signature
6866 is the same as that of <tt>func</tt> with any arguments marked with
6867 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
6868 is allowed, and it must be of pointer type. Calling the new function is
6869 equivalent to calling <tt>func</tt> with the same argument list, but
6870 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
6871 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
6872 by <tt>tramp</tt> is modified, then the effect of any later call to the
6873 returned function pointer is undefined.</p>
6877 <!-- ======================================================================= -->
6878 <div class="doc_subsection">
6879 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6882 <div class="doc_text">
6884 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
6885 hardware constructs for atomic operations and memory synchronization. This
6886 provides an interface to the hardware, not an interface to the programmer. It
6887 is aimed at a low enough level to allow any programming models or APIs
6888 (Application Programming Interfaces) which need atomic behaviors to map
6889 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6890 hardware provides a "universal IR" for source languages, it also provides a
6891 starting point for developing a "universal" atomic operation and
6892 synchronization IR.</p>
6894 <p>These do <em>not</em> form an API such as high-level threading libraries,
6895 software transaction memory systems, atomic primitives, and intrinsic
6896 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6897 application libraries. The hardware interface provided by LLVM should allow
6898 a clean implementation of all of these APIs and parallel programming models.
6899 No one model or paradigm should be selected above others unless the hardware
6900 itself ubiquitously does so.</p>
6904 <!-- _______________________________________________________________________ -->
6905 <div class="doc_subsubsection">
6906 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6908 <div class="doc_text">
6911 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device> )
6915 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6916 specific pairs of memory access types.</p>
6919 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6920 The first four arguments enables a specific barrier as listed below. The
6921 fifth argument specifies that the barrier applies to io or device or uncached
6925 <li><tt>ll</tt>: load-load barrier</li>
6926 <li><tt>ls</tt>: load-store barrier</li>
6927 <li><tt>sl</tt>: store-load barrier</li>
6928 <li><tt>ss</tt>: store-store barrier</li>
6929 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6933 <p>This intrinsic causes the system to enforce some ordering constraints upon
6934 the loads and stores of the program. This barrier does not
6935 indicate <em>when</em> any events will occur, it only enforces
6936 an <em>order</em> in which they occur. For any of the specified pairs of load
6937 and store operations (f.ex. load-load, or store-load), all of the first
6938 operations preceding the barrier will complete before any of the second
6939 operations succeeding the barrier begin. Specifically the semantics for each
6940 pairing is as follows:</p>
6943 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6944 after the barrier begins.</li>
6945 <li><tt>ls</tt>: All loads before the barrier must complete before any
6946 store after the barrier begins.</li>
6947 <li><tt>ss</tt>: All stores before the barrier must complete before any
6948 store after the barrier begins.</li>
6949 <li><tt>sl</tt>: All stores before the barrier must complete before any
6950 load after the barrier begins.</li>
6953 <p>These semantics are applied with a logical "and" behavior when more than one
6954 is enabled in a single memory barrier intrinsic.</p>
6956 <p>Backends may implement stronger barriers than those requested when they do
6957 not support as fine grained a barrier as requested. Some architectures do
6958 not need all types of barriers and on such architectures, these become
6963 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6964 %ptr = bitcast i8* %mallocP to i32*
6967 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6968 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6969 <i>; guarantee the above finishes</i>
6970 store i32 8, %ptr <i>; before this begins</i>
6975 <!-- _______________________________________________________________________ -->
6976 <div class="doc_subsubsection">
6977 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6980 <div class="doc_text">
6983 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6984 any integer bit width and for different address spaces. Not all targets
6985 support all bit widths however.</p>
6988 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6989 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6990 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6991 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6995 <p>This loads a value in memory and compares it to a given value. If they are
6996 equal, it stores a new value into the memory.</p>
6999 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
7000 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
7001 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
7002 this integer type. While any bit width integer may be used, targets may only
7003 lower representations they support in hardware.</p>
7006 <p>This entire intrinsic must be executed atomically. It first loads the value
7007 in memory pointed to by <tt>ptr</tt> and compares it with the
7008 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
7009 memory. The loaded value is yielded in all cases. This provides the
7010 equivalent of an atomic compare-and-swap operation within the SSA
7015 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7016 %ptr = bitcast i8* %mallocP to i32*
7019 %val1 = add i32 4, 4
7020 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
7021 <i>; yields {i32}:result1 = 4</i>
7022 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7023 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7025 %val2 = add i32 1, 1
7026 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
7027 <i>; yields {i32}:result2 = 8</i>
7028 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
7030 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
7035 <!-- _______________________________________________________________________ -->
7036 <div class="doc_subsubsection">
7037 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
7039 <div class="doc_text">
7042 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
7043 integer bit width. Not all targets support all bit widths however.</p>
7046 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
7047 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
7048 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
7049 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
7053 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
7054 the value from memory. It then stores the value in <tt>val</tt> in the memory
7055 at <tt>ptr</tt>.</p>
7058 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
7059 the <tt>val</tt> argument and the result must be integers of the same bit
7060 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
7061 integer type. The targets may only lower integer representations they
7065 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
7066 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
7067 equivalent of an atomic swap operation within the SSA framework.</p>
7071 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7072 %ptr = bitcast i8* %mallocP to i32*
7075 %val1 = add i32 4, 4
7076 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
7077 <i>; yields {i32}:result1 = 4</i>
7078 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7079 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7081 %val2 = add i32 1, 1
7082 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
7083 <i>; yields {i32}:result2 = 8</i>
7085 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
7086 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
7091 <!-- _______________________________________________________________________ -->
7092 <div class="doc_subsubsection">
7093 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
7097 <div class="doc_text">
7100 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
7101 any integer bit width. Not all targets support all bit widths however.</p>
7104 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
7105 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
7106 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
7107 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
7111 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
7112 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7115 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7116 and the second an integer value. The result is also an integer value. These
7117 integer types can have any bit width, but they must all have the same bit
7118 width. The targets may only lower integer representations they support.</p>
7121 <p>This intrinsic does a series of operations atomically. It first loads the
7122 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
7123 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
7127 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7128 %ptr = bitcast i8* %mallocP to i32*
7130 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
7131 <i>; yields {i32}:result1 = 4</i>
7132 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
7133 <i>; yields {i32}:result2 = 8</i>
7134 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
7135 <i>; yields {i32}:result3 = 10</i>
7136 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
7141 <!-- _______________________________________________________________________ -->
7142 <div class="doc_subsubsection">
7143 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
7147 <div class="doc_text">
7150 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
7151 any integer bit width and for different address spaces. Not all targets
7152 support all bit widths however.</p>
7155 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
7156 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
7157 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
7158 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
7162 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
7163 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7166 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7167 and the second an integer value. The result is also an integer value. These
7168 integer types can have any bit width, but they must all have the same bit
7169 width. The targets may only lower integer representations they support.</p>
7172 <p>This intrinsic does a series of operations atomically. It first loads the
7173 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
7174 result to <tt>ptr</tt>. It yields the original value stored
7175 at <tt>ptr</tt>.</p>
7179 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7180 %ptr = bitcast i8* %mallocP to i32*
7182 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
7183 <i>; yields {i32}:result1 = 8</i>
7184 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
7185 <i>; yields {i32}:result2 = 4</i>
7186 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
7187 <i>; yields {i32}:result3 = 2</i>
7188 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
7193 <!-- _______________________________________________________________________ -->
7194 <div class="doc_subsubsection">
7195 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
7196 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
7197 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
7198 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
7201 <div class="doc_text">
7204 <p>These are overloaded intrinsics. You can
7205 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
7206 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
7207 bit width and for different address spaces. Not all targets support all bit
7211 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
7212 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
7213 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
7214 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
7218 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
7219 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
7220 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
7221 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
7225 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
7226 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
7227 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
7228 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
7232 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
7233 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
7234 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
7235 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
7239 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
7240 the value stored in memory at <tt>ptr</tt>. It yields the original value
7241 at <tt>ptr</tt>.</p>
7244 <p>These intrinsics take two arguments, the first a pointer to an integer value
7245 and the second an integer value. The result is also an integer value. These
7246 integer types can have any bit width, but they must all have the same bit
7247 width. The targets may only lower integer representations they support.</p>
7250 <p>These intrinsics does a series of operations atomically. They first load the
7251 value stored at <tt>ptr</tt>. They then do the bitwise
7252 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
7253 original value stored at <tt>ptr</tt>.</p>
7257 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7258 %ptr = bitcast i8* %mallocP to i32*
7259 store i32 0x0F0F, %ptr
7260 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
7261 <i>; yields {i32}:result0 = 0x0F0F</i>
7262 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
7263 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
7264 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
7265 <i>; yields {i32}:result2 = 0xF0</i>
7266 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
7267 <i>; yields {i32}:result3 = FF</i>
7268 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
7273 <!-- _______________________________________________________________________ -->
7274 <div class="doc_subsubsection">
7275 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
7276 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
7277 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
7278 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
7281 <div class="doc_text">
7284 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
7285 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
7286 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
7287 address spaces. Not all targets support all bit widths however.</p>
7290 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
7291 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
7292 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
7293 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
7297 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
7298 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
7299 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
7300 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
7304 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
7305 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
7306 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
7307 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
7311 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
7312 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
7313 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
7314 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
7318 <p>These intrinsics takes the signed or unsigned minimum or maximum of
7319 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
7320 original value at <tt>ptr</tt>.</p>
7323 <p>These intrinsics take two arguments, the first a pointer to an integer value
7324 and the second an integer value. The result is also an integer value. These
7325 integer types can have any bit width, but they must all have the same bit
7326 width. The targets may only lower integer representations they support.</p>
7329 <p>These intrinsics does a series of operations atomically. They first load the
7330 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
7331 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
7332 yield the original value stored at <tt>ptr</tt>.</p>
7336 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7337 %ptr = bitcast i8* %mallocP to i32*
7339 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
7340 <i>; yields {i32}:result0 = 7</i>
7341 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
7342 <i>; yields {i32}:result1 = -2</i>
7343 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
7344 <i>; yields {i32}:result2 = 8</i>
7345 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
7346 <i>; yields {i32}:result3 = 8</i>
7347 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
7353 <!-- ======================================================================= -->
7354 <div class="doc_subsection">
7355 <a name="int_memorymarkers">Memory Use Markers</a>
7358 <div class="doc_text">
7360 <p>This class of intrinsics exists to information about the lifetime of memory
7361 objects and ranges where variables are immutable.</p>
7365 <!-- _______________________________________________________________________ -->
7366 <div class="doc_subsubsection">
7367 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
7370 <div class="doc_text">
7374 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
7378 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
7379 object's lifetime.</p>
7382 <p>The first argument is a constant integer representing the size of the
7383 object, or -1 if it is variable sized. The second argument is a pointer to
7387 <p>This intrinsic indicates that before this point in the code, the value of the
7388 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7389 never be used and has an undefined value. A load from the pointer that
7390 precedes this intrinsic can be replaced with
7391 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
7395 <!-- _______________________________________________________________________ -->
7396 <div class="doc_subsubsection">
7397 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
7400 <div class="doc_text">
7404 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
7408 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
7409 object's lifetime.</p>
7412 <p>The first argument is a constant integer representing the size of the
7413 object, or -1 if it is variable sized. The second argument is a pointer to
7417 <p>This intrinsic indicates that after this point in the code, the value of the
7418 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7419 never be used and has an undefined value. Any stores into the memory object
7420 following this intrinsic may be removed as dead.
7424 <!-- _______________________________________________________________________ -->
7425 <div class="doc_subsubsection">
7426 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
7429 <div class="doc_text">
7433 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>) readonly
7437 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
7438 a memory object will not change.</p>
7441 <p>The first argument is a constant integer representing the size of the
7442 object, or -1 if it is variable sized. The second argument is a pointer to
7446 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
7447 the return value, the referenced memory location is constant and
7452 <!-- _______________________________________________________________________ -->
7453 <div class="doc_subsubsection">
7454 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
7457 <div class="doc_text">
7461 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7465 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
7466 a memory object are mutable.</p>
7469 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
7470 The second argument is a constant integer representing the size of the
7471 object, or -1 if it is variable sized and the third argument is a pointer
7475 <p>This intrinsic indicates that the memory is mutable again.</p>
7479 <!-- ======================================================================= -->
7480 <div class="doc_subsection">
7481 <a name="int_general">General Intrinsics</a>
7484 <div class="doc_text">
7486 <p>This class of intrinsics is designed to be generic and has no specific
7491 <!-- _______________________________________________________________________ -->
7492 <div class="doc_subsubsection">
7493 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
7496 <div class="doc_text">
7500 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
7504 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
7507 <p>The first argument is a pointer to a value, the second is a pointer to a
7508 global string, the third is a pointer to a global string which is the source
7509 file name, and the last argument is the line number.</p>
7512 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7513 This can be useful for special purpose optimizations that want to look for
7514 these annotations. These have no other defined use, they are ignored by code
7515 generation and optimization.</p>
7519 <!-- _______________________________________________________________________ -->
7520 <div class="doc_subsubsection">
7521 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
7524 <div class="doc_text">
7527 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
7528 any integer bit width.</p>
7531 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
7532 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
7533 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
7534 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
7535 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
7539 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
7542 <p>The first argument is an integer value (result of some expression), the
7543 second is a pointer to a global string, the third is a pointer to a global
7544 string which is the source file name, and the last argument is the line
7545 number. It returns the value of the first argument.</p>
7548 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7549 arbitrary strings. This can be useful for special purpose optimizations that
7550 want to look for these annotations. These have no other defined use, they
7551 are ignored by code generation and optimization.</p>
7555 <!-- _______________________________________________________________________ -->
7556 <div class="doc_subsubsection">
7557 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
7560 <div class="doc_text">
7564 declare void @llvm.trap()
7568 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
7574 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7575 target does not have a trap instruction, this intrinsic will be lowered to
7576 the call of the <tt>abort()</tt> function.</p>
7580 <!-- _______________________________________________________________________ -->
7581 <div class="doc_subsubsection">
7582 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
7585 <div class="doc_text">
7589 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
7593 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
7594 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
7595 ensure that it is placed on the stack before local variables.</p>
7598 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
7599 arguments. The first argument is the value loaded from the stack
7600 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
7601 that has enough space to hold the value of the guard.</p>
7604 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7605 the <tt>AllocaInst</tt> stack slot to be before local variables on the
7606 stack. This is to ensure that if a local variable on the stack is
7607 overwritten, it will destroy the value of the guard. When the function exits,
7608 the guard on the stack is checked against the original guard. If they're
7609 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
7614 <!-- _______________________________________________________________________ -->
7615 <div class="doc_subsubsection">
7616 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
7619 <div class="doc_text">
7623 declare i32 @llvm.objectsize.i32( i8* <object>, i1 <type> )
7624 declare i64 @llvm.objectsize.i64( i8* <object>, i1 <type> )
7628 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information
7629 to the optimizers to discover at compile time either a) when an
7630 operation like memcpy will either overflow a buffer that corresponds to
7631 an object, or b) to determine that a runtime check for overflow isn't
7632 necessary. An object in this context means an allocation of a
7633 specific class, structure, array, or other object.</p>
7636 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
7637 argument is a pointer to or into the <tt>object</tt>. The second argument
7638 is a boolean 0 or 1. This argument determines whether you want the
7639 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
7640 1, variables are not allowed.</p>
7643 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
7644 representing the size of the object concerned or <tt>i32/i64 -1 or 0</tt>
7645 (depending on the <tt>type</tt> argument if the size cannot be determined
7646 at compile time.</p>
7650 <!-- *********************************************************************** -->
7653 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
7654 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
7655 <a href="http://validator.w3.org/check/referer"><img
7656 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
7658 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
7659 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
7660 Last modified: $Date$