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="#paramattrs">Parameter Attributes</a></li>
47 <li><a href="#fnattrs">Function Attributes</a></li>
48 <li><a href="#gc">Garbage Collector Names</a></li>
49 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
50 <li><a href="#datalayout">Data Layout</a></li>
51 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
54 <li><a href="#typesystem">Type System</a>
56 <li><a href="#t_classifications">Type Classifications</a></li>
57 <li><a href="#t_primitive">Primitive Types</a>
59 <li><a href="#t_integer">Integer Type</a></li>
60 <li><a href="#t_floating">Floating Point Types</a></li>
61 <li><a href="#t_void">Void Type</a></li>
62 <li><a href="#t_label">Label Type</a></li>
63 <li><a href="#t_metadata">Metadata Type</a></li>
66 <li><a href="#t_derived">Derived Types</a>
68 <li><a href="#t_array">Array Type</a></li>
69 <li><a href="#t_function">Function Type</a></li>
70 <li><a href="#t_pointer">Pointer Type</a></li>
71 <li><a href="#t_struct">Structure Type</a></li>
72 <li><a href="#t_pstruct">Packed Structure Type</a></li>
73 <li><a href="#t_vector">Vector Type</a></li>
74 <li><a href="#t_opaque">Opaque Type</a></li>
77 <li><a href="#t_uprefs">Type Up-references</a></li>
80 <li><a href="#constants">Constants</a>
82 <li><a href="#simpleconstants">Simple Constants</a></li>
83 <li><a href="#complexconstants">Complex Constants</a></li>
84 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
85 <li><a href="#undefvalues">Undefined Values</a></li>
86 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
87 <li><a href="#constantexprs">Constant Expressions</a></li>
88 <li><a href="#metadata">Embedded Metadata</a></li>
91 <li><a href="#othervalues">Other Values</a>
93 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
96 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
98 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
99 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
100 Global Variable</a></li>
101 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
102 Global Variable</a></li>
103 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
104 Global Variable</a></li>
107 <li><a href="#instref">Instruction Reference</a>
109 <li><a href="#terminators">Terminator Instructions</a>
111 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
112 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
113 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
114 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
115 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
116 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
117 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
120 <li><a href="#binaryops">Binary Operations</a>
122 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
123 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
124 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
125 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
126 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
127 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
128 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
129 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
130 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
131 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
132 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
133 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
136 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
138 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
139 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
140 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
141 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
142 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
143 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
146 <li><a href="#vectorops">Vector Operations</a>
148 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
149 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
150 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
153 <li><a href="#aggregateops">Aggregate Operations</a>
155 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
156 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
159 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
161 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
162 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
163 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
164 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
167 <li><a href="#convertops">Conversion Operations</a>
169 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
170 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
171 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
172 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
173 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
174 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
175 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
176 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
177 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
178 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
179 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
180 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
183 <li><a href="#otherops">Other Operations</a>
185 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
186 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
187 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
188 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
189 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
190 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
195 <li><a href="#intrinsics">Intrinsic Functions</a>
197 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
199 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
200 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
201 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
204 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
206 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
207 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
208 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
211 <li><a href="#int_codegen">Code Generator Intrinsics</a>
213 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
214 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
215 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
216 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
217 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
218 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
219 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
222 <li><a href="#int_libc">Standard C Library Intrinsics</a>
224 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
225 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
226 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
227 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
228 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
229 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
230 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
231 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
234 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
236 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
237 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
238 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
239 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
242 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
244 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
245 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
246 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
247 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
248 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
249 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
252 <li><a href="#int_debugger">Debugger intrinsics</a></li>
253 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
254 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
256 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
259 <li><a href="#int_atomics">Atomic intrinsics</a>
261 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
262 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
263 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
264 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
265 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
266 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
267 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
268 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
269 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
270 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
271 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
272 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
273 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
276 <li><a href="#int_memorymarkers">Memory Use Markers</a>
278 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
279 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
280 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
281 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
284 <li><a href="#int_general">General intrinsics</a>
286 <li><a href="#int_var_annotation">
287 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
288 <li><a href="#int_annotation">
289 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
290 <li><a href="#int_trap">
291 '<tt>llvm.trap</tt>' Intrinsic</a></li>
292 <li><a href="#int_stackprotector">
293 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
300 <div class="doc_author">
301 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
302 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
305 <!-- *********************************************************************** -->
306 <div class="doc_section"> <a name="abstract">Abstract </a></div>
307 <!-- *********************************************************************** -->
309 <div class="doc_text">
311 <p>This document is a reference manual for the LLVM assembly language. LLVM is
312 a Static Single Assignment (SSA) based representation that provides type
313 safety, low-level operations, flexibility, and the capability of representing
314 'all' high-level languages cleanly. It is the common code representation
315 used throughout all phases of the LLVM compilation strategy.</p>
319 <!-- *********************************************************************** -->
320 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
321 <!-- *********************************************************************** -->
323 <div class="doc_text">
325 <p>The LLVM code representation is designed to be used in three different forms:
326 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
327 for fast loading by a Just-In-Time compiler), and as a human readable
328 assembly language representation. This allows LLVM to provide a powerful
329 intermediate representation for efficient compiler transformations and
330 analysis, while providing a natural means to debug and visualize the
331 transformations. The three different forms of LLVM are all equivalent. This
332 document describes the human readable representation and notation.</p>
334 <p>The LLVM representation aims to be light-weight and low-level while being
335 expressive, typed, and extensible at the same time. It aims to be a
336 "universal IR" of sorts, by being at a low enough level that high-level ideas
337 may be cleanly mapped to it (similar to how microprocessors are "universal
338 IR's", allowing many source languages to be mapped to them). By providing
339 type information, LLVM can be used as the target of optimizations: for
340 example, through pointer analysis, it can be proven that a C automatic
341 variable is never accessed outside of the current function, allowing it to
342 be promoted to a simple SSA value instead of a memory location.</p>
346 <!-- _______________________________________________________________________ -->
347 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
349 <div class="doc_text">
351 <p>It is important to note that this document describes 'well formed' LLVM
352 assembly language. There is a difference between what the parser accepts and
353 what is considered 'well formed'. For example, the following instruction is
354 syntactically okay, but not well formed:</p>
356 <div class="doc_code">
358 %x = <a href="#i_add">add</a> i32 1, %x
362 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
363 LLVM infrastructure provides a verification pass that may be used to verify
364 that an LLVM module is well formed. This pass is automatically run by the
365 parser after parsing input assembly and by the optimizer before it outputs
366 bitcode. The violations pointed out by the verifier pass indicate bugs in
367 transformation passes or input to the parser.</p>
371 <!-- Describe the typesetting conventions here. -->
373 <!-- *********************************************************************** -->
374 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
375 <!-- *********************************************************************** -->
377 <div class="doc_text">
379 <p>LLVM identifiers come in two basic types: global and local. Global
380 identifiers (functions, global variables) begin with the <tt>'@'</tt>
381 character. Local identifiers (register names, types) begin with
382 the <tt>'%'</tt> character. Additionally, there are three different formats
383 for identifiers, for different purposes:</p>
386 <li>Named values are represented as a string of characters with their prefix.
387 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
388 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
389 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
390 other characters in their names can be surrounded with quotes. Special
391 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
392 ASCII code for the character in hexadecimal. In this way, any character
393 can be used in a name value, even quotes themselves.</li>
395 <li>Unnamed values are represented as an unsigned numeric value with their
396 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
398 <li>Constants, which are described in a <a href="#constants">section about
399 constants</a>, below.</li>
402 <p>LLVM requires that values start with a prefix for two reasons: Compilers
403 don't need to worry about name clashes with reserved words, and the set of
404 reserved words may be expanded in the future without penalty. Additionally,
405 unnamed identifiers allow a compiler to quickly come up with a temporary
406 variable without having to avoid symbol table conflicts.</p>
408 <p>Reserved words in LLVM are very similar to reserved words in other
409 languages. There are keywords for different opcodes
410 ('<tt><a href="#i_add">add</a></tt>',
411 '<tt><a href="#i_bitcast">bitcast</a></tt>',
412 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
413 ('<tt><a href="#t_void">void</a></tt>',
414 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
415 reserved words cannot conflict with variable names, because none of them
416 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
418 <p>Here is an example of LLVM code to multiply the integer variable
419 '<tt>%X</tt>' by 8:</p>
423 <div class="doc_code">
425 %result = <a href="#i_mul">mul</a> i32 %X, 8
429 <p>After strength reduction:</p>
431 <div class="doc_code">
433 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
437 <p>And the hard way:</p>
439 <div class="doc_code">
441 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
442 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
443 %result = <a href="#i_add">add</a> i32 %1, %1
447 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
448 lexical features of LLVM:</p>
451 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
454 <li>Unnamed temporaries are created when the result of a computation is not
455 assigned to a named value.</li>
457 <li>Unnamed temporaries are numbered sequentially</li>
460 <p>It also shows a convention that we follow in this document. When
461 demonstrating instructions, we will follow an instruction with a comment that
462 defines the type and name of value produced. Comments are shown in italic
467 <!-- *********************************************************************** -->
468 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
469 <!-- *********************************************************************** -->
471 <!-- ======================================================================= -->
472 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
475 <div class="doc_text">
477 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
478 of the input programs. Each module consists of functions, global variables,
479 and symbol table entries. Modules may be combined together with the LLVM
480 linker, which merges function (and global variable) definitions, resolves
481 forward declarations, and merges symbol table entries. Here is an example of
482 the "hello world" module:</p>
484 <div class="doc_code">
486 <i>; Declare the string constant as a global constant.</i>
487 <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>
489 <i>; External declaration of the puts function</i>
490 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
492 <i>; Definition of main function</i>
493 define i32 @main() { <i>; i32()* </i>
494 <i>; Convert [13 x i8]* to i8 *...</i>
495 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
497 <i>; Call puts function to write out the string to stdout.</i>
498 <a href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
499 <a href="#i_ret">ret</a> i32 0<br>}<br>
503 <p>This example is made up of a <a href="#globalvars">global variable</a> named
504 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function, and
505 a <a href="#functionstructure">function definition</a> for
508 <p>In general, a module is made up of a list of global values, where both
509 functions and global variables are global values. Global values are
510 represented by a pointer to a memory location (in this case, a pointer to an
511 array of char, and a pointer to a function), and have one of the
512 following <a href="#linkage">linkage types</a>.</p>
516 <!-- ======================================================================= -->
517 <div class="doc_subsection">
518 <a name="linkage">Linkage Types</a>
521 <div class="doc_text">
523 <p>All Global Variables and Functions have one of the following types of
527 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
528 <dd>Global values with private linkage are only directly accessible by objects
529 in the current module. In particular, linking code into a module with an
530 private global value may cause the private to be renamed as necessary to
531 avoid collisions. Because the symbol is private to the module, all
532 references can be updated. This doesn't show up in any symbol table in the
535 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
536 <dd>Similar to private, but the symbol is passed through the assembler and
537 removed by the linker after evaluation. Note that (unlike private
538 symbols) linker_private symbols are subject to coalescing by the linker:
539 weak symbols get merged and redefinitions are rejected. However, unlike
540 normal strong symbols, they are removed by the linker from the final
541 linked image (executable or dynamic library).</dd>
543 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
544 <dd>Similar to private, but the value shows as a local symbol
545 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
546 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
548 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
549 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
550 into the object file corresponding to the LLVM module. They exist to
551 allow inlining and other optimizations to take place given knowledge of
552 the definition of the global, which is known to be somewhere outside the
553 module. Globals with <tt>available_externally</tt> linkage are allowed to
554 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
555 This linkage type is only allowed on definitions, not declarations.</dd>
557 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
558 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
559 the same name when linkage occurs. This is typically used to implement
560 inline functions, templates, or other code which must be generated in each
561 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
562 allowed to be discarded.</dd>
564 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
565 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
566 <tt>linkonce</tt> linkage, except that unreferenced globals with
567 <tt>weak</tt> linkage may not be discarded. This is used for globals that
568 are declared "weak" in C source code.</dd>
570 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
571 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
572 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
574 Symbols with "<tt>common</tt>" linkage are merged in the same way as
575 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
576 <tt>common</tt> symbols may not have an explicit section,
577 must have a zero initializer, and may not be marked '<a
578 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
579 have common linkage.</dd>
582 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
583 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
584 pointer to array type. When two global variables with appending linkage
585 are linked together, the two global arrays are appended together. This is
586 the LLVM, typesafe, equivalent of having the system linker append together
587 "sections" with identical names when .o files are linked.</dd>
589 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
590 <dd>The semantics of this linkage follow the ELF object file model: the symbol
591 is weak until linked, if not linked, the symbol becomes null instead of
592 being an undefined reference.</dd>
594 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
595 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
596 <dd>Some languages allow differing globals to be merged, such as two functions
597 with different semantics. Other languages, such as <tt>C++</tt>, ensure
598 that only equivalent globals are ever merged (the "one definition rule" -
599 "ODR"). Such languages can use the <tt>linkonce_odr</tt>
600 and <tt>weak_odr</tt> linkage types to indicate that the global will only
601 be merged with equivalent globals. These linkage types are otherwise the
602 same as their non-<tt>odr</tt> versions.</dd>
604 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
605 <dd>If none of the above identifiers are used, the global is externally
606 visible, meaning that it participates in linkage and can be used to
607 resolve external symbol references.</dd>
610 <p>The next two types of linkage are targeted for Microsoft Windows platform
611 only. They are designed to support importing (exporting) symbols from (to)
612 DLLs (Dynamic Link Libraries).</p>
615 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
616 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
617 or variable via a global pointer to a pointer that is set up by the DLL
618 exporting the symbol. On Microsoft Windows targets, the pointer name is
619 formed by combining <code>__imp_</code> and the function or variable
622 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
623 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
624 pointer to a pointer in a DLL, so that it can be referenced with the
625 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
626 name is formed by combining <code>__imp_</code> and the function or
630 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
631 another module defined a "<tt>.LC0</tt>" variable and was linked with this
632 one, one of the two would be renamed, preventing a collision. Since
633 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
634 declarations), they are accessible outside of the current module.</p>
636 <p>It is illegal for a function <i>declaration</i> to have any linkage type
637 other than "externally visible", <tt>dllimport</tt>
638 or <tt>extern_weak</tt>.</p>
640 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
641 or <tt>weak_odr</tt> linkages.</p>
645 <!-- ======================================================================= -->
646 <div class="doc_subsection">
647 <a name="callingconv">Calling Conventions</a>
650 <div class="doc_text">
652 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
653 and <a href="#i_invoke">invokes</a> can all have an optional calling
654 convention specified for the call. The calling convention of any pair of
655 dynamic caller/callee must match, or the behavior of the program is
656 undefined. The following calling conventions are supported by LLVM, and more
657 may be added in the future:</p>
660 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
661 <dd>This calling convention (the default if no other calling convention is
662 specified) matches the target C calling conventions. This calling
663 convention supports varargs function calls and tolerates some mismatch in
664 the declared prototype and implemented declaration of the function (as
667 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
668 <dd>This calling convention attempts to make calls as fast as possible
669 (e.g. by passing things in registers). This calling convention allows the
670 target to use whatever tricks it wants to produce fast code for the
671 target, without having to conform to an externally specified ABI
672 (Application Binary Interface). Implementations of this convention should
673 allow arbitrary <a href="CodeGenerator.html#tailcallopt">tail call
674 optimization</a> to be supported. This calling convention does not
675 support varargs and requires the prototype of all callees to exactly match
676 the prototype of the function definition.</dd>
678 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
679 <dd>This calling convention attempts to make code in the caller as efficient
680 as possible under the assumption that the call is not commonly executed.
681 As such, these calls often preserve all registers so that the call does
682 not break any live ranges in the caller side. This calling convention
683 does not support varargs and requires the prototype of all callees to
684 exactly match the prototype of the function definition.</dd>
686 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
687 <dd>Any calling convention may be specified by number, allowing
688 target-specific calling conventions to be used. Target specific calling
689 conventions start at 64.</dd>
692 <p>More calling conventions can be added/defined on an as-needed basis, to
693 support Pascal conventions or any other well-known target-independent
698 <!-- ======================================================================= -->
699 <div class="doc_subsection">
700 <a name="visibility">Visibility Styles</a>
703 <div class="doc_text">
705 <p>All Global Variables and Functions have one of the following visibility
709 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
710 <dd>On targets that use the ELF object file format, default visibility means
711 that the declaration is visible to other modules and, in shared libraries,
712 means that the declared entity may be overridden. On Darwin, default
713 visibility means that the declaration is visible to other modules. Default
714 visibility corresponds to "external linkage" in the language.</dd>
716 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
717 <dd>Two declarations of an object with hidden visibility refer to the same
718 object if they are in the same shared object. Usually, hidden visibility
719 indicates that the symbol will not be placed into the dynamic symbol
720 table, so no other module (executable or shared library) can reference it
723 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
724 <dd>On ELF, protected visibility indicates that the symbol will be placed in
725 the dynamic symbol table, but that references within the defining module
726 will bind to the local symbol. That is, the symbol cannot be overridden by
732 <!-- ======================================================================= -->
733 <div class="doc_subsection">
734 <a name="namedtypes">Named Types</a>
737 <div class="doc_text">
739 <p>LLVM IR allows you to specify name aliases for certain types. This can make
740 it easier to read the IR and make the IR more condensed (particularly when
741 recursive types are involved). An example of a name specification is:</p>
743 <div class="doc_code">
745 %mytype = type { %mytype*, i32 }
749 <p>You may give a name to any <a href="#typesystem">type</a> except
750 "<a href="t_void">void</a>". Type name aliases may be used anywhere a type
751 is expected with the syntax "%mytype".</p>
753 <p>Note that type names are aliases for the structural type that they indicate,
754 and that you can therefore specify multiple names for the same type. This
755 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
756 uses structural typing, the name is not part of the type. When printing out
757 LLVM IR, the printer will pick <em>one name</em> to render all types of a
758 particular shape. This means that if you have code where two different
759 source types end up having the same LLVM type, that the dumper will sometimes
760 print the "wrong" or unexpected type. This is an important design point and
761 isn't going to change.</p>
765 <!-- ======================================================================= -->
766 <div class="doc_subsection">
767 <a name="globalvars">Global Variables</a>
770 <div class="doc_text">
772 <p>Global variables define regions of memory allocated at compilation time
773 instead of run-time. Global variables may optionally be initialized, may
774 have an explicit section to be placed in, and may have an optional explicit
775 alignment specified. A variable may be defined as "thread_local", which
776 means that it will not be shared by threads (each thread will have a
777 separated copy of the variable). A variable may be defined as a global
778 "constant," which indicates that the contents of the variable
779 will <b>never</b> be modified (enabling better optimization, allowing the
780 global data to be placed in the read-only section of an executable, etc).
781 Note that variables that need runtime initialization cannot be marked
782 "constant" as there is a store to the variable.</p>
784 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
785 constant, even if the final definition of the global is not. This capability
786 can be used to enable slightly better optimization of the program, but
787 requires the language definition to guarantee that optimizations based on the
788 'constantness' are valid for the translation units that do not include the
791 <p>As SSA values, global variables define pointer values that are in scope
792 (i.e. they dominate) all basic blocks in the program. Global variables
793 always define a pointer to their "content" type because they describe a
794 region of memory, and all memory objects in LLVM are accessed through
797 <p>A global variable may be declared to reside in a target-specific numbered
798 address space. For targets that support them, address spaces may affect how
799 optimizations are performed and/or what target instructions are used to
800 access the variable. The default address space is zero. The address space
801 qualifier must precede any other attributes.</p>
803 <p>LLVM allows an explicit section to be specified for globals. If the target
804 supports it, it will emit globals to the section specified.</p>
806 <p>An explicit alignment may be specified for a global. If not present, or if
807 the alignment is set to zero, the alignment of the global is set by the
808 target to whatever it feels convenient. If an explicit alignment is
809 specified, the global is forced to have at least that much alignment. All
810 alignments must be a power of 2.</p>
812 <p>For example, the following defines a global in a numbered address space with
813 an initializer, section, and alignment:</p>
815 <div class="doc_code">
817 @G = addrspace(5) constant float 1.0, section "foo", align 4
824 <!-- ======================================================================= -->
825 <div class="doc_subsection">
826 <a name="functionstructure">Functions</a>
829 <div class="doc_text">
831 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord, an
832 optional <a href="#linkage">linkage type</a>, an optional
833 <a href="#visibility">visibility style</a>, an optional
834 <a href="#callingconv">calling convention</a>, a return type, an optional
835 <a href="#paramattrs">parameter attribute</a> for the return type, a function
836 name, a (possibly empty) argument list (each with optional
837 <a href="#paramattrs">parameter attributes</a>), optional
838 <a href="#fnattrs">function attributes</a>, an optional section, an optional
839 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
840 curly brace, a list of basic blocks, and a closing curly brace.</p>
842 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
843 optional <a href="#linkage">linkage type</a>, an optional
844 <a href="#visibility">visibility style</a>, an optional
845 <a href="#callingconv">calling convention</a>, a return type, an optional
846 <a href="#paramattrs">parameter attribute</a> for the return type, a function
847 name, a possibly empty list of arguments, an optional alignment, and an
848 optional <a href="#gc">garbage collector name</a>.</p>
850 <p>A function definition contains a list of basic blocks, forming the CFG
851 (Control Flow Graph) for the function. Each basic block may optionally start
852 with a label (giving the basic block a symbol table entry), contains a list
853 of instructions, and ends with a <a href="#terminators">terminator</a>
854 instruction (such as a branch or function return).</p>
856 <p>The first basic block in a function is special in two ways: it is immediately
857 executed on entrance to the function, and it is not allowed to have
858 predecessor basic blocks (i.e. there can not be any branches to the entry
859 block of a function). Because the block can have no predecessors, it also
860 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
862 <p>LLVM allows an explicit section to be specified for functions. If the target
863 supports it, it will emit functions to the section specified.</p>
865 <p>An explicit alignment may be specified for a function. If not present, or if
866 the alignment is set to zero, the alignment of the function is set by the
867 target to whatever it feels convenient. If an explicit alignment is
868 specified, the function is forced to have at least that much alignment. All
869 alignments must be a power of 2.</p>
872 <div class="doc_code">
874 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
875 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
876 <ResultType> @<FunctionName> ([argument list])
877 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
878 [<a href="#gc">gc</a>] { ... }
884 <!-- ======================================================================= -->
885 <div class="doc_subsection">
886 <a name="aliasstructure">Aliases</a>
889 <div class="doc_text">
891 <p>Aliases act as "second name" for the aliasee value (which can be either
892 function, global variable, another alias or bitcast of global value). Aliases
893 may have an optional <a href="#linkage">linkage type</a>, and an
894 optional <a href="#visibility">visibility style</a>.</p>
897 <div class="doc_code">
899 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
905 <!-- ======================================================================= -->
906 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
908 <div class="doc_text">
910 <p>The return type and each parameter of a function type may have a set of
911 <i>parameter attributes</i> associated with them. Parameter attributes are
912 used to communicate additional information about the result or parameters of
913 a function. Parameter attributes are considered to be part of the function,
914 not of the function type, so functions with different parameter attributes
915 can have the same function type.</p>
917 <p>Parameter attributes are simple keywords that follow the type specified. If
918 multiple parameter attributes are needed, they are space separated. For
921 <div class="doc_code">
923 declare i32 @printf(i8* noalias nocapture, ...)
924 declare i32 @atoi(i8 zeroext)
925 declare signext i8 @returns_signed_char()
929 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
930 <tt>readonly</tt>) come immediately after the argument list.</p>
932 <p>Currently, only the following parameter attributes are defined:</p>
935 <dt><tt><b>zeroext</b></tt></dt>
936 <dd>This indicates to the code generator that the parameter or return value
937 should be zero-extended to a 32-bit value by the caller (for a parameter)
938 or the callee (for a return value).</dd>
940 <dt><tt><b>signext</b></tt></dt>
941 <dd>This indicates to the code generator that the parameter or return value
942 should be sign-extended to a 32-bit value by the caller (for a parameter)
943 or the callee (for a return value).</dd>
945 <dt><tt><b>inreg</b></tt></dt>
946 <dd>This indicates that this parameter or return value should be treated in a
947 special target-dependent fashion during while emitting code for a function
948 call or return (usually, by putting it in a register as opposed to memory,
949 though some targets use it to distinguish between two different kinds of
950 registers). Use of this attribute is target-specific.</dd>
952 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
953 <dd>This indicates that the pointer parameter should really be passed by value
954 to the function. The attribute implies that a hidden copy of the pointee
955 is made between the caller and the callee, so the callee is unable to
956 modify the value in the callee. This attribute is only valid on LLVM
957 pointer arguments. It is generally used to pass structs and arrays by
958 value, but is also valid on pointers to scalars. The copy is considered
959 to belong to the caller not the callee (for example,
960 <tt><a href="#readonly">readonly</a></tt> functions should not write to
961 <tt>byval</tt> parameters). This is not a valid attribute for return
962 values. The byval attribute also supports specifying an alignment with
963 the align attribute. This has a target-specific effect on the code
964 generator that usually indicates a desired alignment for the synthesized
967 <dt><tt><b>sret</b></tt></dt>
968 <dd>This indicates that the pointer parameter specifies the address of a
969 structure that is the return value of the function in the source program.
970 This pointer must be guaranteed by the caller to be valid: loads and
971 stores to the structure may be assumed by the callee to not to trap. This
972 may only be applied to the first parameter. This is not a valid attribute
973 for return values. </dd>
975 <dt><tt><b>noalias</b></tt></dt>
976 <dd>This indicates that the pointer does not alias any global or any other
977 parameter. The caller is responsible for ensuring that this is the
978 case. On a function return value, <tt>noalias</tt> additionally indicates
979 that the pointer does not alias any other pointers visible to the
980 caller. For further details, please see the discussion of the NoAlias
982 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
985 <dt><tt><b>nocapture</b></tt></dt>
986 <dd>This indicates that the callee does not make any copies of the pointer
987 that outlive the callee itself. This is not a valid attribute for return
990 <dt><tt><b>nest</b></tt></dt>
991 <dd>This indicates that the pointer parameter can be excised using the
992 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
993 attribute for return values.</dd>
998 <!-- ======================================================================= -->
999 <div class="doc_subsection">
1000 <a name="gc">Garbage Collector Names</a>
1003 <div class="doc_text">
1005 <p>Each function may specify a garbage collector name, which is simply a
1008 <div class="doc_code">
1010 define void @f() gc "name" { ... }
1014 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1015 collector which will cause the compiler to alter its output in order to
1016 support the named garbage collection algorithm.</p>
1020 <!-- ======================================================================= -->
1021 <div class="doc_subsection">
1022 <a name="fnattrs">Function Attributes</a>
1025 <div class="doc_text">
1027 <p>Function attributes are set to communicate additional information about a
1028 function. Function attributes are considered to be part of the function, not
1029 of the function type, so functions with different parameter attributes can
1030 have the same function type.</p>
1032 <p>Function attributes are simple keywords that follow the type specified. If
1033 multiple attributes are needed, they are space separated. For example:</p>
1035 <div class="doc_code">
1037 define void @f() noinline { ... }
1038 define void @f() alwaysinline { ... }
1039 define void @f() alwaysinline optsize { ... }
1040 define void @f() optsize { ... }
1045 <dt><tt><b>alwaysinline</b></tt></dt>
1046 <dd>This attribute indicates that the inliner should attempt to inline this
1047 function into callers whenever possible, ignoring any active inlining size
1048 threshold for this caller.</dd>
1050 <dt><tt><b>inlinehint</b></tt></dt>
1051 <dd>This attribute indicates that the source code contained a hint that inlining
1052 this function is desirable (such as the "inline" keyword in C/C++). It
1053 is just a hint; it imposes no requirements on the inliner.</dd>
1055 <dt><tt><b>noinline</b></tt></dt>
1056 <dd>This attribute indicates that the inliner should never inline this
1057 function in any situation. This attribute may not be used together with
1058 the <tt>alwaysinline</tt> attribute.</dd>
1060 <dt><tt><b>optsize</b></tt></dt>
1061 <dd>This attribute suggests that optimization passes and code generator passes
1062 make choices that keep the code size of this function low, and otherwise
1063 do optimizations specifically to reduce code size.</dd>
1065 <dt><tt><b>noreturn</b></tt></dt>
1066 <dd>This function attribute indicates that the function never returns
1067 normally. This produces undefined behavior at runtime if the function
1068 ever does dynamically return.</dd>
1070 <dt><tt><b>nounwind</b></tt></dt>
1071 <dd>This function attribute indicates that the function never returns with an
1072 unwind or exceptional control flow. If the function does unwind, its
1073 runtime behavior is undefined.</dd>
1075 <dt><tt><b>readnone</b></tt></dt>
1076 <dd>This attribute indicates that the function computes its result (or decides
1077 to unwind an exception) based strictly on its arguments, without
1078 dereferencing any pointer arguments or otherwise accessing any mutable
1079 state (e.g. memory, control registers, etc) visible to caller functions.
1080 It does not write through any pointer arguments
1081 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1082 changes any state visible to callers. This means that it cannot unwind
1083 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1084 could use the <tt>unwind</tt> instruction.</dd>
1086 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1087 <dd>This attribute indicates that the function does not write through any
1088 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1089 arguments) or otherwise modify any state (e.g. memory, control registers,
1090 etc) visible to caller functions. It may dereference pointer arguments
1091 and read state that may be set in the caller. A readonly function always
1092 returns the same value (or unwinds an exception identically) when called
1093 with the same set of arguments and global state. It cannot unwind an
1094 exception by calling the <tt>C++</tt> exception throwing methods, but may
1095 use the <tt>unwind</tt> instruction.</dd>
1097 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1098 <dd>This attribute indicates that the function should emit a stack smashing
1099 protector. It is in the form of a "canary"—a random value placed on
1100 the stack before the local variables that's checked upon return from the
1101 function to see if it has been overwritten. A heuristic is used to
1102 determine if a function needs stack protectors or not.<br>
1104 If a function that has an <tt>ssp</tt> attribute is inlined into a
1105 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1106 function will have an <tt>ssp</tt> attribute.</dd>
1108 <dt><tt><b>sspreq</b></tt></dt>
1109 <dd>This attribute indicates that the function should <em>always</em> emit a
1110 stack smashing protector. This overrides
1111 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1113 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1114 function that doesn't have an <tt>sspreq</tt> attribute or which has
1115 an <tt>ssp</tt> attribute, then the resulting function will have
1116 an <tt>sspreq</tt> attribute.</dd>
1118 <dt><tt><b>noredzone</b></tt></dt>
1119 <dd>This attribute indicates that the code generator should not use a red
1120 zone, even if the target-specific ABI normally permits it.</dd>
1122 <dt><tt><b>noimplicitfloat</b></tt></dt>
1123 <dd>This attributes disables implicit floating point instructions.</dd>
1125 <dt><tt><b>naked</b></tt></dt>
1126 <dd>This attribute disables prologue / epilogue emission for the function.
1127 This can have very system-specific consequences.</dd>
1132 <!-- ======================================================================= -->
1133 <div class="doc_subsection">
1134 <a name="moduleasm">Module-Level Inline Assembly</a>
1137 <div class="doc_text">
1139 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1140 the GCC "file scope inline asm" blocks. These blocks are internally
1141 concatenated by LLVM and treated as a single unit, but may be separated in
1142 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1144 <div class="doc_code">
1146 module asm "inline asm code goes here"
1147 module asm "more can go here"
1151 <p>The strings can contain any character by escaping non-printable characters.
1152 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1155 <p>The inline asm code is simply printed to the machine code .s file when
1156 assembly code is generated.</p>
1160 <!-- ======================================================================= -->
1161 <div class="doc_subsection">
1162 <a name="datalayout">Data Layout</a>
1165 <div class="doc_text">
1167 <p>A module may specify a target specific data layout string that specifies how
1168 data is to be laid out in memory. The syntax for the data layout is
1171 <div class="doc_code">
1173 target datalayout = "<i>layout specification</i>"
1177 <p>The <i>layout specification</i> consists of a list of specifications
1178 separated by the minus sign character ('-'). Each specification starts with
1179 a letter and may include other information after the letter to define some
1180 aspect of the data layout. The specifications accepted are as follows:</p>
1184 <dd>Specifies that the target lays out data in big-endian form. That is, the
1185 bits with the most significance have the lowest address location.</dd>
1188 <dd>Specifies that the target lays out data in little-endian form. That is,
1189 the bits with the least significance have the lowest address
1192 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1193 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1194 <i>preferred</i> alignments. All sizes are in bits. Specifying
1195 the <i>pref</i> alignment is optional. If omitted, the
1196 preceding <tt>:</tt> should be omitted too.</dd>
1198 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1199 <dd>This specifies the alignment for an integer type of a given bit
1200 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1202 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1203 <dd>This specifies the alignment for a vector type of a given bit
1206 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1207 <dd>This specifies the alignment for a floating point type of a given bit
1208 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1211 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1212 <dd>This specifies the alignment for an aggregate type of a given bit
1215 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1216 <dd>This specifies the alignment for a stack object of a given bit
1220 <p>When constructing the data layout for a given target, LLVM starts with a
1221 default set of specifications which are then (possibly) overriden by the
1222 specifications in the <tt>datalayout</tt> keyword. The default specifications
1223 are given in this list:</p>
1226 <li><tt>E</tt> - big endian</li>
1227 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1228 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1229 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1230 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1231 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1232 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1233 alignment of 64-bits</li>
1234 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1235 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1236 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1237 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1238 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1239 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1242 <p>When LLVM is determining the alignment for a given type, it uses the
1243 following rules:</p>
1246 <li>If the type sought is an exact match for one of the specifications, that
1247 specification is used.</li>
1249 <li>If no match is found, and the type sought is an integer type, then the
1250 smallest integer type that is larger than the bitwidth of the sought type
1251 is used. If none of the specifications are larger than the bitwidth then
1252 the the largest integer type is used. For example, given the default
1253 specifications above, the i7 type will use the alignment of i8 (next
1254 largest) while both i65 and i256 will use the alignment of i64 (largest
1257 <li>If no match is found, and the type sought is a vector type, then the
1258 largest vector type that is smaller than the sought vector type will be
1259 used as a fall back. This happens because <128 x double> can be
1260 implemented in terms of 64 <2 x double>, for example.</li>
1265 <!-- ======================================================================= -->
1266 <div class="doc_subsection">
1267 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1270 <div class="doc_text">
1272 <p>Any memory access must be done through a pointer value associated
1273 with an address range of the memory access, otherwise the behavior
1274 is undefined. Pointer values are associated with address ranges
1275 according to the following rules:</p>
1278 <li>A pointer value formed from a
1279 <tt><a href="#i_getelementptr">getelementptr</a></tt> instruction
1280 is associated with the addresses associated with the first operand
1281 of the <tt>getelementptr</tt>.</li>
1282 <li>An address of a global variable is associated with the address
1283 range of the variable's storage.</li>
1284 <li>The result value of an allocation instruction is associated with
1285 the address range of the allocated storage.</li>
1286 <li>A null pointer in the default address-space is associated with
1288 <li>A pointer value formed by an
1289 <tt><a href="#i_inttoptr">inttoptr</a></tt> is associated with all
1290 address ranges of all pointer values that contribute (directly or
1291 indirectly) to the computation of the pointer's value.</li>
1292 <li>The result value of a
1293 <tt><a href="#i_bitcast">bitcast</a></tt> is associated with all
1294 addresses associated with the operand of the <tt>bitcast</tt>.</li>
1295 <li>An integer constant other than zero or a pointer value returned
1296 from a function not defined within LLVM may be associated with address
1297 ranges allocated through mechanisms other than those provided by
1298 LLVM. Such ranges shall not overlap with any ranges of addresses
1299 allocated by mechanisms provided by LLVM.</li>
1302 <p>LLVM IR does not associate types with memory. The result type of a
1303 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1304 alignment of the memory from which to load, as well as the
1305 interpretation of the value. The first operand of a
1306 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1307 and alignment of the store.</p>
1309 <p>Consequently, type-based alias analysis, aka TBAA, aka
1310 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1311 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1312 additional information which specialized optimization passes may use
1313 to implement type-based alias analysis.</p>
1317 <!-- *********************************************************************** -->
1318 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1319 <!-- *********************************************************************** -->
1321 <div class="doc_text">
1323 <p>The LLVM type system is one of the most important features of the
1324 intermediate representation. Being typed enables a number of optimizations
1325 to be performed on the intermediate representation directly, without having
1326 to do extra analyses on the side before the transformation. A strong type
1327 system makes it easier to read the generated code and enables novel analyses
1328 and transformations that are not feasible to perform on normal three address
1329 code representations.</p>
1333 <!-- ======================================================================= -->
1334 <div class="doc_subsection"> <a name="t_classifications">Type
1335 Classifications</a> </div>
1337 <div class="doc_text">
1339 <p>The types fall into a few useful classifications:</p>
1341 <table border="1" cellspacing="0" cellpadding="4">
1343 <tr><th>Classification</th><th>Types</th></tr>
1345 <td><a href="#t_integer">integer</a></td>
1346 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1349 <td><a href="#t_floating">floating point</a></td>
1350 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1353 <td><a name="t_firstclass">first class</a></td>
1354 <td><a href="#t_integer">integer</a>,
1355 <a href="#t_floating">floating point</a>,
1356 <a href="#t_pointer">pointer</a>,
1357 <a href="#t_vector">vector</a>,
1358 <a href="#t_struct">structure</a>,
1359 <a href="#t_array">array</a>,
1360 <a href="#t_label">label</a>,
1361 <a href="#t_metadata">metadata</a>.
1365 <td><a href="#t_primitive">primitive</a></td>
1366 <td><a href="#t_label">label</a>,
1367 <a href="#t_void">void</a>,
1368 <a href="#t_floating">floating point</a>,
1369 <a href="#t_metadata">metadata</a>.</td>
1372 <td><a href="#t_derived">derived</a></td>
1373 <td><a href="#t_integer">integer</a>,
1374 <a href="#t_array">array</a>,
1375 <a href="#t_function">function</a>,
1376 <a href="#t_pointer">pointer</a>,
1377 <a href="#t_struct">structure</a>,
1378 <a href="#t_pstruct">packed structure</a>,
1379 <a href="#t_vector">vector</a>,
1380 <a href="#t_opaque">opaque</a>.
1386 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1387 important. Values of these types are the only ones which can be produced by
1392 <!-- ======================================================================= -->
1393 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1395 <div class="doc_text">
1397 <p>The primitive types are the fundamental building blocks of the LLVM
1402 <!-- _______________________________________________________________________ -->
1403 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1405 <div class="doc_text">
1408 <p>The integer type is a very simple type that simply specifies an arbitrary
1409 bit width for the integer type desired. Any bit width from 1 bit to
1410 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1417 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1421 <table class="layout">
1423 <td class="left"><tt>i1</tt></td>
1424 <td class="left">a single-bit integer.</td>
1427 <td class="left"><tt>i32</tt></td>
1428 <td class="left">a 32-bit integer.</td>
1431 <td class="left"><tt>i1942652</tt></td>
1432 <td class="left">a really big integer of over 1 million bits.</td>
1436 <p>Note that the code generator does not yet support large integer types to be
1437 used as function return types. The specific limit on how large a return type
1438 the code generator can currently handle is target-dependent; currently it's
1439 often 64 bits for 32-bit targets and 128 bits for 64-bit targets.</p>
1443 <!-- _______________________________________________________________________ -->
1444 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1446 <div class="doc_text">
1450 <tr><th>Type</th><th>Description</th></tr>
1451 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1452 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1453 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1454 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1455 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1461 <!-- _______________________________________________________________________ -->
1462 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1464 <div class="doc_text">
1467 <p>The void type does not represent any value and has no size.</p>
1476 <!-- _______________________________________________________________________ -->
1477 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1479 <div class="doc_text">
1482 <p>The label type represents code labels.</p>
1491 <!-- _______________________________________________________________________ -->
1492 <div class="doc_subsubsection"> <a name="t_metadata">Metadata Type</a> </div>
1494 <div class="doc_text">
1497 <p>The metadata type represents embedded metadata. No derived types may be
1498 created from metadata except for <a href="#t_function">function</a>
1509 <!-- ======================================================================= -->
1510 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1512 <div class="doc_text">
1514 <p>The real power in LLVM comes from the derived types in the system. This is
1515 what allows a programmer to represent arrays, functions, pointers, and other
1516 useful types. Each of these types contain one or more element types which
1517 may be a primitive type, or another derived type. For example, it is
1518 possible to have a two dimensional array, using an array as the element type
1519 of another array.</p>
1523 <!-- _______________________________________________________________________ -->
1524 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1526 <div class="doc_text">
1529 <p>The array type is a very simple derived type that arranges elements
1530 sequentially in memory. The array type requires a size (number of elements)
1531 and an underlying data type.</p>
1535 [<# elements> x <elementtype>]
1538 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1539 be any type with a size.</p>
1542 <table class="layout">
1544 <td class="left"><tt>[40 x i32]</tt></td>
1545 <td class="left">Array of 40 32-bit integer values.</td>
1548 <td class="left"><tt>[41 x i32]</tt></td>
1549 <td class="left">Array of 41 32-bit integer values.</td>
1552 <td class="left"><tt>[4 x i8]</tt></td>
1553 <td class="left">Array of 4 8-bit integer values.</td>
1556 <p>Here are some examples of multidimensional arrays:</p>
1557 <table class="layout">
1559 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1560 <td class="left">3x4 array of 32-bit integer values.</td>
1563 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1564 <td class="left">12x10 array of single precision floating point values.</td>
1567 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1568 <td class="left">2x3x4 array of 16-bit integer values.</td>
1572 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1573 length array. Normally, accesses past the end of an array are undefined in
1574 LLVM (e.g. it is illegal to access the 5th element of a 3 element array). As
1575 a special case, however, zero length arrays are recognized to be variable
1576 length. This allows implementation of 'pascal style arrays' with the LLVM
1577 type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1579 <p>Note that the code generator does not yet support large aggregate types to be
1580 used as function return types. The specific limit on how large an aggregate
1581 return type the code generator can currently handle is target-dependent, and
1582 also dependent on the aggregate element types.</p>
1586 <!-- _______________________________________________________________________ -->
1587 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1589 <div class="doc_text">
1592 <p>The function type can be thought of as a function signature. It consists of
1593 a return type and a list of formal parameter types. The return type of a
1594 function type is a scalar type, a void type, or a struct type. If the return
1595 type is a struct type then all struct elements must be of first class types,
1596 and the struct must have at least one element.</p>
1600 <returntype> (<parameter list>)
1603 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1604 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1605 which indicates that the function takes a variable number of arguments.
1606 Variable argument functions can access their arguments with
1607 the <a href="#int_varargs">variable argument handling intrinsic</a>
1608 functions. '<tt><returntype></tt>' is a any type except
1609 <a href="#t_label">label</a>.</p>
1612 <table class="layout">
1614 <td class="left"><tt>i32 (i32)</tt></td>
1615 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1617 </tr><tr class="layout">
1618 <td class="left"><tt>float (i16 signext, i32 *) *
1620 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1621 an <tt>i16</tt> that should be sign extended and a
1622 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1625 </tr><tr class="layout">
1626 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1627 <td class="left">A vararg function that takes at least one
1628 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1629 which returns an integer. This is the signature for <tt>printf</tt> in
1632 </tr><tr class="layout">
1633 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1634 <td class="left">A function taking an <tt>i32</tt>, returning a
1635 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
1642 <!-- _______________________________________________________________________ -->
1643 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1645 <div class="doc_text">
1648 <p>The structure type is used to represent a collection of data members together
1649 in memory. The packing of the field types is defined to match the ABI of the
1650 underlying processor. The elements of a structure may be any type that has a
1653 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1654 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1655 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1659 { <type list> }
1663 <table class="layout">
1665 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1666 <td class="left">A triple of three <tt>i32</tt> values</td>
1667 </tr><tr class="layout">
1668 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1669 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1670 second element is a <a href="#t_pointer">pointer</a> to a
1671 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1672 an <tt>i32</tt>.</td>
1676 <p>Note that the code generator does not yet support large aggregate types to be
1677 used as function return types. The specific limit on how large an aggregate
1678 return type the code generator can currently handle is target-dependent, and
1679 also dependent on the aggregate element types.</p>
1683 <!-- _______________________________________________________________________ -->
1684 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1687 <div class="doc_text">
1690 <p>The packed structure type is used to represent a collection of data members
1691 together in memory. There is no padding between fields. Further, the
1692 alignment of a packed structure is 1 byte. The elements of a packed
1693 structure may be any type that has a size.</p>
1695 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1696 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1697 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1701 < { <type list> } >
1705 <table class="layout">
1707 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1708 <td class="left">A triple of three <tt>i32</tt> values</td>
1709 </tr><tr class="layout">
1711 <tt>< { float, i32 (i32)* } ></tt></td>
1712 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1713 second element is a <a href="#t_pointer">pointer</a> to a
1714 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1715 an <tt>i32</tt>.</td>
1721 <!-- _______________________________________________________________________ -->
1722 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1724 <div class="doc_text">
1727 <p>As in many languages, the pointer type represents a pointer or reference to
1728 another object, which must live in memory. Pointer types may have an optional
1729 address space attribute defining the target-specific numbered address space
1730 where the pointed-to object resides. The default address space is zero.</p>
1732 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
1733 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1741 <table class="layout">
1743 <td class="left"><tt>[4 x i32]*</tt></td>
1744 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1745 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1748 <td class="left"><tt>i32 (i32 *) *</tt></td>
1749 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1750 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1754 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1755 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1756 that resides in address space #5.</td>
1762 <!-- _______________________________________________________________________ -->
1763 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1765 <div class="doc_text">
1768 <p>A vector type is a simple derived type that represents a vector of elements.
1769 Vector types are used when multiple primitive data are operated in parallel
1770 using a single instruction (SIMD). A vector type requires a size (number of
1771 elements) and an underlying primitive data type. Vectors must have a power
1772 of two length (1, 2, 4, 8, 16 ...). Vector types are considered
1773 <a href="#t_firstclass">first class</a>.</p>
1777 < <# elements> x <elementtype> >
1780 <p>The number of elements is a constant integer value; elementtype may be any
1781 integer or floating point type.</p>
1784 <table class="layout">
1786 <td class="left"><tt><4 x i32></tt></td>
1787 <td class="left">Vector of 4 32-bit integer values.</td>
1790 <td class="left"><tt><8 x float></tt></td>
1791 <td class="left">Vector of 8 32-bit floating-point values.</td>
1794 <td class="left"><tt><2 x i64></tt></td>
1795 <td class="left">Vector of 2 64-bit integer values.</td>
1799 <p>Note that the code generator does not yet support large vector types to be
1800 used as function return types. The specific limit on how large a vector
1801 return type codegen can currently handle is target-dependent; currently it's
1802 often a few times longer than a hardware vector register.</p>
1806 <!-- _______________________________________________________________________ -->
1807 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1808 <div class="doc_text">
1811 <p>Opaque types are used to represent unknown types in the system. This
1812 corresponds (for example) to the C notion of a forward declared structure
1813 type. In LLVM, opaque types can eventually be resolved to any type (not just
1814 a structure type).</p>
1822 <table class="layout">
1824 <td class="left"><tt>opaque</tt></td>
1825 <td class="left">An opaque type.</td>
1831 <!-- ======================================================================= -->
1832 <div class="doc_subsection">
1833 <a name="t_uprefs">Type Up-references</a>
1836 <div class="doc_text">
1839 <p>An "up reference" allows you to refer to a lexically enclosing type without
1840 requiring it to have a name. For instance, a structure declaration may
1841 contain a pointer to any of the types it is lexically a member of. Example
1842 of up references (with their equivalent as named type declarations)
1846 { \2 * } %x = type { %x* }
1847 { \2 }* %y = type { %y }*
1851 <p>An up reference is needed by the asmprinter for printing out cyclic types
1852 when there is no declared name for a type in the cycle. Because the
1853 asmprinter does not want to print out an infinite type string, it needs a
1854 syntax to handle recursive types that have no names (all names are optional
1862 <p>The level is the count of the lexical type that is being referred to.</p>
1865 <table class="layout">
1867 <td class="left"><tt>\1*</tt></td>
1868 <td class="left">Self-referential pointer.</td>
1871 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1872 <td class="left">Recursive structure where the upref refers to the out-most
1879 <!-- *********************************************************************** -->
1880 <div class="doc_section"> <a name="constants">Constants</a> </div>
1881 <!-- *********************************************************************** -->
1883 <div class="doc_text">
1885 <p>LLVM has several different basic types of constants. This section describes
1886 them all and their syntax.</p>
1890 <!-- ======================================================================= -->
1891 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1893 <div class="doc_text">
1896 <dt><b>Boolean constants</b></dt>
1897 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1898 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
1900 <dt><b>Integer constants</b></dt>
1901 <dd>Standard integers (such as '4') are constants of
1902 the <a href="#t_integer">integer</a> type. Negative numbers may be used
1903 with integer types.</dd>
1905 <dt><b>Floating point constants</b></dt>
1906 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1907 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1908 notation (see below). The assembler requires the exact decimal value of a
1909 floating-point constant. For example, the assembler accepts 1.25 but
1910 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1911 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1913 <dt><b>Null pointer constants</b></dt>
1914 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1915 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1918 <p>The one non-intuitive notation for constants is the hexadecimal form of
1919 floating point constants. For example, the form '<tt>double
1920 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
1921 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
1922 constants are required (and the only time that they are generated by the
1923 disassembler) is when a floating point constant must be emitted but it cannot
1924 be represented as a decimal floating point number in a reasonable number of
1925 digits. For example, NaN's, infinities, and other special values are
1926 represented in their IEEE hexadecimal format so that assembly and disassembly
1927 do not cause any bits to change in the constants.</p>
1929 <p>When using the hexadecimal form, constants of types float and double are
1930 represented using the 16-digit form shown above (which matches the IEEE754
1931 representation for double); float values must, however, be exactly
1932 representable as IEE754 single precision. Hexadecimal format is always used
1933 for long double, and there are three forms of long double. The 80-bit format
1934 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
1935 The 128-bit format used by PowerPC (two adjacent doubles) is represented
1936 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
1937 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
1938 currently supported target uses this format. Long doubles will only work if
1939 they match the long double format on your target. All hexadecimal formats
1940 are big-endian (sign bit at the left).</p>
1944 <!-- ======================================================================= -->
1945 <div class="doc_subsection">
1946 <a name="aggregateconstants"></a> <!-- old anchor -->
1947 <a name="complexconstants">Complex Constants</a>
1950 <div class="doc_text">
1952 <p>Complex constants are a (potentially recursive) combination of simple
1953 constants and smaller complex constants.</p>
1956 <dt><b>Structure constants</b></dt>
1957 <dd>Structure constants are represented with notation similar to structure
1958 type definitions (a comma separated list of elements, surrounded by braces
1959 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1960 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
1961 Structure constants must have <a href="#t_struct">structure type</a>, and
1962 the number and types of elements must match those specified by the
1965 <dt><b>Array constants</b></dt>
1966 <dd>Array constants are represented with notation similar to array type
1967 definitions (a comma separated list of elements, surrounded by square
1968 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
1969 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
1970 the number and types of elements must match those specified by the
1973 <dt><b>Vector constants</b></dt>
1974 <dd>Vector constants are represented with notation similar to vector type
1975 definitions (a comma separated list of elements, surrounded by
1976 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
1977 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
1978 have <a href="#t_vector">vector type</a>, and the number and types of
1979 elements must match those specified by the type.</dd>
1981 <dt><b>Zero initialization</b></dt>
1982 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1983 value to zero of <em>any</em> type, including scalar and aggregate types.
1984 This is often used to avoid having to print large zero initializers
1985 (e.g. for large arrays) and is always exactly equivalent to using explicit
1986 zero initializers.</dd>
1988 <dt><b>Metadata node</b></dt>
1989 <dd>A metadata node is a structure-like constant with
1990 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
1991 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
1992 be interpreted as part of the instruction stream, metadata is a place to
1993 attach additional information such as debug info.</dd>
1998 <!-- ======================================================================= -->
1999 <div class="doc_subsection">
2000 <a name="globalconstants">Global Variable and Function Addresses</a>
2003 <div class="doc_text">
2005 <p>The addresses of <a href="#globalvars">global variables</a>
2006 and <a href="#functionstructure">functions</a> are always implicitly valid
2007 (link-time) constants. These constants are explicitly referenced when
2008 the <a href="#identifiers">identifier for the global</a> is used and always
2009 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2010 legal LLVM file:</p>
2012 <div class="doc_code">
2016 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2022 <!-- ======================================================================= -->
2023 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
2024 <div class="doc_text">
2026 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2027 indicates that the user of the value may receive an unspecified bit-pattern.
2028 Undefined values may be of any type (other than label or void) and be used
2029 anywhere a constant is permitted.</p>
2031 <p>Undefined values are useful because they indicate to the compiler that the
2032 program is well defined no matter what value is used. This gives the
2033 compiler more freedom to optimize. Here are some examples of (potentially
2034 surprising) transformations that are valid (in pseudo IR):</p>
2037 <div class="doc_code">
2049 <p>This is safe because all of the output bits are affected by the undef bits.
2050 Any output bit can have a zero or one depending on the input bits.</p>
2052 <div class="doc_code">
2065 <p>These logical operations have bits that are not always affected by the input.
2066 For example, if "%X" has a zero bit, then the output of the 'and' operation will
2067 always be a zero, no matter what the corresponding bit from the undef is. As
2068 such, it is unsafe to optimize or assume that the result of the and is undef.
2069 However, it is safe to assume that all bits of the undef could be 0, and
2070 optimize the and to 0. Likewise, it is safe to assume that all the bits of
2071 the undef operand to the or could be set, allowing the or to be folded to
2074 <div class="doc_code">
2076 %A = select undef, %X, %Y
2077 %B = select undef, 42, %Y
2078 %C = select %X, %Y, undef
2090 <p>This set of examples show that undefined select (and conditional branch)
2091 conditions can go "either way" but they have to come from one of the two
2092 operands. In the %A example, if %X and %Y were both known to have a clear low
2093 bit, then %A would have to have a cleared low bit. However, in the %C example,
2094 the optimizer is allowed to assume that the undef operand could be the same as
2095 %Y, allowing the whole select to be eliminated.</p>
2098 <div class="doc_code">
2100 %A = xor undef, undef
2119 <p>This example points out that two undef operands are not necessarily the same.
2120 This can be surprising to people (and also matches C semantics) where they
2121 assume that "X^X" is always zero, even if X is undef. This isn't true for a
2122 number of reasons, but the short answer is that an undef "variable" can
2123 arbitrarily change its value over its "live range". This is true because the
2124 "variable" doesn't actually <em>have a live range</em>. Instead, the value is
2125 logically read from arbitrary registers that happen to be around when needed,
2126 so the value is not necessarily consistent over time. In fact, %A and %C need
2127 to have the same semantics or the core LLVM "replace all uses with" concept
2130 <div class="doc_code">
2140 <p>These examples show the crucial difference between an <em>undefined
2141 value</em> and <em>undefined behavior</em>. An undefined value (like undef) is
2142 allowed to have an arbitrary bit-pattern. This means that the %A operation
2143 can be constant folded to undef because the undef could be an SNaN, and fdiv is
2144 not (currently) defined on SNaN's. However, in the second example, we can make
2145 a more aggressive assumption: because the undef is allowed to be an arbitrary
2146 value, we are allowed to assume that it could be zero. Since a divide by zero
2147 has <em>undefined behavior</em>, we are allowed to assume that the operation
2148 does not execute at all. This allows us to delete the divide and all code after
2149 it: since the undefined operation "can't happen", the optimizer can assume that
2150 it occurs in dead code.
2153 <div class="doc_code">
2155 a: store undef -> %X
2156 b: store %X -> undef
2163 <p>These examples reiterate the fdiv example: a store "of" an undefined value
2164 can be assumed to not have any effect: we can assume that the value is
2165 overwritten with bits that happen to match what was already there. However, a
2166 store "to" an undefined location could clobber arbitrary memory, therefore, it
2167 has undefined behavior.</p>
2171 <!-- ======================================================================= -->
2172 <div class="doc_subsection"><a name="blockaddress">Addresses of Basic
2174 <div class="doc_text">
2176 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2178 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2179 basic block in the specified function, and always has an i8* type. Taking
2180 the address of the entry block is illegal.</p>
2182 <p>This value only has defined behavior when used as an operand to the
2183 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction or for comparisons
2184 against null. Pointer equality tests between labels addresses is undefined
2185 behavior - though, again, comparison against null is ok, and no label is
2186 equal to the null pointer. This may also be passed around as an opaque
2187 pointer sized value as long as the bits are not inspected. This allows
2188 <tt>ptrtoint</tt> and arithmetic to be performed on these values so long as
2189 the original value is reconstituted before the <tt>indirectbr</tt>.</p>
2191 <p>Finally, some targets may provide defined semantics when
2192 using the value as the operand to an inline assembly, but that is target
2199 <!-- ======================================================================= -->
2200 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
2203 <div class="doc_text">
2205 <p>Constant expressions are used to allow expressions involving other constants
2206 to be used as constants. Constant expressions may be of
2207 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2208 operation that does not have side effects (e.g. load and call are not
2209 supported). The following is the syntax for constant expressions:</p>
2212 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
2213 <dd>Truncate a constant to another type. The bit size of CST must be larger
2214 than the bit size of TYPE. Both types must be integers.</dd>
2216 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
2217 <dd>Zero extend a constant to another type. The bit size of CST must be
2218 smaller or equal to the bit size of TYPE. Both types must be
2221 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
2222 <dd>Sign extend a constant to another type. The bit size of CST must be
2223 smaller or equal to the bit size of TYPE. Both types must be
2226 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
2227 <dd>Truncate a floating point constant to another floating point type. The
2228 size of CST must be larger than the size of TYPE. Both types must be
2229 floating point.</dd>
2231 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
2232 <dd>Floating point extend a constant to another type. The size of CST must be
2233 smaller or equal to the size of TYPE. Both types must be floating
2236 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
2237 <dd>Convert a floating point constant to the corresponding unsigned integer
2238 constant. TYPE must be a scalar or vector integer type. CST must be of
2239 scalar or vector floating point type. Both CST and TYPE must be scalars,
2240 or vectors of the same number of elements. If the value won't fit in the
2241 integer type, the results are undefined.</dd>
2243 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
2244 <dd>Convert a floating point constant to the corresponding signed integer
2245 constant. TYPE must be a scalar or vector integer type. CST must be of
2246 scalar or vector floating point type. Both CST and TYPE must be scalars,
2247 or vectors of the same number of elements. If the value won't fit in the
2248 integer type, the results are undefined.</dd>
2250 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
2251 <dd>Convert an unsigned integer constant to the corresponding floating point
2252 constant. TYPE must be a scalar or vector floating point type. CST must be
2253 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2254 vectors of the same number of elements. If the value won't fit in the
2255 floating point type, the results are undefined.</dd>
2257 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
2258 <dd>Convert a signed integer constant to the corresponding floating point
2259 constant. TYPE must be a scalar or vector floating point type. CST must be
2260 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2261 vectors of the same number of elements. If the value won't fit in the
2262 floating point type, the results are undefined.</dd>
2264 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
2265 <dd>Convert a pointer typed constant to the corresponding integer constant
2266 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2267 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2268 make it fit in <tt>TYPE</tt>.</dd>
2270 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
2271 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2272 type. CST must be of integer type. The CST value is zero extended,
2273 truncated, or unchanged to make it fit in a pointer size. This one is
2274 <i>really</i> dangerous!</dd>
2276 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
2277 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2278 are the same as those for the <a href="#i_bitcast">bitcast
2279 instruction</a>.</dd>
2281 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2282 <dt><b><tt>getelementptr inbounds ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2283 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2284 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2285 instruction, the index list may have zero or more indexes, which are
2286 required to make sense for the type of "CSTPTR".</dd>
2288 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
2289 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2291 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
2292 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2294 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
2295 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2297 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2298 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2301 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2302 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2305 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2306 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2309 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2310 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2311 be any of the <a href="#binaryops">binary</a>
2312 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2313 on operands are the same as those for the corresponding instruction
2314 (e.g. no bitwise operations on floating point values are allowed).</dd>
2319 <!-- ======================================================================= -->
2320 <div class="doc_subsection"><a name="metadata">Embedded Metadata</a>
2323 <div class="doc_text">
2325 <p>Embedded metadata provides a way to attach arbitrary data to the instruction
2326 stream without affecting the behaviour of the program. There are two
2327 metadata primitives, strings and nodes. All metadata has the
2328 <tt>metadata</tt> type and is identified in syntax by a preceding exclamation
2329 point ('<tt>!</tt>').</p>
2331 <p>A metadata string is a string surrounded by double quotes. It can contain
2332 any character by escaping non-printable characters with "\xx" where "xx" is
2333 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2335 <p>Metadata nodes are represented with notation similar to structure constants
2336 (a comma separated list of elements, surrounded by braces and preceded by an
2337 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2340 <p>A metadata node will attempt to track changes to the values it holds. In the
2341 event that a value is deleted, it will be replaced with a typeless
2342 "<tt>null</tt>", such as "<tt>metadata !{null, i32 10}</tt>".</p>
2344 <p>Optimizations may rely on metadata to provide additional information about
2345 the program that isn't available in the instructions, or that isn't easily
2346 computable. Similarly, the code generator may expect a certain metadata
2347 format to be used to express debugging information.</p>
2351 <!-- *********************************************************************** -->
2352 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2353 <!-- *********************************************************************** -->
2355 <!-- ======================================================================= -->
2356 <div class="doc_subsection">
2357 <a name="inlineasm">Inline Assembler Expressions</a>
2360 <div class="doc_text">
2362 <p>LLVM supports inline assembler expressions (as opposed
2363 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2364 a special value. This value represents the inline assembler as a string
2365 (containing the instructions to emit), a list of operand constraints (stored
2366 as a string), a flag that indicates whether or not the inline asm
2367 expression has side effects, and a flag indicating whether the function
2368 containing the asm needs to align its stack conservatively. An example
2369 inline assembler expression is:</p>
2371 <div class="doc_code">
2373 i32 (i32) asm "bswap $0", "=r,r"
2377 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2378 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2381 <div class="doc_code">
2383 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2387 <p>Inline asms with side effects not visible in the constraint list must be
2388 marked as having side effects. This is done through the use of the
2389 '<tt>sideeffect</tt>' keyword, like so:</p>
2391 <div class="doc_code">
2393 call void asm sideeffect "eieio", ""()
2397 <p>In some cases inline asms will contain code that will not work unless the
2398 stack is aligned in some way, such as calls or SSE instructions on x86,
2399 yet will not contain code that does that alignment within the asm.
2400 The compiler should make conservative assumptions about what the asm might
2401 contain and should generate its usual stack alignment code in the prologue
2402 if the '<tt>alignstack</tt>' keyword is present:</p>
2404 <div class="doc_code">
2406 call void asm alignstack "eieio", ""()
2410 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2413 <p>TODO: The format of the asm and constraints string still need to be
2414 documented here. Constraints on what can be done (e.g. duplication, moving,
2415 etc need to be documented). This is probably best done by reference to
2416 another document that covers inline asm from a holistic perspective.</p>
2421 <!-- *********************************************************************** -->
2422 <div class="doc_section">
2423 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2425 <!-- *********************************************************************** -->
2427 <p>LLVM has a number of "magic" global variables that contain data that affect
2428 code generation or other IR semantics. These are documented here. All globals
2429 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2430 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2433 <!-- ======================================================================= -->
2434 <div class="doc_subsection">
2435 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2438 <div class="doc_text">
2440 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2441 href="#linkage_appending">appending linkage</a>. This array contains a list of
2442 pointers to global variables and functions which may optionally have a pointer
2443 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2449 @llvm.used = appending global [2 x i8*] [
2451 i8* bitcast (i32* @Y to i8*)
2452 ], section "llvm.metadata"
2455 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2456 compiler, assembler, and linker are required to treat the symbol as if there is
2457 a reference to the global that it cannot see. For example, if a variable has
2458 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2459 list, it cannot be deleted. This is commonly used to represent references from
2460 inline asms and other things the compiler cannot "see", and corresponds to
2461 "attribute((used))" in GNU C.</p>
2463 <p>On some targets, the code generator must emit a directive to the assembler or
2464 object file to prevent the assembler and linker from molesting the symbol.</p>
2468 <!-- ======================================================================= -->
2469 <div class="doc_subsection">
2470 <a name="intg_compiler_used">The '<tt>llvm.compiler.used</tt>' Global Variable</a>
2473 <div class="doc_text">
2475 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2476 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2477 touching the symbol. On targets that support it, this allows an intelligent
2478 linker to optimize references to the symbol without being impeded as it would be
2479 by <tt>@llvm.used</tt>.</p>
2481 <p>This is a rare construct that should only be used in rare circumstances, and
2482 should not be exposed to source languages.</p>
2486 <!-- ======================================================================= -->
2487 <div class="doc_subsection">
2488 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2491 <div class="doc_text">
2493 <p>TODO: Describe this.</p>
2497 <!-- ======================================================================= -->
2498 <div class="doc_subsection">
2499 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2502 <div class="doc_text">
2504 <p>TODO: Describe this.</p>
2509 <!-- *********************************************************************** -->
2510 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2511 <!-- *********************************************************************** -->
2513 <div class="doc_text">
2515 <p>The LLVM instruction set consists of several different classifications of
2516 instructions: <a href="#terminators">terminator
2517 instructions</a>, <a href="#binaryops">binary instructions</a>,
2518 <a href="#bitwiseops">bitwise binary instructions</a>,
2519 <a href="#memoryops">memory instructions</a>, and
2520 <a href="#otherops">other instructions</a>.</p>
2524 <!-- ======================================================================= -->
2525 <div class="doc_subsection"> <a name="terminators">Terminator
2526 Instructions</a> </div>
2528 <div class="doc_text">
2530 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
2531 in a program ends with a "Terminator" instruction, which indicates which
2532 block should be executed after the current block is finished. These
2533 terminator instructions typically yield a '<tt>void</tt>' value: they produce
2534 control flow, not values (the one exception being the
2535 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2537 <p>There are six different terminator instructions: the
2538 '<a href="#i_ret"><tt>ret</tt></a>' instruction, the
2539 '<a href="#i_br"><tt>br</tt></a>' instruction, the
2540 '<a href="#i_switch"><tt>switch</tt></a>' instruction, the
2541 '<a href="#i_indirectbr">'<tt>indirectbr</tt></a>' Instruction, the
2542 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the
2543 '<a href="#i_unwind"><tt>unwind</tt></a>' instruction, and the
2544 '<a href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2548 <!-- _______________________________________________________________________ -->
2549 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2550 Instruction</a> </div>
2552 <div class="doc_text">
2556 ret <type> <value> <i>; Return a value from a non-void function</i>
2557 ret void <i>; Return from void function</i>
2561 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
2562 a value) from a function back to the caller.</p>
2564 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
2565 value and then causes control flow, and one that just causes control flow to
2569 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
2570 return value. The type of the return value must be a
2571 '<a href="#t_firstclass">first class</a>' type.</p>
2573 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
2574 non-void return type and contains a '<tt>ret</tt>' instruction with no return
2575 value or a return value with a type that does not match its type, or if it
2576 has a void return type and contains a '<tt>ret</tt>' instruction with a
2580 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
2581 the calling function's context. If the caller is a
2582 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
2583 instruction after the call. If the caller was an
2584 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
2585 the beginning of the "normal" destination block. If the instruction returns
2586 a value, that value shall set the call or invoke instruction's return
2591 ret i32 5 <i>; Return an integer value of 5</i>
2592 ret void <i>; Return from a void function</i>
2593 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2596 <p>Note that the code generator does not yet fully support large
2597 return values. The specific sizes that are currently supported are
2598 dependent on the target. For integers, on 32-bit targets the limit
2599 is often 64 bits, and on 64-bit targets the limit is often 128 bits.
2600 For aggregate types, the current limits are dependent on the element
2601 types; for example targets are often limited to 2 total integer
2602 elements and 2 total floating-point elements.</p>
2605 <!-- _______________________________________________________________________ -->
2606 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2608 <div class="doc_text">
2612 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2616 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
2617 different basic block in the current function. There are two forms of this
2618 instruction, corresponding to a conditional branch and an unconditional
2622 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
2623 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
2624 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
2628 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2629 argument is evaluated. If the value is <tt>true</tt>, control flows to the
2630 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2631 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2636 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
2637 br i1 %cond, label %IfEqual, label %IfUnequal
2639 <a href="#i_ret">ret</a> i32 1
2641 <a href="#i_ret">ret</a> i32 0
2646 <!-- _______________________________________________________________________ -->
2647 <div class="doc_subsubsection">
2648 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2651 <div class="doc_text">
2655 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2659 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2660 several different places. It is a generalization of the '<tt>br</tt>'
2661 instruction, allowing a branch to occur to one of many possible
2665 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2666 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
2667 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
2668 The table is not allowed to contain duplicate constant entries.</p>
2671 <p>The <tt>switch</tt> instruction specifies a table of values and
2672 destinations. When the '<tt>switch</tt>' instruction is executed, this table
2673 is searched for the given value. If the value is found, control flow is
2674 transferred to the corresponding destination; otherwise, control flow is
2675 transferred to the default destination.</p>
2677 <h5>Implementation:</h5>
2678 <p>Depending on properties of the target machine and the particular
2679 <tt>switch</tt> instruction, this instruction may be code generated in
2680 different ways. For example, it could be generated as a series of chained
2681 conditional branches or with a lookup table.</p>
2685 <i>; Emulate a conditional br instruction</i>
2686 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2687 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2689 <i>; Emulate an unconditional br instruction</i>
2690 switch i32 0, label %dest [ ]
2692 <i>; Implement a jump table:</i>
2693 switch i32 %val, label %otherwise [ i32 0, label %onzero
2695 i32 2, label %ontwo ]
2701 <!-- _______________________________________________________________________ -->
2702 <div class="doc_subsubsection">
2703 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
2706 <div class="doc_text">
2710 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
2715 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
2716 within the current function, whose address is specified by
2717 "<tt>address</tt>". Address must be derived from a <a
2718 href="#blockaddress">blockaddress</a> constant.</p>
2722 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
2723 rest of the arguments indicate the full set of possible destinations that the
2724 address may point to. Blocks are allowed to occur multiple times in the
2725 destination list, though this isn't particularly useful.</p>
2727 <p>This destination list is required so that dataflow analysis has an accurate
2728 understanding of the CFG.</p>
2732 <p>Control transfers to the block specified in the address argument. All
2733 possible destination blocks must be listed in the label list, otherwise this
2734 instruction has undefined behavior. This implies that jumps to labels
2735 defined in other functions have undefined behavior as well.</p>
2737 <h5>Implementation:</h5>
2739 <p>This is typically implemented with a jump through a register.</p>
2743 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
2749 <!-- _______________________________________________________________________ -->
2750 <div class="doc_subsubsection">
2751 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2754 <div class="doc_text">
2758 <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>]
2759 to label <normal label> unwind label <exception label>
2763 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2764 function, with the possibility of control flow transfer to either the
2765 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
2766 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
2767 control flow will return to the "normal" label. If the callee (or any
2768 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
2769 instruction, control is interrupted and continued at the dynamically nearest
2770 "exception" label.</p>
2773 <p>This instruction requires several arguments:</p>
2776 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
2777 convention</a> the call should use. If none is specified, the call
2778 defaults to using C calling conventions.</li>
2780 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2781 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
2782 '<tt>inreg</tt>' attributes are valid here.</li>
2784 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2785 function value being invoked. In most cases, this is a direct function
2786 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
2787 off an arbitrary pointer to function value.</li>
2789 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2790 function to be invoked. </li>
2792 <li>'<tt>function args</tt>': argument list whose types match the function
2793 signature argument types. If the function signature indicates the
2794 function accepts a variable number of arguments, the extra arguments can
2797 <li>'<tt>normal label</tt>': the label reached when the called function
2798 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2800 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2801 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2803 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2804 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2805 '<tt>readnone</tt>' attributes are valid here.</li>
2809 <p>This instruction is designed to operate as a standard
2810 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
2811 primary difference is that it establishes an association with a label, which
2812 is used by the runtime library to unwind the stack.</p>
2814 <p>This instruction is used in languages with destructors to ensure that proper
2815 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2816 exception. Additionally, this is important for implementation of
2817 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2819 <p>For the purposes of the SSA form, the definition of the value returned by the
2820 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
2821 block to the "normal" label. If the callee unwinds then no return value is
2826 %retval = invoke i32 @Test(i32 15) to label %Continue
2827 unwind label %TestCleanup <i>; {i32}:retval set</i>
2828 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2829 unwind label %TestCleanup <i>; {i32}:retval set</i>
2834 <!-- _______________________________________________________________________ -->
2836 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2837 Instruction</a> </div>
2839 <div class="doc_text">
2847 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2848 at the first callee in the dynamic call stack which used
2849 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
2850 This is primarily used to implement exception handling.</p>
2853 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2854 immediately halt. The dynamic call stack is then searched for the
2855 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
2856 Once found, execution continues at the "exceptional" destination block
2857 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
2858 instruction in the dynamic call chain, undefined behavior results.</p>
2862 <!-- _______________________________________________________________________ -->
2864 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2865 Instruction</a> </div>
2867 <div class="doc_text">
2875 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2876 instruction is used to inform the optimizer that a particular portion of the
2877 code is not reachable. This can be used to indicate that the code after a
2878 no-return function cannot be reached, and other facts.</p>
2881 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2885 <!-- ======================================================================= -->
2886 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2888 <div class="doc_text">
2890 <p>Binary operators are used to do most of the computation in a program. They
2891 require two operands of the same type, execute an operation on them, and
2892 produce a single value. The operands might represent multiple data, as is
2893 the case with the <a href="#t_vector">vector</a> data type. The result value
2894 has the same type as its operands.</p>
2896 <p>There are several different binary operators:</p>
2900 <!-- _______________________________________________________________________ -->
2901 <div class="doc_subsubsection">
2902 <a name="i_add">'<tt>add</tt>' Instruction</a>
2905 <div class="doc_text">
2909 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2910 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2911 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2912 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2916 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2919 <p>The two arguments to the '<tt>add</tt>' instruction must
2920 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2921 integer values. Both arguments must have identical types.</p>
2924 <p>The value produced is the integer sum of the two operands.</p>
2926 <p>If the sum has unsigned overflow, the result returned is the mathematical
2927 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
2929 <p>Because LLVM integers use a two's complement representation, this instruction
2930 is appropriate for both signed and unsigned integers.</p>
2932 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
2933 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
2934 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
2935 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2939 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2944 <!-- _______________________________________________________________________ -->
2945 <div class="doc_subsubsection">
2946 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
2949 <div class="doc_text">
2953 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2957 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
2960 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
2961 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2962 floating point values. Both arguments must have identical types.</p>
2965 <p>The value produced is the floating point sum of the two operands.</p>
2969 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
2974 <!-- _______________________________________________________________________ -->
2975 <div class="doc_subsubsection">
2976 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2979 <div class="doc_text">
2983 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2984 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2985 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2986 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2990 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2993 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2994 '<tt>neg</tt>' instruction present in most other intermediate
2995 representations.</p>
2998 <p>The two arguments to the '<tt>sub</tt>' instruction must
2999 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3000 integer values. Both arguments must have identical types.</p>
3003 <p>The value produced is the integer difference of the two operands.</p>
3005 <p>If the difference has unsigned overflow, the result returned is the
3006 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3009 <p>Because LLVM integers use a two's complement representation, this instruction
3010 is appropriate for both signed and unsigned integers.</p>
3012 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3013 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3014 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3015 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
3019 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3020 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3025 <!-- _______________________________________________________________________ -->
3026 <div class="doc_subsubsection">
3027 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3030 <div class="doc_text">
3034 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3038 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3041 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3042 '<tt>fneg</tt>' instruction present in most other intermediate
3043 representations.</p>
3046 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3047 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3048 floating point values. Both arguments must have identical types.</p>
3051 <p>The value produced is the floating point difference of the two operands.</p>
3055 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3056 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3061 <!-- _______________________________________________________________________ -->
3062 <div class="doc_subsubsection">
3063 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3066 <div class="doc_text">
3070 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3071 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3072 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3073 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3077 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3080 <p>The two arguments to the '<tt>mul</tt>' instruction must
3081 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3082 integer values. Both arguments must have identical types.</p>
3085 <p>The value produced is the integer product of the two operands.</p>
3087 <p>If the result of the multiplication has unsigned overflow, the result
3088 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3089 width of the result.</p>
3091 <p>Because LLVM integers use a two's complement representation, and the result
3092 is the same width as the operands, this instruction returns the correct
3093 result for both signed and unsigned integers. If a full product
3094 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3095 be sign-extended or zero-extended as appropriate to the width of the full
3098 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3099 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3100 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3101 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
3105 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3110 <!-- _______________________________________________________________________ -->
3111 <div class="doc_subsubsection">
3112 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3115 <div class="doc_text">
3119 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3123 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3126 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3127 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3128 floating point values. Both arguments must have identical types.</p>
3131 <p>The value produced is the floating point product of the two operands.</p>
3135 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3140 <!-- _______________________________________________________________________ -->
3141 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
3144 <div class="doc_text">
3148 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3152 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3155 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3156 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3157 values. Both arguments must have identical types.</p>
3160 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3162 <p>Note that unsigned integer division and signed integer division are distinct
3163 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3165 <p>Division by zero leads to undefined behavior.</p>
3169 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3174 <!-- _______________________________________________________________________ -->
3175 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
3178 <div class="doc_text">
3182 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3183 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3187 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3190 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3191 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3192 values. Both arguments must have identical types.</p>
3195 <p>The value produced is the signed integer quotient of the two operands rounded
3198 <p>Note that signed integer division and unsigned integer division are distinct
3199 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3201 <p>Division by zero leads to undefined behavior. Overflow also leads to
3202 undefined behavior; this is a rare case, but can occur, for example, by doing
3203 a 32-bit division of -2147483648 by -1.</p>
3205 <p>If the <tt>exact</tt> keyword is present, the result value of the
3206 <tt>sdiv</tt> is undefined if the result would be rounded or if overflow
3211 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3216 <!-- _______________________________________________________________________ -->
3217 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
3218 Instruction</a> </div>
3220 <div class="doc_text">
3224 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3228 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3231 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3232 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3233 floating point values. Both arguments must have identical types.</p>
3236 <p>The value produced is the floating point quotient of the two operands.</p>
3240 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3245 <!-- _______________________________________________________________________ -->
3246 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3249 <div class="doc_text">
3253 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3257 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3258 division of its two arguments.</p>
3261 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3262 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3263 values. Both arguments must have identical types.</p>
3266 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3267 This instruction always performs an unsigned division to get the
3270 <p>Note that unsigned integer remainder and signed integer remainder are
3271 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3273 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3277 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3282 <!-- _______________________________________________________________________ -->
3283 <div class="doc_subsubsection">
3284 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3287 <div class="doc_text">
3291 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3295 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3296 division of its two operands. This instruction can also take
3297 <a href="#t_vector">vector</a> versions of the values in which case the
3298 elements must be integers.</p>
3301 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3302 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3303 values. Both arguments must have identical types.</p>
3306 <p>This instruction returns the <i>remainder</i> of a division (where the result
3307 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
3308 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
3309 a value. For more information about the difference,
3310 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3311 Math Forum</a>. For a table of how this is implemented in various languages,
3312 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3313 Wikipedia: modulo operation</a>.</p>
3315 <p>Note that signed integer remainder and unsigned integer remainder are
3316 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3318 <p>Taking the remainder of a division by zero leads to undefined behavior.
3319 Overflow also leads to undefined behavior; this is a rare case, but can
3320 occur, for example, by taking the remainder of a 32-bit division of
3321 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3322 lets srem be implemented using instructions that return both the result of
3323 the division and the remainder.)</p>
3327 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3332 <!-- _______________________________________________________________________ -->
3333 <div class="doc_subsubsection">
3334 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
3336 <div class="doc_text">
3340 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3344 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3345 its two operands.</p>
3348 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3349 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3350 floating point values. Both arguments must have identical types.</p>
3353 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3354 has the same sign as the dividend.</p>
3358 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3363 <!-- ======================================================================= -->
3364 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
3365 Operations</a> </div>
3367 <div class="doc_text">
3369 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3370 program. They are generally very efficient instructions and can commonly be
3371 strength reduced from other instructions. They require two operands of the
3372 same type, execute an operation on them, and produce a single value. The
3373 resulting value is the same type as its operands.</p>
3377 <!-- _______________________________________________________________________ -->
3378 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
3379 Instruction</a> </div>
3381 <div class="doc_text">
3385 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3389 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3390 a specified number of bits.</p>
3393 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3394 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3395 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3398 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3399 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3400 is (statically or dynamically) negative or equal to or larger than the number
3401 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3402 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3403 shift amount in <tt>op2</tt>.</p>
3407 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3408 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3409 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3410 <result> = shl i32 1, 32 <i>; undefined</i>
3411 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3416 <!-- _______________________________________________________________________ -->
3417 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
3418 Instruction</a> </div>
3420 <div class="doc_text">
3424 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3428 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
3429 operand shifted to the right a specified number of bits with zero fill.</p>
3432 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
3433 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3434 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3437 <p>This instruction always performs a logical shift right operation. The most
3438 significant bits of the result will be filled with zero bits after the shift.
3439 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
3440 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3441 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3442 shift amount in <tt>op2</tt>.</p>
3446 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
3447 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
3448 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
3449 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
3450 <result> = lshr i32 1, 32 <i>; undefined</i>
3451 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3456 <!-- _______________________________________________________________________ -->
3457 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
3458 Instruction</a> </div>
3459 <div class="doc_text">
3463 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3467 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
3468 operand shifted to the right a specified number of bits with sign
3472 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
3473 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3474 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3477 <p>This instruction always performs an arithmetic shift right operation, The
3478 most significant bits of the result will be filled with the sign bit
3479 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
3480 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
3481 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
3482 the corresponding shift amount in <tt>op2</tt>.</p>
3486 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
3487 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
3488 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
3489 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
3490 <result> = ashr i32 1, 32 <i>; undefined</i>
3491 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3496 <!-- _______________________________________________________________________ -->
3497 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
3498 Instruction</a> </div>
3500 <div class="doc_text">
3504 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3508 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
3512 <p>The two arguments to the '<tt>and</tt>' instruction must be
3513 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3514 values. Both arguments must have identical types.</p>
3517 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
3519 <table border="1" cellspacing="0" cellpadding="4">
3551 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
3552 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
3553 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
3556 <!-- _______________________________________________________________________ -->
3557 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
3559 <div class="doc_text">
3563 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3567 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
3571 <p>The two arguments to the '<tt>or</tt>' instruction must be
3572 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3573 values. Both arguments must have identical types.</p>
3576 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
3578 <table border="1" cellspacing="0" cellpadding="4">
3610 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3611 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3612 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3617 <!-- _______________________________________________________________________ -->
3618 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3619 Instruction</a> </div>
3621 <div class="doc_text">
3625 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3629 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
3630 its two operands. The <tt>xor</tt> is used to implement the "one's
3631 complement" operation, which is the "~" operator in C.</p>
3634 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3635 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3636 values. Both arguments must have identical types.</p>
3639 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3641 <table border="1" cellspacing="0" cellpadding="4">
3673 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3674 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3675 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3676 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3681 <!-- ======================================================================= -->
3682 <div class="doc_subsection">
3683 <a name="vectorops">Vector Operations</a>
3686 <div class="doc_text">
3688 <p>LLVM supports several instructions to represent vector operations in a
3689 target-independent manner. These instructions cover the element-access and
3690 vector-specific operations needed to process vectors effectively. While LLVM
3691 does directly support these vector operations, many sophisticated algorithms
3692 will want to use target-specific intrinsics to take full advantage of a
3693 specific target.</p>
3697 <!-- _______________________________________________________________________ -->
3698 <div class="doc_subsubsection">
3699 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3702 <div class="doc_text">
3706 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3710 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
3711 from a vector at a specified index.</p>
3715 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
3716 of <a href="#t_vector">vector</a> type. The second operand is an index
3717 indicating the position from which to extract the element. The index may be
3721 <p>The result is a scalar of the same type as the element type of
3722 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3723 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3724 results are undefined.</p>
3728 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3733 <!-- _______________________________________________________________________ -->
3734 <div class="doc_subsubsection">
3735 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3738 <div class="doc_text">
3742 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3746 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
3747 vector at a specified index.</p>
3750 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
3751 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
3752 whose type must equal the element type of the first operand. The third
3753 operand is an index indicating the position at which to insert the value.
3754 The index may be a variable.</p>
3757 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
3758 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
3759 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3760 results are undefined.</p>
3764 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3769 <!-- _______________________________________________________________________ -->
3770 <div class="doc_subsubsection">
3771 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3774 <div class="doc_text">
3778 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3782 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3783 from two input vectors, returning a vector with the same element type as the
3784 input and length that is the same as the shuffle mask.</p>
3787 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3788 with types that match each other. The third argument is a shuffle mask whose
3789 element type is always 'i32'. The result of the instruction is a vector
3790 whose length is the same as the shuffle mask and whose element type is the
3791 same as the element type of the first two operands.</p>
3793 <p>The shuffle mask operand is required to be a constant vector with either
3794 constant integer or undef values.</p>
3797 <p>The elements of the two input vectors are numbered from left to right across
3798 both of the vectors. The shuffle mask operand specifies, for each element of
3799 the result vector, which element of the two input vectors the result element
3800 gets. The element selector may be undef (meaning "don't care") and the
3801 second operand may be undef if performing a shuffle from only one vector.</p>
3805 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3806 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3807 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
3808 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3809 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
3810 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3811 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3812 <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>
3817 <!-- ======================================================================= -->
3818 <div class="doc_subsection">
3819 <a name="aggregateops">Aggregate Operations</a>
3822 <div class="doc_text">
3824 <p>LLVM supports several instructions for working with aggregate values.</p>
3828 <!-- _______________________________________________________________________ -->
3829 <div class="doc_subsubsection">
3830 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3833 <div class="doc_text">
3837 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3841 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3842 or array element from an aggregate value.</p>
3845 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
3846 of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type. The
3847 operands are constant indices to specify which value to extract in a similar
3848 manner as indices in a
3849 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
3852 <p>The result is the value at the position in the aggregate specified by the
3857 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3862 <!-- _______________________________________________________________________ -->
3863 <div class="doc_subsubsection">
3864 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3867 <div class="doc_text">
3871 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3875 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a struct field or
3876 array element in an aggregate.</p>
3880 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
3881 of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type. The
3882 second operand is a first-class value to insert. The following operands are
3883 constant indices indicating the position at which to insert the value in a
3884 similar manner as indices in a
3885 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction. The
3886 value to insert must have the same type as the value identified by the
3890 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
3891 that of <tt>val</tt> except that the value at the position specified by the
3892 indices is that of <tt>elt</tt>.</p>
3896 <result> = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3902 <!-- ======================================================================= -->
3903 <div class="doc_subsection">
3904 <a name="memoryops">Memory Access and Addressing Operations</a>
3907 <div class="doc_text">
3909 <p>A key design point of an SSA-based representation is how it represents
3910 memory. In LLVM, no memory locations are in SSA form, which makes things
3911 very simple. This section describes how to read, write, and allocate
3916 <!-- _______________________________________________________________________ -->
3917 <div class="doc_subsubsection">
3918 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3921 <div class="doc_text">
3925 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3929 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3930 currently executing function, to be automatically released when this function
3931 returns to its caller. The object is always allocated in the generic address
3932 space (address space zero).</p>
3935 <p>The '<tt>alloca</tt>' instruction
3936 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
3937 runtime stack, returning a pointer of the appropriate type to the program.
3938 If "NumElements" is specified, it is the number of elements allocated,
3939 otherwise "NumElements" is defaulted to be one. If a constant alignment is
3940 specified, the value result of the allocation is guaranteed to be aligned to
3941 at least that boundary. If not specified, or if zero, the target can choose
3942 to align the allocation on any convenient boundary compatible with the
3945 <p>'<tt>type</tt>' may be any sized type.</p>
3948 <p>Memory is allocated; a pointer is returned. The operation is undefined if
3949 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3950 memory is automatically released when the function returns. The
3951 '<tt>alloca</tt>' instruction is commonly used to represent automatic
3952 variables that must have an address available. When the function returns
3953 (either with the <tt><a href="#i_ret">ret</a></tt>
3954 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
3955 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
3959 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3960 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3961 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3962 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3967 <!-- _______________________________________________________________________ -->
3968 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3969 Instruction</a> </div>
3971 <div class="doc_text">
3975 <result> = load <ty>* <pointer>[, align <alignment>]
3976 <result> = volatile load <ty>* <pointer>[, align <alignment>]
3980 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3983 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
3984 from which to load. The pointer must point to
3985 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3986 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
3987 number or order of execution of this <tt>load</tt> with other
3988 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3991 <p>The optional constant "align" argument specifies the alignment of the
3992 operation (that is, the alignment of the memory address). A value of 0 or an
3993 omitted "align" argument means that the operation has the preferential
3994 alignment for the target. It is the responsibility of the code emitter to
3995 ensure that the alignment information is correct. Overestimating the
3996 alignment results in an undefined behavior. Underestimating the alignment may
3997 produce less efficient code. An alignment of 1 is always safe.</p>
4000 <p>The location of memory pointed to is loaded. If the value being loaded is of
4001 scalar type then the number of bytes read does not exceed the minimum number
4002 of bytes needed to hold all bits of the type. For example, loading an
4003 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4004 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4005 is undefined if the value was not originally written using a store of the
4010 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4011 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4012 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4017 <!-- _______________________________________________________________________ -->
4018 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
4019 Instruction</a> </div>
4021 <div class="doc_text">
4025 store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
4026 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
4030 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4033 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4034 and an address at which to store it. The type of the
4035 '<tt><pointer></tt>' operand must be a pointer to
4036 the <a href="#t_firstclass">first class</a> type of the
4037 '<tt><value></tt>' operand. If the <tt>store</tt> is marked
4038 as <tt>volatile</tt>, then the optimizer is not allowed to modify the number
4039 or order of execution of this <tt>store</tt> with other
4040 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
4043 <p>The optional constant "align" argument specifies the alignment of the
4044 operation (that is, the alignment of the memory address). A value of 0 or an
4045 omitted "align" argument means that the operation has the preferential
4046 alignment for the target. It is the responsibility of the code emitter to
4047 ensure that the alignment information is correct. Overestimating the
4048 alignment results in an undefined behavior. Underestimating the alignment may
4049 produce less efficient code. An alignment of 1 is always safe.</p>
4052 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4053 location specified by the '<tt><pointer></tt>' operand. If
4054 '<tt><value></tt>' is of scalar type then the number of bytes written
4055 does not exceed the minimum number of bytes needed to hold all bits of the
4056 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4057 writing a value of a type like <tt>i20</tt> with a size that is not an
4058 integral number of bytes, it is unspecified what happens to the extra bits
4059 that do not belong to the type, but they will typically be overwritten.</p>
4063 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4064 store i32 3, i32* %ptr <i>; yields {void}</i>
4065 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4070 <!-- _______________________________________________________________________ -->
4071 <div class="doc_subsubsection">
4072 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
4075 <div class="doc_text">
4079 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4080 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4084 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4085 subelement of an aggregate data structure. It performs address calculation
4086 only and does not access memory.</p>
4089 <p>The first argument is always a pointer, and forms the basis of the
4090 calculation. The remaining arguments are indices that indicate which of the
4091 elements of the aggregate object are indexed. The interpretation of each
4092 index is dependent on the type being indexed into. The first index always
4093 indexes the pointer value given as the first argument, the second index
4094 indexes a value of the type pointed to (not necessarily the value directly
4095 pointed to, since the first index can be non-zero), etc. The first type
4096 indexed into must be a pointer value, subsequent types can be arrays, vectors
4097 and structs. Note that subsequent types being indexed into can never be
4098 pointers, since that would require loading the pointer before continuing
4101 <p>The type of each index argument depends on the type it is indexing into.
4102 When indexing into a (optionally packed) structure, only <tt>i32</tt> integer
4103 <b>constants</b> are allowed. When indexing into an array, pointer or
4104 vector, integers of any width are allowed, and they are not required to be
4107 <p>For example, let's consider a C code fragment and how it gets compiled to
4110 <div class="doc_code">
4123 int *foo(struct ST *s) {
4124 return &s[1].Z.B[5][13];
4129 <p>The LLVM code generated by the GCC frontend is:</p>
4131 <div class="doc_code">
4133 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4134 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
4136 define i32* @foo(%ST* %s) {
4138 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
4145 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
4146 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
4147 }</tt>' type, a structure. The second index indexes into the third element
4148 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
4149 i8 }</tt>' type, another structure. The third index indexes into the second
4150 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
4151 array. The two dimensions of the array are subscripted into, yielding an
4152 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
4153 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
4155 <p>Note that it is perfectly legal to index partially through a structure,
4156 returning a pointer to an inner element. Because of this, the LLVM code for
4157 the given testcase is equivalent to:</p>
4160 define i32* @foo(%ST* %s) {
4161 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
4162 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
4163 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
4164 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
4165 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
4170 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
4171 <tt>getelementptr</tt> is undefined if the base pointer is not an
4172 <i>in bounds</i> address of an allocated object, or if any of the addresses
4173 that would be formed by successive addition of the offsets implied by the
4174 indices to the base address with infinitely precise arithmetic are not an
4175 <i>in bounds</i> address of that allocated object.
4176 The <i>in bounds</i> addresses for an allocated object are all the addresses
4177 that point into the object, plus the address one byte past the end.</p>
4179 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
4180 the base address with silently-wrapping two's complement arithmetic, and
4181 the result value of the <tt>getelementptr</tt> may be outside the object
4182 pointed to by the base pointer. The result value may not necessarily be
4183 used to access memory though, even if it happens to point into allocated
4184 storage. See the <a href="#pointeraliasing">Pointer Aliasing Rules</a>
4185 section for more information.</p>
4187 <p>The getelementptr instruction is often confusing. For some more insight into
4188 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
4192 <i>; yields [12 x i8]*:aptr</i>
4193 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4194 <i>; yields i8*:vptr</i>
4195 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4196 <i>; yields i8*:eptr</i>
4197 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4198 <i>; yields i32*:iptr</i>
4199 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4204 <!-- ======================================================================= -->
4205 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
4208 <div class="doc_text">
4210 <p>The instructions in this category are the conversion instructions (casting)
4211 which all take a single operand and a type. They perform various bit
4212 conversions on the operand.</p>
4216 <!-- _______________________________________________________________________ -->
4217 <div class="doc_subsubsection">
4218 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
4220 <div class="doc_text">
4224 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
4228 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
4229 type <tt>ty2</tt>.</p>
4232 <p>The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
4233 be an <a href="#t_integer">integer</a> type, and a type that specifies the
4234 size and type of the result, which must be
4235 an <a href="#t_integer">integer</a> type. The bit size of <tt>value</tt> must
4236 be larger than the bit size of <tt>ty2</tt>. Equal sized types are not
4240 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
4241 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
4242 source size must be larger than the destination size, <tt>trunc</tt> cannot
4243 be a <i>no-op cast</i>. It will always truncate bits.</p>
4247 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
4248 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
4249 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
4254 <!-- _______________________________________________________________________ -->
4255 <div class="doc_subsubsection">
4256 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
4258 <div class="doc_text">
4262 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
4266 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
4271 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
4272 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4273 also be of <a href="#t_integer">integer</a> type. The bit size of the
4274 <tt>value</tt> must be smaller than the bit size of the destination type,
4278 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
4279 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
4281 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
4285 %X = zext i32 257 to i64 <i>; yields i64:257</i>
4286 %Y = zext i1 true to i32 <i>; yields i32:1</i>
4291 <!-- _______________________________________________________________________ -->
4292 <div class="doc_subsubsection">
4293 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
4295 <div class="doc_text">
4299 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
4303 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
4306 <p>The '<tt>sext</tt>' instruction takes a value to cast, which must be of
4307 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4308 also be of <a href="#t_integer">integer</a> type. The bit size of the
4309 <tt>value</tt> must be smaller than the bit size of the destination type,
4313 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
4314 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
4315 of the type <tt>ty2</tt>.</p>
4317 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
4321 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
4322 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
4327 <!-- _______________________________________________________________________ -->
4328 <div class="doc_subsubsection">
4329 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
4332 <div class="doc_text">
4336 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
4340 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
4344 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
4345 point</a> value to cast and a <a href="#t_floating">floating point</a> type
4346 to cast it to. The size of <tt>value</tt> must be larger than the size of
4347 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
4348 <i>no-op cast</i>.</p>
4351 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
4352 <a href="#t_floating">floating point</a> type to a smaller
4353 <a href="#t_floating">floating point</a> type. If the value cannot fit
4354 within the destination type, <tt>ty2</tt>, then the results are
4359 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
4360 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
4365 <!-- _______________________________________________________________________ -->
4366 <div class="doc_subsubsection">
4367 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
4369 <div class="doc_text">
4373 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
4377 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
4378 floating point value.</p>
4381 <p>The '<tt>fpext</tt>' instruction takes a
4382 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
4383 a <a href="#t_floating">floating point</a> type to cast it to. The source
4384 type must be smaller than the destination type.</p>
4387 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
4388 <a href="#t_floating">floating point</a> type to a larger
4389 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
4390 used to make a <i>no-op cast</i> because it always changes bits. Use
4391 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
4395 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
4396 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
4401 <!-- _______________________________________________________________________ -->
4402 <div class="doc_subsubsection">
4403 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
4405 <div class="doc_text">
4409 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
4413 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
4414 unsigned integer equivalent of type <tt>ty2</tt>.</p>
4417 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
4418 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4419 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4420 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4421 vector integer type with the same number of elements as <tt>ty</tt></p>
4424 <p>The '<tt>fptoui</tt>' instruction converts its
4425 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4426 towards zero) unsigned integer value. If the value cannot fit
4427 in <tt>ty2</tt>, the results are undefined.</p>
4431 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
4432 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
4433 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
4438 <!-- _______________________________________________________________________ -->
4439 <div class="doc_subsubsection">
4440 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
4442 <div class="doc_text">
4446 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
4450 <p>The '<tt>fptosi</tt>' instruction converts
4451 <a href="#t_floating">floating point</a> <tt>value</tt> to
4452 type <tt>ty2</tt>.</p>
4455 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
4456 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4457 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4458 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4459 vector integer type with the same number of elements as <tt>ty</tt></p>
4462 <p>The '<tt>fptosi</tt>' instruction converts its
4463 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4464 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
4465 the results are undefined.</p>
4469 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
4470 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
4471 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
4476 <!-- _______________________________________________________________________ -->
4477 <div class="doc_subsubsection">
4478 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
4480 <div class="doc_text">
4484 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4488 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4489 integer and converts that value to the <tt>ty2</tt> type.</p>
4492 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4493 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4494 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4495 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4496 floating point type with the same number of elements as <tt>ty</tt></p>
4499 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4500 integer quantity and converts it to the corresponding floating point
4501 value. If the value cannot fit in the floating point value, the results are
4506 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4507 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4512 <!-- _______________________________________________________________________ -->
4513 <div class="doc_subsubsection">
4514 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4516 <div class="doc_text">
4520 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4524 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
4525 and converts that value to the <tt>ty2</tt> type.</p>
4528 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4529 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4530 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4531 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4532 floating point type with the same number of elements as <tt>ty</tt></p>
4535 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
4536 quantity and converts it to the corresponding floating point value. If the
4537 value cannot fit in the floating point value, the results are undefined.</p>
4541 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4542 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4547 <!-- _______________________________________________________________________ -->
4548 <div class="doc_subsubsection">
4549 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4551 <div class="doc_text">
4555 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4559 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4560 the integer type <tt>ty2</tt>.</p>
4563 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4564 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4565 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4568 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4569 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4570 truncating or zero extending that value to the size of the integer type. If
4571 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4572 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4573 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4578 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4579 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4584 <!-- _______________________________________________________________________ -->
4585 <div class="doc_subsubsection">
4586 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4588 <div class="doc_text">
4592 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4596 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
4597 pointer type, <tt>ty2</tt>.</p>
4600 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4601 value to cast, and a type to cast it to, which must be a
4602 <a href="#t_pointer">pointer</a> type.</p>
4605 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4606 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4607 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4608 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
4609 than the size of a pointer then a zero extension is done. If they are the
4610 same size, nothing is done (<i>no-op cast</i>).</p>
4614 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4615 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4616 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4621 <!-- _______________________________________________________________________ -->
4622 <div class="doc_subsubsection">
4623 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4625 <div class="doc_text">
4629 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4633 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4634 <tt>ty2</tt> without changing any bits.</p>
4637 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
4638 non-aggregate first class value, and a type to cast it to, which must also be
4639 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
4640 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
4641 identical. If the source type is a pointer, the destination type must also be
4642 a pointer. This instruction supports bitwise conversion of vectors to
4643 integers and to vectors of other types (as long as they have the same
4647 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4648 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4649 this conversion. The conversion is done as if the <tt>value</tt> had been
4650 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
4651 be converted to other pointer types with this instruction. To convert
4652 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
4653 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4657 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4658 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4659 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4664 <!-- ======================================================================= -->
4665 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4667 <div class="doc_text">
4669 <p>The instructions in this category are the "miscellaneous" instructions, which
4670 defy better classification.</p>
4674 <!-- _______________________________________________________________________ -->
4675 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4678 <div class="doc_text">
4682 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4686 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
4687 boolean values based on comparison of its two integer, integer vector, or
4688 pointer operands.</p>
4691 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4692 the condition code indicating the kind of comparison to perform. It is not a
4693 value, just a keyword. The possible condition code are:</p>
4696 <li><tt>eq</tt>: equal</li>
4697 <li><tt>ne</tt>: not equal </li>
4698 <li><tt>ugt</tt>: unsigned greater than</li>
4699 <li><tt>uge</tt>: unsigned greater or equal</li>
4700 <li><tt>ult</tt>: unsigned less than</li>
4701 <li><tt>ule</tt>: unsigned less or equal</li>
4702 <li><tt>sgt</tt>: signed greater than</li>
4703 <li><tt>sge</tt>: signed greater or equal</li>
4704 <li><tt>slt</tt>: signed less than</li>
4705 <li><tt>sle</tt>: signed less or equal</li>
4708 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4709 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
4710 typed. They must also be identical types.</p>
4713 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
4714 condition code given as <tt>cond</tt>. The comparison performed always yields
4715 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
4716 result, as follows:</p>
4719 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4720 <tt>false</tt> otherwise. No sign interpretation is necessary or
4723 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4724 <tt>false</tt> otherwise. No sign interpretation is necessary or
4727 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4728 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4730 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4731 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4732 to <tt>op2</tt>.</li>
4734 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4735 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4737 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4738 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4740 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4741 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4743 <li><tt>sge</tt>: interprets the operands as signed values and yields
4744 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4745 to <tt>op2</tt>.</li>
4747 <li><tt>slt</tt>: interprets the operands as signed values and yields
4748 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4750 <li><tt>sle</tt>: interprets the operands as signed values and yields
4751 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4754 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4755 values are compared as if they were integers.</p>
4757 <p>If the operands are integer vectors, then they are compared element by
4758 element. The result is an <tt>i1</tt> vector with the same number of elements
4759 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
4763 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4764 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4765 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4766 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4767 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4768 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4771 <p>Note that the code generator does not yet support vector types with
4772 the <tt>icmp</tt> instruction.</p>
4776 <!-- _______________________________________________________________________ -->
4777 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4780 <div class="doc_text">
4784 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4788 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
4789 values based on comparison of its operands.</p>
4791 <p>If the operands are floating point scalars, then the result type is a boolean
4792 (<a href="#t_integer"><tt>i1</tt></a>).</p>
4794 <p>If the operands are floating point vectors, then the result type is a vector
4795 of boolean with the same number of elements as the operands being
4799 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4800 the condition code indicating the kind of comparison to perform. It is not a
4801 value, just a keyword. The possible condition code are:</p>
4804 <li><tt>false</tt>: no comparison, always returns false</li>
4805 <li><tt>oeq</tt>: ordered and equal</li>
4806 <li><tt>ogt</tt>: ordered and greater than </li>
4807 <li><tt>oge</tt>: ordered and greater than or equal</li>
4808 <li><tt>olt</tt>: ordered and less than </li>
4809 <li><tt>ole</tt>: ordered and less than or equal</li>
4810 <li><tt>one</tt>: ordered and not equal</li>
4811 <li><tt>ord</tt>: ordered (no nans)</li>
4812 <li><tt>ueq</tt>: unordered or equal</li>
4813 <li><tt>ugt</tt>: unordered or greater than </li>
4814 <li><tt>uge</tt>: unordered or greater than or equal</li>
4815 <li><tt>ult</tt>: unordered or less than </li>
4816 <li><tt>ule</tt>: unordered or less than or equal</li>
4817 <li><tt>une</tt>: unordered or not equal</li>
4818 <li><tt>uno</tt>: unordered (either nans)</li>
4819 <li><tt>true</tt>: no comparison, always returns true</li>
4822 <p><i>Ordered</i> means that neither operand is a QNAN while
4823 <i>unordered</i> means that either operand may be a QNAN.</p>
4825 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
4826 a <a href="#t_floating">floating point</a> type or
4827 a <a href="#t_vector">vector</a> of floating point type. They must have
4828 identical types.</p>
4831 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4832 according to the condition code given as <tt>cond</tt>. If the operands are
4833 vectors, then the vectors are compared element by element. Each comparison
4834 performed always yields an <a href="#t_integer">i1</a> result, as
4838 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4840 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4841 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4843 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4844 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4846 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4847 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4849 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4850 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4852 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4853 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4855 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4856 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4858 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4860 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4861 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4863 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4864 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4866 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4867 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4869 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4870 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4872 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4873 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4875 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4876 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4878 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4880 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4885 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4886 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4887 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4888 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4891 <p>Note that the code generator does not yet support vector types with
4892 the <tt>fcmp</tt> instruction.</p>
4896 <!-- _______________________________________________________________________ -->
4897 <div class="doc_subsubsection">
4898 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4901 <div class="doc_text">
4905 <result> = phi <ty> [ <val0>, <label0>], ...
4909 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
4910 SSA graph representing the function.</p>
4913 <p>The type of the incoming values is specified with the first type field. After
4914 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
4915 one pair for each predecessor basic block of the current block. Only values
4916 of <a href="#t_firstclass">first class</a> type may be used as the value
4917 arguments to the PHI node. Only labels may be used as the label
4920 <p>There must be no non-phi instructions between the start of a basic block and
4921 the PHI instructions: i.e. PHI instructions must be first in a basic
4924 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
4925 occur on the edge from the corresponding predecessor block to the current
4926 block (but after any definition of an '<tt>invoke</tt>' instruction's return
4927 value on the same edge).</p>
4930 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4931 specified by the pair corresponding to the predecessor basic block that
4932 executed just prior to the current block.</p>
4936 Loop: ; Infinite loop that counts from 0 on up...
4937 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4938 %nextindvar = add i32 %indvar, 1
4944 <!-- _______________________________________________________________________ -->
4945 <div class="doc_subsubsection">
4946 <a name="i_select">'<tt>select</tt>' Instruction</a>
4949 <div class="doc_text">
4953 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4955 <i>selty</i> is either i1 or {<N x i1>}
4959 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
4960 condition, without branching.</p>
4964 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
4965 values indicating the condition, and two values of the
4966 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
4967 vectors and the condition is a scalar, then entire vectors are selected, not
4968 individual elements.</p>
4971 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
4972 first value argument; otherwise, it returns the second value argument.</p>
4974 <p>If the condition is a vector of i1, then the value arguments must be vectors
4975 of the same size, and the selection is done element by element.</p>
4979 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4982 <p>Note that the code generator does not yet support conditions
4983 with vector type.</p>
4987 <!-- _______________________________________________________________________ -->
4988 <div class="doc_subsubsection">
4989 <a name="i_call">'<tt>call</tt>' Instruction</a>
4992 <div class="doc_text">
4996 <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>]
5000 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
5003 <p>This instruction requires several arguments:</p>
5006 <li>The optional "tail" marker indicates whether the callee function accesses
5007 any allocas or varargs in the caller. If the "tail" marker is present,
5008 the function call is eligible for tail call optimization. Note that calls
5009 may be marked "tail" even if they do not occur before
5010 a <a href="#i_ret"><tt>ret</tt></a> instruction.</li>
5012 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5013 convention</a> the call should use. If none is specified, the call
5014 defaults to using C calling conventions.</li>
5016 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5017 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5018 '<tt>inreg</tt>' attributes are valid here.</li>
5020 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5021 type of the return value. Functions that return no value are marked
5022 <tt><a href="#t_void">void</a></tt>.</li>
5024 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5025 being invoked. The argument types must match the types implied by this
5026 signature. This type can be omitted if the function is not varargs and if
5027 the function type does not return a pointer to a function.</li>
5029 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5030 be invoked. In most cases, this is a direct function invocation, but
5031 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5032 to function value.</li>
5034 <li>'<tt>function args</tt>': argument list whose types match the function
5035 signature argument types. All arguments must be of
5036 <a href="#t_firstclass">first class</a> type. If the function signature
5037 indicates the function accepts a variable number of arguments, the extra
5038 arguments can be specified.</li>
5040 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5041 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5042 '<tt>readnone</tt>' attributes are valid here.</li>
5046 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5047 a specified function, with its incoming arguments bound to the specified
5048 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5049 function, control flow continues with the instruction after the function
5050 call, and the return value of the function is bound to the result
5055 %retval = call i32 @test(i32 %argc)
5056 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
5057 %X = tail call i32 @foo() <i>; yields i32</i>
5058 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5059 call void %foo(i8 97 signext)
5061 %struct.A = type { i32, i8 }
5062 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
5063 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5064 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
5065 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
5066 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
5069 <p>llvm treats calls to some functions with names and arguments that match the
5070 standard C99 library as being the C99 library functions, and may perform
5071 optimizations or generate code for them under that assumption. This is
5072 something we'd like to change in the future to provide better support for
5073 freestanding environments and non-C-based langauges.</p>
5077 <!-- _______________________________________________________________________ -->
5078 <div class="doc_subsubsection">
5079 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5082 <div class="doc_text">
5086 <resultval> = va_arg <va_list*> <arglist>, <argty>
5090 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
5091 the "variable argument" area of a function call. It is used to implement the
5092 <tt>va_arg</tt> macro in C.</p>
5095 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
5096 argument. It returns a value of the specified argument type and increments
5097 the <tt>va_list</tt> to point to the next argument. The actual type
5098 of <tt>va_list</tt> is target specific.</p>
5101 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
5102 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
5103 to the next argument. For more information, see the variable argument
5104 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
5106 <p>It is legal for this instruction to be called in a function which does not
5107 take a variable number of arguments, for example, the <tt>vfprintf</tt>
5110 <p><tt>va_arg</tt> is an LLVM instruction instead of
5111 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
5115 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
5117 <p>Note that the code generator does not yet fully support va_arg on many
5118 targets. Also, it does not currently support va_arg with aggregate types on
5123 <!-- *********************************************************************** -->
5124 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
5125 <!-- *********************************************************************** -->
5127 <div class="doc_text">
5129 <p>LLVM supports the notion of an "intrinsic function". These functions have
5130 well known names and semantics and are required to follow certain
5131 restrictions. Overall, these intrinsics represent an extension mechanism for
5132 the LLVM language that does not require changing all of the transformations
5133 in LLVM when adding to the language (or the bitcode reader/writer, the
5134 parser, etc...).</p>
5136 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
5137 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5138 begin with this prefix. Intrinsic functions must always be external
5139 functions: you cannot define the body of intrinsic functions. Intrinsic
5140 functions may only be used in call or invoke instructions: it is illegal to
5141 take the address of an intrinsic function. Additionally, because intrinsic
5142 functions are part of the LLVM language, it is required if any are added that
5143 they be documented here.</p>
5145 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5146 family of functions that perform the same operation but on different data
5147 types. Because LLVM can represent over 8 million different integer types,
5148 overloading is used commonly to allow an intrinsic function to operate on any
5149 integer type. One or more of the argument types or the result type can be
5150 overloaded to accept any integer type. Argument types may also be defined as
5151 exactly matching a previous argument's type or the result type. This allows
5152 an intrinsic function which accepts multiple arguments, but needs all of them
5153 to be of the same type, to only be overloaded with respect to a single
5154 argument or the result.</p>
5156 <p>Overloaded intrinsics will have the names of its overloaded argument types
5157 encoded into its function name, each preceded by a period. Only those types
5158 which are overloaded result in a name suffix. Arguments whose type is matched
5159 against another type do not. For example, the <tt>llvm.ctpop</tt> function
5160 can take an integer of any width and returns an integer of exactly the same
5161 integer width. This leads to a family of functions such as
5162 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
5163 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5164 suffix is required. Because the argument's type is matched against the return
5165 type, it does not require its own name suffix.</p>
5167 <p>To learn how to add an intrinsic function, please see the
5168 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
5172 <!-- ======================================================================= -->
5173 <div class="doc_subsection">
5174 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
5177 <div class="doc_text">
5179 <p>Variable argument support is defined in LLVM with
5180 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
5181 intrinsic functions. These functions are related to the similarly named
5182 macros defined in the <tt><stdarg.h></tt> header file.</p>
5184 <p>All of these functions operate on arguments that use a target-specific value
5185 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5186 not define what this type is, so all transformations should be prepared to
5187 handle these functions regardless of the type used.</p>
5189 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
5190 instruction and the variable argument handling intrinsic functions are
5193 <div class="doc_code">
5195 define i32 @test(i32 %X, ...) {
5196 ; Initialize variable argument processing
5198 %ap2 = bitcast i8** %ap to i8*
5199 call void @llvm.va_start(i8* %ap2)
5201 ; Read a single integer argument
5202 %tmp = va_arg i8** %ap, i32
5204 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5206 %aq2 = bitcast i8** %aq to i8*
5207 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5208 call void @llvm.va_end(i8* %aq2)
5210 ; Stop processing of arguments.
5211 call void @llvm.va_end(i8* %ap2)
5215 declare void @llvm.va_start(i8*)
5216 declare void @llvm.va_copy(i8*, i8*)
5217 declare void @llvm.va_end(i8*)
5223 <!-- _______________________________________________________________________ -->
5224 <div class="doc_subsubsection">
5225 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
5229 <div class="doc_text">
5233 declare void %llvm.va_start(i8* <arglist>)
5237 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
5238 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
5241 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
5244 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
5245 macro available in C. In a target-dependent way, it initializes
5246 the <tt>va_list</tt> element to which the argument points, so that the next
5247 call to <tt>va_arg</tt> will produce the first variable argument passed to
5248 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
5249 need to know the last argument of the function as the compiler can figure
5254 <!-- _______________________________________________________________________ -->
5255 <div class="doc_subsubsection">
5256 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
5259 <div class="doc_text">
5263 declare void @llvm.va_end(i8* <arglist>)
5267 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
5268 which has been initialized previously
5269 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
5270 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
5273 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
5276 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
5277 macro available in C. In a target-dependent way, it destroys
5278 the <tt>va_list</tt> element to which the argument points. Calls
5279 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
5280 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
5281 with calls to <tt>llvm.va_end</tt>.</p>
5285 <!-- _______________________________________________________________________ -->
5286 <div class="doc_subsubsection">
5287 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
5290 <div class="doc_text">
5294 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
5298 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
5299 from the source argument list to the destination argument list.</p>
5302 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
5303 The second argument is a pointer to a <tt>va_list</tt> element to copy
5307 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
5308 macro available in C. In a target-dependent way, it copies the
5309 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
5310 element. This intrinsic is necessary because
5311 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
5312 arbitrarily complex and require, for example, memory allocation.</p>
5316 <!-- ======================================================================= -->
5317 <div class="doc_subsection">
5318 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
5321 <div class="doc_text">
5323 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
5324 Collection</a> (GC) requires the implementation and generation of these
5325 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
5326 roots on the stack</a>, as well as garbage collector implementations that
5327 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
5328 barriers. Front-ends for type-safe garbage collected languages should generate
5329 these intrinsics to make use of the LLVM garbage collectors. For more details,
5330 see <a href="GarbageCollection.html">Accurate Garbage Collection with
5333 <p>The garbage collection intrinsics only operate on objects in the generic
5334 address space (address space zero).</p>
5338 <!-- _______________________________________________________________________ -->
5339 <div class="doc_subsubsection">
5340 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
5343 <div class="doc_text">
5347 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5351 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
5352 the code generator, and allows some metadata to be associated with it.</p>
5355 <p>The first argument specifies the address of a stack object that contains the
5356 root pointer. The second pointer (which must be either a constant or a
5357 global value address) contains the meta-data to be associated with the
5361 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
5362 location. At compile-time, the code generator generates information to allow
5363 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
5364 intrinsic may only be used in a function which <a href="#gc">specifies a GC
5369 <!-- _______________________________________________________________________ -->
5370 <div class="doc_subsubsection">
5371 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
5374 <div class="doc_text">
5378 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5382 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
5383 locations, allowing garbage collector implementations that require read
5387 <p>The second argument is the address to read from, which should be an address
5388 allocated from the garbage collector. The first object is a pointer to the
5389 start of the referenced object, if needed by the language runtime (otherwise
5393 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
5394 instruction, but may be replaced with substantially more complex code by the
5395 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
5396 may only be used in a function which <a href="#gc">specifies a GC
5401 <!-- _______________________________________________________________________ -->
5402 <div class="doc_subsubsection">
5403 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5406 <div class="doc_text">
5410 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5414 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5415 locations, allowing garbage collector implementations that require write
5416 barriers (such as generational or reference counting collectors).</p>
5419 <p>The first argument is the reference to store, the second is the start of the
5420 object to store it to, and the third is the address of the field of Obj to
5421 store to. If the runtime does not require a pointer to the object, Obj may
5425 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5426 instruction, but may be replaced with substantially more complex code by the
5427 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5428 may only be used in a function which <a href="#gc">specifies a GC
5433 <!-- ======================================================================= -->
5434 <div class="doc_subsection">
5435 <a name="int_codegen">Code Generator Intrinsics</a>
5438 <div class="doc_text">
5440 <p>These intrinsics are provided by LLVM to expose special features that may
5441 only be implemented with code generator support.</p>
5445 <!-- _______________________________________________________________________ -->
5446 <div class="doc_subsubsection">
5447 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5450 <div class="doc_text">
5454 declare i8 *@llvm.returnaddress(i32 <level>)
5458 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5459 target-specific value indicating the return address of the current function
5460 or one of its callers.</p>
5463 <p>The argument to this intrinsic indicates which function to return the address
5464 for. Zero indicates the calling function, one indicates its caller, etc.
5465 The argument is <b>required</b> to be a constant integer value.</p>
5468 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
5469 indicating the return address of the specified call frame, or zero if it
5470 cannot be identified. The value returned by this intrinsic is likely to be
5471 incorrect or 0 for arguments other than zero, so it should only be used for
5472 debugging purposes.</p>
5474 <p>Note that calling this intrinsic does not prevent function inlining or other
5475 aggressive transformations, so the value returned may not be that of the
5476 obvious source-language caller.</p>
5480 <!-- _______________________________________________________________________ -->
5481 <div class="doc_subsubsection">
5482 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5485 <div class="doc_text">
5489 declare i8 *@llvm.frameaddress(i32 <level>)
5493 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5494 target-specific frame pointer value for the specified stack frame.</p>
5497 <p>The argument to this intrinsic indicates which function to return the frame
5498 pointer for. Zero indicates the calling function, one indicates its caller,
5499 etc. The argument is <b>required</b> to be a constant integer value.</p>
5502 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
5503 indicating the frame address of the specified call frame, or zero if it
5504 cannot be identified. The value returned by this intrinsic is likely to be
5505 incorrect or 0 for arguments other than zero, so it should only be used for
5506 debugging purposes.</p>
5508 <p>Note that calling this intrinsic does not prevent function inlining or other
5509 aggressive transformations, so the value returned may not be that of the
5510 obvious source-language caller.</p>
5514 <!-- _______________________________________________________________________ -->
5515 <div class="doc_subsubsection">
5516 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5519 <div class="doc_text">
5523 declare i8 *@llvm.stacksave()
5527 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
5528 of the function stack, for use
5529 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
5530 useful for implementing language features like scoped automatic variable
5531 sized arrays in C99.</p>
5534 <p>This intrinsic returns a opaque pointer value that can be passed
5535 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
5536 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
5537 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
5538 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
5539 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
5540 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
5544 <!-- _______________________________________________________________________ -->
5545 <div class="doc_subsubsection">
5546 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5549 <div class="doc_text">
5553 declare void @llvm.stackrestore(i8 * %ptr)
5557 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5558 the function stack to the state it was in when the
5559 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
5560 executed. This is useful for implementing language features like scoped
5561 automatic variable sized arrays in C99.</p>
5564 <p>See the description
5565 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
5569 <!-- _______________________________________________________________________ -->
5570 <div class="doc_subsubsection">
5571 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5574 <div class="doc_text">
5578 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5582 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
5583 insert a prefetch instruction if supported; otherwise, it is a noop.
5584 Prefetches have no effect on the behavior of the program but can change its
5585 performance characteristics.</p>
5588 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
5589 specifier determining if the fetch should be for a read (0) or write (1),
5590 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5591 locality, to (3) - extremely local keep in cache. The <tt>rw</tt>
5592 and <tt>locality</tt> arguments must be constant integers.</p>
5595 <p>This intrinsic does not modify the behavior of the program. In particular,
5596 prefetches cannot trap and do not produce a value. On targets that support
5597 this intrinsic, the prefetch can provide hints to the processor cache for
5598 better performance.</p>
5602 <!-- _______________________________________________________________________ -->
5603 <div class="doc_subsubsection">
5604 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5607 <div class="doc_text">
5611 declare void @llvm.pcmarker(i32 <id>)
5615 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
5616 Counter (PC) in a region of code to simulators and other tools. The method
5617 is target specific, but it is expected that the marker will use exported
5618 symbols to transmit the PC of the marker. The marker makes no guarantees
5619 that it will remain with any specific instruction after optimizations. It is
5620 possible that the presence of a marker will inhibit optimizations. The
5621 intended use is to be inserted after optimizations to allow correlations of
5622 simulation runs.</p>
5625 <p><tt>id</tt> is a numerical id identifying the marker.</p>
5628 <p>This intrinsic does not modify the behavior of the program. Backends that do
5629 not support this intrinisic may ignore it.</p>
5633 <!-- _______________________________________________________________________ -->
5634 <div class="doc_subsubsection">
5635 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5638 <div class="doc_text">
5642 declare i64 @llvm.readcyclecounter( )
5646 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5647 counter register (or similar low latency, high accuracy clocks) on those
5648 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5649 should map to RPCC. As the backing counters overflow quickly (on the order
5650 of 9 seconds on alpha), this should only be used for small timings.</p>
5653 <p>When directly supported, reading the cycle counter should not modify any
5654 memory. Implementations are allowed to either return a application specific
5655 value or a system wide value. On backends without support, this is lowered
5656 to a constant 0.</p>
5660 <!-- ======================================================================= -->
5661 <div class="doc_subsection">
5662 <a name="int_libc">Standard C Library Intrinsics</a>
5665 <div class="doc_text">
5667 <p>LLVM provides intrinsics for a few important standard C library functions.
5668 These intrinsics allow source-language front-ends to pass information about
5669 the alignment of the pointer arguments to the code generator, providing
5670 opportunity for more efficient code generation.</p>
5674 <!-- _______________________________________________________________________ -->
5675 <div class="doc_subsubsection">
5676 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5679 <div class="doc_text">
5682 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
5683 integer bit width. Not all targets support all bit widths however.</p>
5686 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5687 i8 <len>, i32 <align>)
5688 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5689 i16 <len>, i32 <align>)
5690 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5691 i32 <len>, i32 <align>)
5692 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5693 i64 <len>, i32 <align>)
5697 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5698 source location to the destination location.</p>
5700 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5701 intrinsics do not return a value, and takes an extra alignment argument.</p>
5704 <p>The first argument is a pointer to the destination, the second is a pointer
5705 to the source. The third argument is an integer argument specifying the
5706 number of bytes to copy, and the fourth argument is the alignment of the
5707 source and destination locations.</p>
5709 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5710 then the caller guarantees that both the source and destination pointers are
5711 aligned to that boundary.</p>
5714 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5715 source location to the destination location, which are not allowed to
5716 overlap. It copies "len" bytes of memory over. If the argument is known to
5717 be aligned to some boundary, this can be specified as the fourth argument,
5718 otherwise it should be set to 0 or 1.</p>
5722 <!-- _______________________________________________________________________ -->
5723 <div class="doc_subsubsection">
5724 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5727 <div class="doc_text">
5730 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5731 width. Not all targets support all bit widths however.</p>
5734 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5735 i8 <len>, i32 <align>)
5736 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5737 i16 <len>, i32 <align>)
5738 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5739 i32 <len>, i32 <align>)
5740 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5741 i64 <len>, i32 <align>)
5745 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
5746 source location to the destination location. It is similar to the
5747 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
5750 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5751 intrinsics do not return a value, and takes an extra alignment argument.</p>
5754 <p>The first argument is a pointer to the destination, the second is a pointer
5755 to the source. The third argument is an integer argument specifying the
5756 number of bytes to copy, and the fourth argument is the alignment of the
5757 source and destination locations.</p>
5759 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5760 then the caller guarantees that the source and destination pointers are
5761 aligned to that boundary.</p>
5764 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
5765 source location to the destination location, which may overlap. It copies
5766 "len" bytes of memory over. If the argument is known to be aligned to some
5767 boundary, this can be specified as the fourth argument, otherwise it should
5768 be set to 0 or 1.</p>
5772 <!-- _______________________________________________________________________ -->
5773 <div class="doc_subsubsection">
5774 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5777 <div class="doc_text">
5780 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5781 width. Not all targets support all bit widths however.</p>
5784 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5785 i8 <len>, i32 <align>)
5786 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5787 i16 <len>, i32 <align>)
5788 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5789 i32 <len>, i32 <align>)
5790 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5791 i64 <len>, i32 <align>)
5795 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
5796 particular byte value.</p>
5798 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
5799 intrinsic does not return a value, and takes an extra alignment argument.</p>
5802 <p>The first argument is a pointer to the destination to fill, the second is the
5803 byte value to fill it with, the third argument is an integer argument
5804 specifying the number of bytes to fill, and the fourth argument is the known
5805 alignment of destination location.</p>
5807 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5808 then the caller guarantees that the destination pointer is aligned to that
5812 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
5813 at the destination location. If the argument is known to be aligned to some
5814 boundary, this can be specified as the fourth argument, otherwise it should
5815 be set to 0 or 1.</p>
5819 <!-- _______________________________________________________________________ -->
5820 <div class="doc_subsubsection">
5821 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5824 <div class="doc_text">
5827 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5828 floating point or vector of floating point type. Not all targets support all
5832 declare float @llvm.sqrt.f32(float %Val)
5833 declare double @llvm.sqrt.f64(double %Val)
5834 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5835 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5836 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5840 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5841 returning the same value as the libm '<tt>sqrt</tt>' functions would.
5842 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
5843 behavior for negative numbers other than -0.0 (which allows for better
5844 optimization, because there is no need to worry about errno being
5845 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
5848 <p>The argument and return value are floating point numbers of the same
5852 <p>This function returns the sqrt of the specified operand if it is a
5853 nonnegative floating point number.</p>
5857 <!-- _______________________________________________________________________ -->
5858 <div class="doc_subsubsection">
5859 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5862 <div class="doc_text">
5865 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5866 floating point or vector of floating point type. Not all targets support all
5870 declare float @llvm.powi.f32(float %Val, i32 %power)
5871 declare double @llvm.powi.f64(double %Val, i32 %power)
5872 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5873 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5874 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5878 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5879 specified (positive or negative) power. The order of evaluation of
5880 multiplications is not defined. When a vector of floating point type is
5881 used, the second argument remains a scalar integer value.</p>
5884 <p>The second argument is an integer power, and the first is a value to raise to
5888 <p>This function returns the first value raised to the second power with an
5889 unspecified sequence of rounding operations.</p>
5893 <!-- _______________________________________________________________________ -->
5894 <div class="doc_subsubsection">
5895 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5898 <div class="doc_text">
5901 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5902 floating point or vector of floating point type. Not all targets support all
5906 declare float @llvm.sin.f32(float %Val)
5907 declare double @llvm.sin.f64(double %Val)
5908 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5909 declare fp128 @llvm.sin.f128(fp128 %Val)
5910 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5914 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
5917 <p>The argument and return value are floating point numbers of the same
5921 <p>This function returns the sine of the specified operand, returning the same
5922 values as the libm <tt>sin</tt> functions would, and handles error conditions
5923 in the same way.</p>
5927 <!-- _______________________________________________________________________ -->
5928 <div class="doc_subsubsection">
5929 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5932 <div class="doc_text">
5935 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5936 floating point or vector of floating point type. Not all targets support all
5940 declare float @llvm.cos.f32(float %Val)
5941 declare double @llvm.cos.f64(double %Val)
5942 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5943 declare fp128 @llvm.cos.f128(fp128 %Val)
5944 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5948 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
5951 <p>The argument and return value are floating point numbers of the same
5955 <p>This function returns the cosine of the specified operand, returning the same
5956 values as the libm <tt>cos</tt> functions would, and handles error conditions
5957 in the same way.</p>
5961 <!-- _______________________________________________________________________ -->
5962 <div class="doc_subsubsection">
5963 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5966 <div class="doc_text">
5969 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5970 floating point or vector of floating point type. Not all targets support all
5974 declare float @llvm.pow.f32(float %Val, float %Power)
5975 declare double @llvm.pow.f64(double %Val, double %Power)
5976 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5977 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5978 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5982 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5983 specified (positive or negative) power.</p>
5986 <p>The second argument is a floating point power, and the first is a value to
5987 raise to that power.</p>
5990 <p>This function returns the first value raised to the second power, returning
5991 the same values as the libm <tt>pow</tt> functions would, and handles error
5992 conditions in the same way.</p>
5996 <!-- ======================================================================= -->
5997 <div class="doc_subsection">
5998 <a name="int_manip">Bit Manipulation Intrinsics</a>
6001 <div class="doc_text">
6003 <p>LLVM provides intrinsics for a few important bit manipulation operations.
6004 These allow efficient code generation for some algorithms.</p>
6008 <!-- _______________________________________________________________________ -->
6009 <div class="doc_subsubsection">
6010 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
6013 <div class="doc_text">
6016 <p>This is an overloaded intrinsic function. You can use bswap on any integer
6017 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
6020 declare i16 @llvm.bswap.i16(i16 <id>)
6021 declare i32 @llvm.bswap.i32(i32 <id>)
6022 declare i64 @llvm.bswap.i64(i64 <id>)
6026 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
6027 values with an even number of bytes (positive multiple of 16 bits). These
6028 are useful for performing operations on data that is not in the target's
6029 native byte order.</p>
6032 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
6033 and low byte of the input i16 swapped. Similarly,
6034 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
6035 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
6036 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
6037 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
6038 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
6039 more, respectively).</p>
6043 <!-- _______________________________________________________________________ -->
6044 <div class="doc_subsubsection">
6045 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
6048 <div class="doc_text">
6051 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6052 width. Not all targets support all bit widths however.</p>
6055 declare i8 @llvm.ctpop.i8(i8 <src>)
6056 declare i16 @llvm.ctpop.i16(i16 <src>)
6057 declare i32 @llvm.ctpop.i32(i32 <src>)
6058 declare i64 @llvm.ctpop.i64(i64 <src>)
6059 declare i256 @llvm.ctpop.i256(i256 <src>)
6063 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
6067 <p>The only argument is the value to be counted. The argument may be of any
6068 integer type. The return type must match the argument type.</p>
6071 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.</p>
6075 <!-- _______________________________________________________________________ -->
6076 <div class="doc_subsubsection">
6077 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
6080 <div class="doc_text">
6083 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
6084 integer bit width. Not all targets support all bit widths however.</p>
6087 declare i8 @llvm.ctlz.i8 (i8 <src>)
6088 declare i16 @llvm.ctlz.i16(i16 <src>)
6089 declare i32 @llvm.ctlz.i32(i32 <src>)
6090 declare i64 @llvm.ctlz.i64(i64 <src>)
6091 declare i256 @llvm.ctlz.i256(i256 <src>)
6095 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
6096 leading zeros in a variable.</p>
6099 <p>The only argument is the value to be counted. The argument may be of any
6100 integer type. The return type must match the argument type.</p>
6103 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
6104 zeros in a variable. If the src == 0 then the result is the size in bits of
6105 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
6109 <!-- _______________________________________________________________________ -->
6110 <div class="doc_subsubsection">
6111 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
6114 <div class="doc_text">
6117 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
6118 integer bit width. Not all targets support all bit widths however.</p>
6121 declare i8 @llvm.cttz.i8 (i8 <src>)
6122 declare i16 @llvm.cttz.i16(i16 <src>)
6123 declare i32 @llvm.cttz.i32(i32 <src>)
6124 declare i64 @llvm.cttz.i64(i64 <src>)
6125 declare i256 @llvm.cttz.i256(i256 <src>)
6129 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
6133 <p>The only argument is the value to be counted. The argument may be of any
6134 integer type. The return type must match the argument type.</p>
6137 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
6138 zeros in a variable. If the src == 0 then the result is the size in bits of
6139 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
6143 <!-- ======================================================================= -->
6144 <div class="doc_subsection">
6145 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
6148 <div class="doc_text">
6150 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
6154 <!-- _______________________________________________________________________ -->
6155 <div class="doc_subsubsection">
6156 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
6159 <div class="doc_text">
6162 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
6163 on any integer bit width.</p>
6166 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6167 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6168 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6172 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6173 a signed addition of the two arguments, and indicate whether an overflow
6174 occurred during the signed summation.</p>
6177 <p>The arguments (%a and %b) and the first element of the result structure may
6178 be of integer types of any bit width, but they must have the same bit
6179 width. The second element of the result structure must be of
6180 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6181 undergo signed addition.</p>
6184 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6185 a signed addition of the two variables. They return a structure — the
6186 first element of which is the signed summation, and the second element of
6187 which is a bit specifying if the signed summation resulted in an
6192 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6193 %sum = extractvalue {i32, i1} %res, 0
6194 %obit = extractvalue {i32, i1} %res, 1
6195 br i1 %obit, label %overflow, label %normal
6200 <!-- _______________________________________________________________________ -->
6201 <div class="doc_subsubsection">
6202 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
6205 <div class="doc_text">
6208 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
6209 on any integer bit width.</p>
6212 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6213 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6214 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6218 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6219 an unsigned addition of the two arguments, and indicate whether a carry
6220 occurred during the unsigned summation.</p>
6223 <p>The arguments (%a and %b) and the first element of the result structure may
6224 be of integer types of any bit width, but they must have the same bit
6225 width. The second element of the result structure must be of
6226 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6227 undergo unsigned addition.</p>
6230 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6231 an unsigned addition of the two arguments. They return a structure —
6232 the first element of which is the sum, and the second element of which is a
6233 bit specifying if the unsigned summation resulted in a carry.</p>
6237 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6238 %sum = extractvalue {i32, i1} %res, 0
6239 %obit = extractvalue {i32, i1} %res, 1
6240 br i1 %obit, label %carry, label %normal
6245 <!-- _______________________________________________________________________ -->
6246 <div class="doc_subsubsection">
6247 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6250 <div class="doc_text">
6253 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6254 on any integer bit width.</p>
6257 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6258 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6259 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6263 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6264 a signed subtraction of the two arguments, and indicate whether an overflow
6265 occurred during the signed subtraction.</p>
6268 <p>The arguments (%a and %b) and the first element of the result structure may
6269 be of integer types of any bit width, but they must have the same bit
6270 width. The second element of the result structure must be of
6271 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6272 undergo signed subtraction.</p>
6275 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6276 a signed subtraction of the two arguments. They return a structure —
6277 the first element of which is the subtraction, and the second element of
6278 which is a bit specifying if the signed subtraction resulted in an
6283 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6284 %sum = extractvalue {i32, i1} %res, 0
6285 %obit = extractvalue {i32, i1} %res, 1
6286 br i1 %obit, label %overflow, label %normal
6291 <!-- _______________________________________________________________________ -->
6292 <div class="doc_subsubsection">
6293 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6296 <div class="doc_text">
6299 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6300 on any integer bit width.</p>
6303 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6304 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6305 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6309 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6310 an unsigned subtraction of the two arguments, and indicate whether an
6311 overflow occurred during the unsigned subtraction.</p>
6314 <p>The arguments (%a and %b) and the first element of the result structure may
6315 be of integer types of any bit width, but they must have the same bit
6316 width. The second element of the result structure must be of
6317 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6318 undergo unsigned subtraction.</p>
6321 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6322 an unsigned subtraction of the two arguments. They return a structure —
6323 the first element of which is the subtraction, and the second element of
6324 which is a bit specifying if the unsigned subtraction resulted in an
6329 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6330 %sum = extractvalue {i32, i1} %res, 0
6331 %obit = extractvalue {i32, i1} %res, 1
6332 br i1 %obit, label %overflow, label %normal
6337 <!-- _______________________________________________________________________ -->
6338 <div class="doc_subsubsection">
6339 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6342 <div class="doc_text">
6345 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6346 on any integer bit width.</p>
6349 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6350 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6351 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6356 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6357 a signed multiplication of the two arguments, and indicate whether an
6358 overflow occurred during the signed multiplication.</p>
6361 <p>The arguments (%a and %b) and the first element of the result structure may
6362 be of integer types of any bit width, but they must have the same bit
6363 width. The second element of the result structure must be of
6364 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6365 undergo signed multiplication.</p>
6368 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6369 a signed multiplication of the two arguments. They return a structure —
6370 the first element of which is the multiplication, and the second element of
6371 which is a bit specifying if the signed multiplication resulted in an
6376 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6377 %sum = extractvalue {i32, i1} %res, 0
6378 %obit = extractvalue {i32, i1} %res, 1
6379 br i1 %obit, label %overflow, label %normal
6384 <!-- _______________________________________________________________________ -->
6385 <div class="doc_subsubsection">
6386 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6389 <div class="doc_text">
6392 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6393 on any integer bit width.</p>
6396 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6397 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6398 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6402 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6403 a unsigned multiplication of the two arguments, and indicate whether an
6404 overflow occurred during the unsigned multiplication.</p>
6407 <p>The arguments (%a and %b) and the first element of the result structure may
6408 be of integer types of any bit width, but they must have the same bit
6409 width. The second element of the result structure must be of
6410 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6411 undergo unsigned multiplication.</p>
6414 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6415 an unsigned multiplication of the two arguments. They return a structure
6416 — the first element of which is the multiplication, and the second
6417 element of which is a bit specifying if the unsigned multiplication resulted
6422 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6423 %sum = extractvalue {i32, i1} %res, 0
6424 %obit = extractvalue {i32, i1} %res, 1
6425 br i1 %obit, label %overflow, label %normal
6430 <!-- ======================================================================= -->
6431 <div class="doc_subsection">
6432 <a name="int_debugger">Debugger Intrinsics</a>
6435 <div class="doc_text">
6437 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
6438 prefix), are described in
6439 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
6440 Level Debugging</a> document.</p>
6444 <!-- ======================================================================= -->
6445 <div class="doc_subsection">
6446 <a name="int_eh">Exception Handling Intrinsics</a>
6449 <div class="doc_text">
6451 <p>The LLVM exception handling intrinsics (which all start with
6452 <tt>llvm.eh.</tt> prefix), are described in
6453 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6454 Handling</a> document.</p>
6458 <!-- ======================================================================= -->
6459 <div class="doc_subsection">
6460 <a name="int_trampoline">Trampoline Intrinsic</a>
6463 <div class="doc_text">
6465 <p>This intrinsic makes it possible to excise one parameter, marked with
6466 the <tt>nest</tt> attribute, from a function. The result is a callable
6467 function pointer lacking the nest parameter - the caller does not need to
6468 provide a value for it. Instead, the value to use is stored in advance in a
6469 "trampoline", a block of memory usually allocated on the stack, which also
6470 contains code to splice the nest value into the argument list. This is used
6471 to implement the GCC nested function address extension.</p>
6473 <p>For example, if the function is
6474 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6475 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
6478 <div class="doc_code">
6480 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6481 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6482 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6483 %fp = bitcast i8* %p to i32 (i32, i32)*
6487 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6488 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6492 <!-- _______________________________________________________________________ -->
6493 <div class="doc_subsubsection">
6494 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6497 <div class="doc_text">
6501 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6505 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
6506 function pointer suitable for executing it.</p>
6509 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6510 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
6511 sufficiently aligned block of memory; this memory is written to by the
6512 intrinsic. Note that the size and the alignment are target-specific - LLVM
6513 currently provides no portable way of determining them, so a front-end that
6514 generates this intrinsic needs to have some target-specific knowledge.
6515 The <tt>func</tt> argument must hold a function bitcast to
6516 an <tt>i8*</tt>.</p>
6519 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
6520 dependent code, turning it into a function. A pointer to this function is
6521 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
6522 function pointer type</a> before being called. The new function's signature
6523 is the same as that of <tt>func</tt> with any arguments marked with
6524 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
6525 is allowed, and it must be of pointer type. Calling the new function is
6526 equivalent to calling <tt>func</tt> with the same argument list, but
6527 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
6528 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
6529 by <tt>tramp</tt> is modified, then the effect of any later call to the
6530 returned function pointer is undefined.</p>
6534 <!-- ======================================================================= -->
6535 <div class="doc_subsection">
6536 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6539 <div class="doc_text">
6541 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
6542 hardware constructs for atomic operations and memory synchronization. This
6543 provides an interface to the hardware, not an interface to the programmer. It
6544 is aimed at a low enough level to allow any programming models or APIs
6545 (Application Programming Interfaces) which need atomic behaviors to map
6546 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6547 hardware provides a "universal IR" for source languages, it also provides a
6548 starting point for developing a "universal" atomic operation and
6549 synchronization IR.</p>
6551 <p>These do <em>not</em> form an API such as high-level threading libraries,
6552 software transaction memory systems, atomic primitives, and intrinsic
6553 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6554 application libraries. The hardware interface provided by LLVM should allow
6555 a clean implementation of all of these APIs and parallel programming models.
6556 No one model or paradigm should be selected above others unless the hardware
6557 itself ubiquitously does so.</p>
6561 <!-- _______________________________________________________________________ -->
6562 <div class="doc_subsubsection">
6563 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6565 <div class="doc_text">
6568 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device> )
6572 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6573 specific pairs of memory access types.</p>
6576 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6577 The first four arguments enables a specific barrier as listed below. The
6578 fith argument specifies that the barrier applies to io or device or uncached
6582 <li><tt>ll</tt>: load-load barrier</li>
6583 <li><tt>ls</tt>: load-store barrier</li>
6584 <li><tt>sl</tt>: store-load barrier</li>
6585 <li><tt>ss</tt>: store-store barrier</li>
6586 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6590 <p>This intrinsic causes the system to enforce some ordering constraints upon
6591 the loads and stores of the program. This barrier does not
6592 indicate <em>when</em> any events will occur, it only enforces
6593 an <em>order</em> in which they occur. For any of the specified pairs of load
6594 and store operations (f.ex. load-load, or store-load), all of the first
6595 operations preceding the barrier will complete before any of the second
6596 operations succeeding the barrier begin. Specifically the semantics for each
6597 pairing is as follows:</p>
6600 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6601 after the barrier begins.</li>
6602 <li><tt>ls</tt>: All loads before the barrier must complete before any
6603 store after the barrier begins.</li>
6604 <li><tt>ss</tt>: All stores before the barrier must complete before any
6605 store after the barrier begins.</li>
6606 <li><tt>sl</tt>: All stores before the barrier must complete before any
6607 load after the barrier begins.</li>
6610 <p>These semantics are applied with a logical "and" behavior when more than one
6611 is enabled in a single memory barrier intrinsic.</p>
6613 <p>Backends may implement stronger barriers than those requested when they do
6614 not support as fine grained a barrier as requested. Some architectures do
6615 not need all types of barriers and on such architectures, these become
6620 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6621 %ptr = bitcast i8* %mallocP to i32*
6624 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6625 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6626 <i>; guarantee the above finishes</i>
6627 store i32 8, %ptr <i>; before this begins</i>
6632 <!-- _______________________________________________________________________ -->
6633 <div class="doc_subsubsection">
6634 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6637 <div class="doc_text">
6640 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6641 any integer bit width and for different address spaces. Not all targets
6642 support all bit widths however.</p>
6645 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6646 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6647 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6648 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6652 <p>This loads a value in memory and compares it to a given value. If they are
6653 equal, it stores a new value into the memory.</p>
6656 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
6657 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6658 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6659 this integer type. While any bit width integer may be used, targets may only
6660 lower representations they support in hardware.</p>
6663 <p>This entire intrinsic must be executed atomically. It first loads the value
6664 in memory pointed to by <tt>ptr</tt> and compares it with the
6665 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
6666 memory. The loaded value is yielded in all cases. This provides the
6667 equivalent of an atomic compare-and-swap operation within the SSA
6672 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6673 %ptr = bitcast i8* %mallocP to i32*
6676 %val1 = add i32 4, 4
6677 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6678 <i>; yields {i32}:result1 = 4</i>
6679 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6680 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6682 %val2 = add i32 1, 1
6683 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6684 <i>; yields {i32}:result2 = 8</i>
6685 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6687 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6692 <!-- _______________________________________________________________________ -->
6693 <div class="doc_subsubsection">
6694 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6696 <div class="doc_text">
6699 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6700 integer bit width. Not all targets support all bit widths however.</p>
6703 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6704 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6705 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6706 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6710 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6711 the value from memory. It then stores the value in <tt>val</tt> in the memory
6712 at <tt>ptr</tt>.</p>
6715 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
6716 the <tt>val</tt> argument and the result must be integers of the same bit
6717 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6718 integer type. The targets may only lower integer representations they
6722 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6723 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6724 equivalent of an atomic swap operation within the SSA framework.</p>
6728 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6729 %ptr = bitcast i8* %mallocP to i32*
6732 %val1 = add i32 4, 4
6733 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6734 <i>; yields {i32}:result1 = 4</i>
6735 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6736 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6738 %val2 = add i32 1, 1
6739 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6740 <i>; yields {i32}:result2 = 8</i>
6742 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6743 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6748 <!-- _______________________________________________________________________ -->
6749 <div class="doc_subsubsection">
6750 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6754 <div class="doc_text">
6757 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
6758 any integer bit width. Not all targets support all bit widths however.</p>
6761 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6762 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6763 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6764 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6768 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
6769 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6772 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6773 and the second an integer value. The result is also an integer value. These
6774 integer types can have any bit width, but they must all have the same bit
6775 width. The targets may only lower integer representations they support.</p>
6778 <p>This intrinsic does a series of operations atomically. It first loads the
6779 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6780 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
6784 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6785 %ptr = bitcast i8* %mallocP to i32*
6787 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6788 <i>; yields {i32}:result1 = 4</i>
6789 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6790 <i>; yields {i32}:result2 = 8</i>
6791 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6792 <i>; yields {i32}:result3 = 10</i>
6793 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6798 <!-- _______________________________________________________________________ -->
6799 <div class="doc_subsubsection">
6800 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6804 <div class="doc_text">
6807 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6808 any integer bit width and for different address spaces. Not all targets
6809 support all bit widths however.</p>
6812 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6813 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6814 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6815 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6819 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6820 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6823 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6824 and the second an integer value. The result is also an integer value. These
6825 integer types can have any bit width, but they must all have the same bit
6826 width. The targets may only lower integer representations they support.</p>
6829 <p>This intrinsic does a series of operations atomically. It first loads the
6830 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6831 result to <tt>ptr</tt>. It yields the original value stored
6832 at <tt>ptr</tt>.</p>
6836 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6837 %ptr = bitcast i8* %mallocP to i32*
6839 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6840 <i>; yields {i32}:result1 = 8</i>
6841 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6842 <i>; yields {i32}:result2 = 4</i>
6843 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6844 <i>; yields {i32}:result3 = 2</i>
6845 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6850 <!-- _______________________________________________________________________ -->
6851 <div class="doc_subsubsection">
6852 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6853 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6854 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6855 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6858 <div class="doc_text">
6861 <p>These are overloaded intrinsics. You can
6862 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
6863 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
6864 bit width and for different address spaces. Not all targets support all bit
6868 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6869 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6870 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6871 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6875 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6876 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6877 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6878 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6882 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6883 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6884 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6885 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6889 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6890 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6891 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6892 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6896 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6897 the value stored in memory at <tt>ptr</tt>. It yields the original value
6898 at <tt>ptr</tt>.</p>
6901 <p>These intrinsics take two arguments, the first a pointer to an integer value
6902 and the second an integer value. The result is also an integer value. These
6903 integer types can have any bit width, but they must all have the same bit
6904 width. The targets may only lower integer representations they support.</p>
6907 <p>These intrinsics does a series of operations atomically. They first load the
6908 value stored at <tt>ptr</tt>. They then do the bitwise
6909 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
6910 original value stored at <tt>ptr</tt>.</p>
6914 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6915 %ptr = bitcast i8* %mallocP to i32*
6916 store i32 0x0F0F, %ptr
6917 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6918 <i>; yields {i32}:result0 = 0x0F0F</i>
6919 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6920 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6921 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6922 <i>; yields {i32}:result2 = 0xF0</i>
6923 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6924 <i>; yields {i32}:result3 = FF</i>
6925 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6930 <!-- _______________________________________________________________________ -->
6931 <div class="doc_subsubsection">
6932 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6933 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6934 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6935 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6938 <div class="doc_text">
6941 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6942 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6943 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6944 address spaces. Not all targets support all bit widths however.</p>
6947 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6948 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6949 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6950 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6954 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6955 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6956 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6957 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6961 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6962 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6963 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6964 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6968 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6969 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6970 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6971 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6975 <p>These intrinsics takes the signed or unsigned minimum or maximum of
6976 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6977 original value at <tt>ptr</tt>.</p>
6980 <p>These intrinsics take two arguments, the first a pointer to an integer value
6981 and the second an integer value. The result is also an integer value. These
6982 integer types can have any bit width, but they must all have the same bit
6983 width. The targets may only lower integer representations they support.</p>
6986 <p>These intrinsics does a series of operations atomically. They first load the
6987 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
6988 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
6989 yield the original value stored at <tt>ptr</tt>.</p>
6993 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6994 %ptr = bitcast i8* %mallocP to i32*
6996 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6997 <i>; yields {i32}:result0 = 7</i>
6998 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6999 <i>; yields {i32}:result1 = -2</i>
7000 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
7001 <i>; yields {i32}:result2 = 8</i>
7002 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
7003 <i>; yields {i32}:result3 = 8</i>
7004 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
7010 <!-- ======================================================================= -->
7011 <div class="doc_subsection">
7012 <a name="int_memorymarkers">Memory Use Markers</a>
7015 <div class="doc_text">
7017 <p>This class of intrinsics exists to information about the lifetime of memory
7018 objects and ranges where variables are immutable.</p>
7022 <!-- _______________________________________________________________________ -->
7023 <div class="doc_subsubsection">
7024 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
7027 <div class="doc_text">
7031 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
7035 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
7036 object's lifetime.</p>
7039 <p>The first argument is a constant integer representing the size of the
7040 object, or -1 if it is variable sized. The second argument is a pointer to
7044 <p>This intrinsic indicates that before this point in the code, the value of the
7045 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7046 never be used and has an undefined value. A load from the pointer that
7047 precedes this intrinsic can be replaced with
7048 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
7052 <!-- _______________________________________________________________________ -->
7053 <div class="doc_subsubsection">
7054 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
7057 <div class="doc_text">
7061 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
7065 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
7066 object's lifetime.</p>
7069 <p>The first argument is a constant integer representing the size of the
7070 object, or -1 if it is variable sized. The second argument is a pointer to
7074 <p>This intrinsic indicates that after this point in the code, the value of the
7075 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7076 never be used and has an undefined value. Any stores into the memory object
7077 following this intrinsic may be removed as dead.
7081 <!-- _______________________________________________________________________ -->
7082 <div class="doc_subsubsection">
7083 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
7086 <div class="doc_text">
7090 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>) readonly
7094 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
7095 a memory object will not change.</p>
7098 <p>The first argument is a constant integer representing the size of the
7099 object, or -1 if it is variable sized. The second argument is a pointer to
7103 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
7104 the return value, the referenced memory location is constant and
7109 <!-- _______________________________________________________________________ -->
7110 <div class="doc_subsubsection">
7111 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
7114 <div class="doc_text">
7118 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7122 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
7123 a memory object are mutable.</p>
7126 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
7127 The second argument is a constant integer representing the size of the
7128 object, or -1 if it is variable sized and the third argument is a pointer
7132 <p>This intrinsic indicates that the memory is mutable again.</p>
7136 <!-- ======================================================================= -->
7137 <div class="doc_subsection">
7138 <a name="int_general">General Intrinsics</a>
7141 <div class="doc_text">
7143 <p>This class of intrinsics is designed to be generic and has no specific
7148 <!-- _______________________________________________________________________ -->
7149 <div class="doc_subsubsection">
7150 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
7153 <div class="doc_text">
7157 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
7161 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
7164 <p>The first argument is a pointer to a value, the second is a pointer to a
7165 global string, the third is a pointer to a global string which is the source
7166 file name, and the last argument is the line number.</p>
7169 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7170 This can be useful for special purpose optimizations that want to look for
7171 these annotations. These have no other defined use, they are ignored by code
7172 generation and optimization.</p>
7176 <!-- _______________________________________________________________________ -->
7177 <div class="doc_subsubsection">
7178 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
7181 <div class="doc_text">
7184 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
7185 any integer bit width.</p>
7188 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
7189 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
7190 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
7191 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
7192 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
7196 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
7199 <p>The first argument is an integer value (result of some expression), the
7200 second is a pointer to a global string, the third is a pointer to a global
7201 string which is the source file name, and the last argument is the line
7202 number. It returns the value of the first argument.</p>
7205 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7206 arbitrary strings. This can be useful for special purpose optimizations that
7207 want to look for these annotations. These have no other defined use, they
7208 are ignored by code generation and optimization.</p>
7212 <!-- _______________________________________________________________________ -->
7213 <div class="doc_subsubsection">
7214 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
7217 <div class="doc_text">
7221 declare void @llvm.trap()
7225 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
7231 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7232 target does not have a trap instruction, this intrinsic will be lowered to
7233 the call of the <tt>abort()</tt> function.</p>
7237 <!-- _______________________________________________________________________ -->
7238 <div class="doc_subsubsection">
7239 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
7242 <div class="doc_text">
7246 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
7250 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
7251 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
7252 ensure that it is placed on the stack before local variables.</p>
7255 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
7256 arguments. The first argument is the value loaded from the stack
7257 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
7258 that has enough space to hold the value of the guard.</p>
7261 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7262 the <tt>AllocaInst</tt> stack slot to be before local variables on the
7263 stack. This is to ensure that if a local variable on the stack is
7264 overwritten, it will destroy the value of the guard. When the function exits,
7265 the guard on the stack is checked against the original guard. If they're
7266 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
7271 <!-- *********************************************************************** -->
7274 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
7275 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
7276 <a href="http://validator.w3.org/check/referer"><img
7277 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
7279 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
7280 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
7281 Last modified: $Date$