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">'<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>
53 <li><a href="#typesystem">Type System</a>
55 <li><a href="#t_classifications">Type Classifications</a></li>
56 <li><a href="#t_primitive">Primitive Types</a>
58 <li><a href="#t_floating">Floating Point Types</a></li>
59 <li><a href="#t_void">Void Type</a></li>
60 <li><a href="#t_label">Label Type</a></li>
61 <li><a href="#t_metadata">Metadata Type</a></li>
64 <li><a href="#t_derived">Derived Types</a>
66 <li><a href="#t_integer">Integer Type</a></li>
67 <li><a href="#t_array">Array Type</a></li>
68 <li><a href="#t_function">Function Type</a></li>
69 <li><a href="#t_pointer">Pointer Type</a></li>
70 <li><a href="#t_struct">Structure Type</a></li>
71 <li><a href="#t_pstruct">Packed Structure Type</a></li>
72 <li><a href="#t_vector">Vector Type</a></li>
73 <li><a href="#t_opaque">Opaque Type</a></li>
76 <li><a href="#t_uprefs">Type Up-references</a></li>
79 <li><a href="#constants">Constants</a>
81 <li><a href="#simpleconstants">Simple Constants</a></li>
82 <li><a href="#complexconstants">Complex Constants</a></li>
83 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
84 <li><a href="#undefvalues">Undefined Values</a></li>
85 <li><a href="#constantexprs">Constant Expressions</a></li>
86 <li><a href="#metadata">Embedded Metadata</a></li>
89 <li><a href="#othervalues">Other Values</a>
91 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
94 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
96 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
97 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
98 Global Variable</a></li>
99 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
100 Global Variable</a></li>
101 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
102 Global Variable</a></li>
105 <li><a href="#instref">Instruction Reference</a>
107 <li><a href="#terminators">Terminator Instructions</a>
109 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
110 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
111 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
112 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
113 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
114 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
117 <li><a href="#binaryops">Binary Operations</a>
119 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
120 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
121 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
122 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
123 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
124 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
125 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
126 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
127 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
128 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
129 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
130 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
133 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
135 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
136 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
137 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
138 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
139 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
140 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
143 <li><a href="#vectorops">Vector Operations</a>
145 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
146 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
147 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
150 <li><a href="#aggregateops">Aggregate Operations</a>
152 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
153 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
156 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
158 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
159 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
160 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
161 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
162 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
163 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
166 <li><a href="#convertops">Conversion Operations</a>
168 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
169 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
170 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
171 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
172 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
173 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
174 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
175 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
176 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
177 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
178 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
179 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
182 <li><a href="#otherops">Other Operations</a>
184 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
185 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
186 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
187 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
188 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
189 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
194 <li><a href="#intrinsics">Intrinsic Functions</a>
196 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
198 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
199 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
200 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
203 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
205 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
206 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
207 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
210 <li><a href="#int_codegen">Code Generator Intrinsics</a>
212 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
213 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
214 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
215 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
216 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
217 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
218 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
221 <li><a href="#int_libc">Standard C Library Intrinsics</a>
223 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
224 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
225 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
226 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
227 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
228 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
229 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
230 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
233 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
235 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
236 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
237 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
238 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
241 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
243 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
244 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
245 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
246 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
247 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
248 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
251 <li><a href="#int_debugger">Debugger intrinsics</a></li>
252 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
253 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
255 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
258 <li><a href="#int_atomics">Atomic intrinsics</a>
260 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
261 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
262 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
263 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
264 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
265 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
266 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
267 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
268 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
269 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
270 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
271 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
272 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
275 <li><a href="#int_general">General intrinsics</a>
277 <li><a href="#int_var_annotation">
278 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
279 <li><a href="#int_annotation">
280 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
281 <li><a href="#int_trap">
282 '<tt>llvm.trap</tt>' Intrinsic</a></li>
283 <li><a href="#int_stackprotector">
284 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
291 <div class="doc_author">
292 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
293 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
296 <!-- *********************************************************************** -->
297 <div class="doc_section"> <a name="abstract">Abstract </a></div>
298 <!-- *********************************************************************** -->
300 <div class="doc_text">
302 <p>This document is a reference manual for the LLVM assembly language. LLVM is
303 a Static Single Assignment (SSA) based representation that provides type
304 safety, low-level operations, flexibility, and the capability of representing
305 'all' high-level languages cleanly. It is the common code representation
306 used throughout all phases of the LLVM compilation strategy.</p>
310 <!-- *********************************************************************** -->
311 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
312 <!-- *********************************************************************** -->
314 <div class="doc_text">
316 <p>The LLVM code representation is designed to be used in three different forms:
317 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
318 for fast loading by a Just-In-Time compiler), and as a human readable
319 assembly language representation. This allows LLVM to provide a powerful
320 intermediate representation for efficient compiler transformations and
321 analysis, while providing a natural means to debug and visualize the
322 transformations. The three different forms of LLVM are all equivalent. This
323 document describes the human readable representation and notation.</p>
325 <p>The LLVM representation aims to be light-weight and low-level while being
326 expressive, typed, and extensible at the same time. It aims to be a
327 "universal IR" of sorts, by being at a low enough level that high-level ideas
328 may be cleanly mapped to it (similar to how microprocessors are "universal
329 IR's", allowing many source languages to be mapped to them). By providing
330 type information, LLVM can be used as the target of optimizations: for
331 example, through pointer analysis, it can be proven that a C automatic
332 variable is never accessed outside of the current function... allowing it to
333 be promoted to a simple SSA value instead of a memory location.</p>
337 <!-- _______________________________________________________________________ -->
338 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
340 <div class="doc_text">
342 <p>It is important to note that this document describes 'well formed' LLVM
343 assembly language. There is a difference between what the parser accepts and
344 what is considered 'well formed'. For example, the following instruction is
345 syntactically okay, but not well formed:</p>
347 <div class="doc_code">
349 %x = <a href="#i_add">add</a> i32 1, %x
353 <p>...because the definition of <tt>%x</tt> does not dominate all of its
354 uses. The LLVM infrastructure provides a verification pass that may be used
355 to verify that an LLVM module is well formed. This pass is automatically run
356 by the parser after parsing input assembly and by the optimizer before it
357 outputs bitcode. The violations pointed out by the verifier pass indicate
358 bugs in transformation passes or input to the parser.</p>
362 <!-- Describe the typesetting conventions here. -->
364 <!-- *********************************************************************** -->
365 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
366 <!-- *********************************************************************** -->
368 <div class="doc_text">
370 <p>LLVM identifiers come in two basic types: global and local. Global
371 identifiers (functions, global variables) begin with the <tt>'@'</tt>
372 character. Local identifiers (register names, types) begin with
373 the <tt>'%'</tt> character. Additionally, there are three different formats
374 for identifiers, for different purposes:</p>
377 <li>Named values are represented as a string of characters with their prefix.
378 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
379 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
380 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
381 other characters in their names can be surrounded with quotes. Special
382 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
383 ASCII code for the character in hexadecimal. In this way, any character
384 can be used in a name value, even quotes themselves.</li>
386 <li>Unnamed values are represented as an unsigned numeric value with their
387 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
389 <li>Constants, which are described in a <a href="#constants">section about
390 constants</a>, below.</li>
393 <p>LLVM requires that values start with a prefix for two reasons: Compilers
394 don't need to worry about name clashes with reserved words, and the set of
395 reserved words may be expanded in the future without penalty. Additionally,
396 unnamed identifiers allow a compiler to quickly come up with a temporary
397 variable without having to avoid symbol table conflicts.</p>
399 <p>Reserved words in LLVM are very similar to reserved words in other
400 languages. There are keywords for different opcodes
401 ('<tt><a href="#i_add">add</a></tt>',
402 '<tt><a href="#i_bitcast">bitcast</a></tt>',
403 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
404 ('<tt><a href="#t_void">void</a></tt>',
405 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
406 reserved words cannot conflict with variable names, because none of them
407 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
409 <p>Here is an example of LLVM code to multiply the integer variable
410 '<tt>%X</tt>' by 8:</p>
414 <div class="doc_code">
416 %result = <a href="#i_mul">mul</a> i32 %X, 8
420 <p>After strength reduction:</p>
422 <div class="doc_code">
424 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
428 <p>And the hard way:</p>
430 <div class="doc_code">
432 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
433 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
434 %result = <a href="#i_add">add</a> i32 %1, %1
438 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
439 lexical features of LLVM:</p>
442 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
445 <li>Unnamed temporaries are created when the result of a computation is not
446 assigned to a named value.</li>
448 <li>Unnamed temporaries are numbered sequentially</li>
451 <p>...and it also shows a convention that we follow in this document. When
452 demonstrating instructions, we will follow an instruction with a comment that
453 defines the type and name of value produced. Comments are shown in italic
458 <!-- *********************************************************************** -->
459 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
460 <!-- *********************************************************************** -->
462 <!-- ======================================================================= -->
463 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
466 <div class="doc_text">
468 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
469 of the input programs. Each module consists of functions, global variables,
470 and symbol table entries. Modules may be combined together with the LLVM
471 linker, which merges function (and global variable) definitions, resolves
472 forward declarations, and merges symbol table entries. Here is an example of
473 the "hello world" module:</p>
475 <div class="doc_code">
476 <pre><i>; Declare the string constant as a global constant...</i>
477 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
478 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
480 <i>; External declaration of the puts function</i>
481 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
483 <i>; Definition of main function</i>
484 define i32 @main() { <i>; i32()* </i>
485 <i>; Convert [13 x i8]* to i8 *...</i>
487 href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
489 <i>; Call puts function to write out the string to stdout...</i>
491 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
493 href="#i_ret">ret</a> i32 0<br>}<br>
497 <p>This example is made up of a <a href="#globalvars">global variable</a> named
498 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function, and
499 a <a href="#functionstructure">function definition</a> for
502 <p>In general, a module is made up of a list of global values, where both
503 functions and global variables are global values. Global values are
504 represented by a pointer to a memory location (in this case, a pointer to an
505 array of char, and a pointer to a function), and have one of the
506 following <a href="#linkage">linkage types</a>.</p>
510 <!-- ======================================================================= -->
511 <div class="doc_subsection">
512 <a name="linkage">Linkage Types</a>
515 <div class="doc_text">
517 <p>All Global Variables and Functions have one of the following types of
521 <dt><tt><b><a name="linkage_private">private</a></b></tt>: </dt>
522 <dd>Global values with private linkage are only directly accessible by objects
523 in the current module. In particular, linking code into a module with an
524 private global value may cause the private to be renamed as necessary to
525 avoid collisions. Because the symbol is private to the module, all
526 references can be updated. This doesn't show up in any symbol table in the
529 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt>: </dt>
530 <dd>Similar to private, but the symbol is passed through the assembler and
531 removed by the linker after evaluation.</dd>
533 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
534 <dd>Similar to private, but the value shows as a local symbol
535 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
536 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
538 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt>: </dt>
539 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
540 into the object file corresponding to the LLVM module. They exist to
541 allow inlining and other optimizations to take place given knowledge of
542 the definition of the global, which is known to be somewhere outside the
543 module. Globals with <tt>available_externally</tt> linkage are allowed to
544 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
545 This linkage type is only allowed on definitions, not declarations.</dd>
547 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
548 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
549 the same name when linkage occurs. This is typically used to implement
550 inline functions, templates, or other code which must be generated in each
551 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
552 allowed to be discarded.</dd>
554 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
555 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
556 linkage, except that unreferenced <tt>common</tt> globals may not be
557 discarded. This is used for globals that may be emitted in multiple
558 translation units, but that are not guaranteed to be emitted into every
559 translation unit that uses them. One example of this is tentative
560 definitions in C, such as "<tt>int X;</tt>" at global scope.</dd>
562 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
563 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
564 that some targets may choose to emit different assembly sequences for them
565 for target-dependent reasons. This is used for globals that are declared
566 "weak" in C source code.</dd>
568 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
569 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
570 pointer to array type. When two global variables with appending linkage
571 are linked together, the two global arrays are appended together. This is
572 the LLVM, typesafe, equivalent of having the system linker append together
573 "sections" with identical names when .o files are linked.</dd>
575 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
576 <dd>The semantics of this linkage follow the ELF object file model: the symbol
577 is weak until linked, if not linked, the symbol becomes null instead of
578 being an undefined reference.</dd>
580 <dt><tt><b><a name="linkage_linkonce">linkonce_odr</a></b></tt>: </dt>
581 <dt><tt><b><a name="linkage_weak">weak_odr</a></b></tt>: </dt>
582 <dd>Some languages allow differing globals to be merged, such as two functions
583 with different semantics. Other languages, such as <tt>C++</tt>, ensure
584 that only equivalent globals are ever merged (the "one definition rule" -
585 "ODR"). Such languages can use the <tt>linkonce_odr</tt>
586 and <tt>weak_odr</tt> linkage types to indicate that the global will only
587 be merged with equivalent globals. These linkage types are otherwise the
588 same as their non-<tt>odr</tt> versions.</dd>
590 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
591 <dd>If none of the above identifiers are used, the global is externally
592 visible, meaning that it participates in linkage and can be used to
593 resolve external symbol references.</dd>
596 <p>The next two types of linkage are targeted for Microsoft Windows platform
597 only. They are designed to support importing (exporting) symbols from (to)
598 DLLs (Dynamic Link Libraries).</p>
601 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
602 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
603 or variable via a global pointer to a pointer that is set up by the DLL
604 exporting the symbol. On Microsoft Windows targets, the pointer name is
605 formed by combining <code>__imp_</code> and the function or variable
608 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
609 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
610 pointer to a pointer in a DLL, so that it can be referenced with the
611 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
612 name is formed by combining <code>__imp_</code> and the function or
616 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
617 another module defined a "<tt>.LC0</tt>" variable and was linked with this
618 one, one of the two would be renamed, preventing a collision. Since
619 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
620 declarations), they are accessible outside of the current module.</p>
622 <p>It is illegal for a function <i>declaration</i> to have any linkage type
623 other than "externally visible", <tt>dllimport</tt>
624 or <tt>extern_weak</tt>.</p>
626 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
627 or <tt>weak_odr</tt> linkages.</p>
631 <!-- ======================================================================= -->
632 <div class="doc_subsection">
633 <a name="callingconv">Calling Conventions</a>
636 <div class="doc_text">
638 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
639 and <a href="#i_invoke">invokes</a> can all have an optional calling
640 convention specified for the call. The calling convention of any pair of
641 dynamic caller/callee must match, or the behavior of the program is
642 undefined. The following calling conventions are supported by LLVM, and more
643 may be added in the future:</p>
646 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
647 <dd>This calling convention (the default if no other calling convention is
648 specified) matches the target C calling conventions. This calling
649 convention supports varargs function calls and tolerates some mismatch in
650 the declared prototype and implemented declaration of the function (as
653 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
654 <dd>This calling convention attempts to make calls as fast as possible
655 (e.g. by passing things in registers). This calling convention allows the
656 target to use whatever tricks it wants to produce fast code for the
657 target, without having to conform to an externally specified ABI
658 (Application Binary Interface). Implementations of this convention should
659 allow arbitrary <a href="CodeGenerator.html#tailcallopt">tail call
660 optimization</a> to be supported. This calling convention does not
661 support varargs and requires the prototype of all callees to exactly match
662 the prototype of the function definition.</dd>
664 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
665 <dd>This calling convention attempts to make code in the caller as efficient
666 as possible under the assumption that the call is not commonly executed.
667 As such, these calls often preserve all registers so that the call does
668 not break any live ranges in the caller side. This calling convention
669 does not support varargs and requires the prototype of all callees to
670 exactly match the prototype of the function definition.</dd>
672 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
673 <dd>Any calling convention may be specified by number, allowing
674 target-specific calling conventions to be used. Target specific calling
675 conventions start at 64.</dd>
678 <p>More calling conventions can be added/defined on an as-needed basis, to
679 support Pascal conventions or any other well-known target-independent
684 <!-- ======================================================================= -->
685 <div class="doc_subsection">
686 <a name="visibility">Visibility Styles</a>
689 <div class="doc_text">
691 <p>All Global Variables and Functions have one of the following visibility
695 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
696 <dd>On targets that use the ELF object file format, default visibility means
697 that the declaration is visible to other modules and, in shared libraries,
698 means that the declared entity may be overridden. On Darwin, default
699 visibility means that the declaration is visible to other modules. Default
700 visibility corresponds to "external linkage" in the language.</dd>
702 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
703 <dd>Two declarations of an object with hidden visibility refer to the same
704 object if they are in the same shared object. Usually, hidden visibility
705 indicates that the symbol will not be placed into the dynamic symbol
706 table, so no other module (executable or shared library) can reference it
709 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
710 <dd>On ELF, protected visibility indicates that the symbol will be placed in
711 the dynamic symbol table, but that references within the defining module
712 will bind to the local symbol. That is, the symbol cannot be overridden by
718 <!-- ======================================================================= -->
719 <div class="doc_subsection">
720 <a name="namedtypes">Named Types</a>
723 <div class="doc_text">
725 <p>LLVM IR allows you to specify name aliases for certain types. This can make
726 it easier to read the IR and make the IR more condensed (particularly when
727 recursive types are involved). An example of a name specification is:</p>
729 <div class="doc_code">
731 %mytype = type { %mytype*, i32 }
735 <p>You may give a name to any <a href="#typesystem">type</a> except
736 "<a href="t_void">void</a>". Type name aliases may be used anywhere a type
737 is expected with the syntax "%mytype".</p>
739 <p>Note that type names are aliases for the structural type that they indicate,
740 and that you can therefore specify multiple names for the same type. This
741 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
742 uses structural typing, the name is not part of the type. When printing out
743 LLVM IR, the printer will pick <em>one name</em> to render all types of a
744 particular shape. This means that if you have code where two different
745 source types end up having the same LLVM type, that the dumper will sometimes
746 print the "wrong" or unexpected type. This is an important design point and
747 isn't going to change.</p>
751 <!-- ======================================================================= -->
752 <div class="doc_subsection">
753 <a name="globalvars">Global Variables</a>
756 <div class="doc_text">
758 <p>Global variables define regions of memory allocated at compilation time
759 instead of run-time. Global variables may optionally be initialized, may
760 have an explicit section to be placed in, and may have an optional explicit
761 alignment specified. A variable may be defined as "thread_local", which
762 means that it will not be shared by threads (each thread will have a
763 separated copy of the variable). A variable may be defined as a global
764 "constant," which indicates that the contents of the variable
765 will <b>never</b> be modified (enabling better optimization, allowing the
766 global data to be placed in the read-only section of an executable, etc).
767 Note that variables that need runtime initialization cannot be marked
768 "constant" as there is a store to the variable.</p>
770 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
771 constant, even if the final definition of the global is not. This capability
772 can be used to enable slightly better optimization of the program, but
773 requires the language definition to guarantee that optimizations based on the
774 'constantness' are valid for the translation units that do not include the
777 <p>As SSA values, global variables define pointer values that are in scope
778 (i.e. they dominate) all basic blocks in the program. Global variables
779 always define a pointer to their "content" type because they describe a
780 region of memory, and all memory objects in LLVM are accessed through
783 <p>A global variable may be declared to reside in a target-specific numbered
784 address space. For targets that support them, address spaces may affect how
785 optimizations are performed and/or what target instructions are used to
786 access the variable. The default address space is zero. The address space
787 qualifier must precede any other attributes.</p>
789 <p>LLVM allows an explicit section to be specified for globals. If the target
790 supports it, it will emit globals to the section specified.</p>
792 <p>An explicit alignment may be specified for a global. If not present, or if
793 the alignment is set to zero, the alignment of the global is set by the
794 target to whatever it feels convenient. If an explicit alignment is
795 specified, the global is forced to have at least that much alignment. All
796 alignments must be a power of 2.</p>
798 <p>For example, the following defines a global in a numbered address space with
799 an initializer, section, and alignment:</p>
801 <div class="doc_code">
803 @G = addrspace(5) constant float 1.0, section "foo", align 4
810 <!-- ======================================================================= -->
811 <div class="doc_subsection">
812 <a name="functionstructure">Functions</a>
815 <div class="doc_text">
817 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord, an
818 optional <a href="#linkage">linkage type</a>, an optional
819 <a href="#visibility">visibility style</a>, an optional
820 <a href="#callingconv">calling convention</a>, a return type, an optional
821 <a href="#paramattrs">parameter attribute</a> for the return type, a function
822 name, a (possibly empty) argument list (each with optional
823 <a href="#paramattrs">parameter attributes</a>), optional
824 <a href="#fnattrs">function attributes</a>, an optional section, an optional
825 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
826 curly brace, a list of basic blocks, and a closing curly brace.</p>
828 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
829 optional <a href="#linkage">linkage type</a>, an optional
830 <a href="#visibility">visibility style</a>, an optional
831 <a href="#callingconv">calling convention</a>, a return type, an optional
832 <a href="#paramattrs">parameter attribute</a> for the return type, a function
833 name, a possibly empty list of arguments, an optional alignment, and an
834 optional <a href="#gc">garbage collector name</a>.</p>
836 <p>A function definition contains a list of basic blocks, forming the CFG
837 (Control Flow Graph) for the function. Each basic block may optionally start
838 with a label (giving the basic block a symbol table entry), contains a list
839 of instructions, and ends with a <a href="#terminators">terminator</a>
840 instruction (such as a branch or function return).</p>
842 <p>The first basic block in a function is special in two ways: it is immediately
843 executed on entrance to the function, and it is not allowed to have
844 predecessor basic blocks (i.e. there can not be any branches to the entry
845 block of a function). Because the block can have no predecessors, it also
846 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
848 <p>LLVM allows an explicit section to be specified for functions. If the target
849 supports it, it will emit functions to the section specified.</p>
851 <p>An explicit alignment may be specified for a function. If not present, or if
852 the alignment is set to zero, the alignment of the function is set by the
853 target to whatever it feels convenient. If an explicit alignment is
854 specified, the function is forced to have at least that much alignment. All
855 alignments must be a power of 2.</p>
858 <div class="doc_code">
860 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
861 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
862 <ResultType> @<FunctionName> ([argument list])
863 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
864 [<a href="#gc">gc</a>] { ... }
870 <!-- ======================================================================= -->
871 <div class="doc_subsection">
872 <a name="aliasstructure">Aliases</a>
875 <div class="doc_text">
877 <p>Aliases act as "second name" for the aliasee value (which can be either
878 function, global variable, another alias or bitcast of global value). Aliases
879 may have an optional <a href="#linkage">linkage type</a>, and an
880 optional <a href="#visibility">visibility style</a>.</p>
883 <div class="doc_code">
885 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
891 <!-- ======================================================================= -->
892 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
894 <div class="doc_text">
896 <p>The return type and each parameter of a function type may have a set of
897 <i>parameter attributes</i> associated with them. Parameter attributes are
898 used to communicate additional information about the result or parameters of
899 a function. Parameter attributes are considered to be part of the function,
900 not of the function type, so functions with different parameter attributes
901 can have the same function type.</p>
903 <p>Parameter attributes are simple keywords that follow the type specified. If
904 multiple parameter attributes are needed, they are space separated. For
907 <div class="doc_code">
909 declare i32 @printf(i8* noalias nocapture, ...)
910 declare i32 @atoi(i8 zeroext)
911 declare signext i8 @returns_signed_char()
915 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
916 <tt>readonly</tt>) come immediately after the argument list.</p>
918 <p>Currently, only the following parameter attributes are defined:</p>
921 <dt><tt>zeroext</tt></dt>
922 <dd>This indicates to the code generator that the parameter or return value
923 should be zero-extended to a 32-bit value by the caller (for a parameter)
924 or the callee (for a return value).</dd>
926 <dt><tt>signext</tt></dt>
927 <dd>This indicates to the code generator that the parameter or return value
928 should be sign-extended to a 32-bit value by the caller (for a parameter)
929 or the callee (for a return value).</dd>
931 <dt><tt>inreg</tt></dt>
932 <dd>This indicates that this parameter or return value should be treated in a
933 special target-dependent fashion during while emitting code for a function
934 call or return (usually, by putting it in a register as opposed to memory,
935 though some targets use it to distinguish between two different kinds of
936 registers). Use of this attribute is target-specific.</dd>
938 <dt><tt><a name="byval">byval</a></tt></dt>
939 <dd>This indicates that the pointer parameter should really be passed by value
940 to the function. The attribute implies that a hidden copy of the pointee
941 is made between the caller and the callee, so the callee is unable to
942 modify the value in the callee. This attribute is only valid on LLVM
943 pointer arguments. It is generally used to pass structs and arrays by
944 value, but is also valid on pointers to scalars. The copy is considered
945 to belong to the caller not the callee (for example,
946 <tt><a href="#readonly">readonly</a></tt> functions should not write to
947 <tt>byval</tt> parameters). This is not a valid attribute for return
948 values. The byval attribute also supports specifying an alignment with
949 the align attribute. This has a target-specific effect on the code
950 generator that usually indicates a desired alignment for the synthesized
953 <dt><tt>sret</tt></dt>
954 <dd>This indicates that the pointer parameter specifies the address of a
955 structure that is the return value of the function in the source program.
956 This pointer must be guaranteed by the caller to be valid: loads and
957 stores to the structure may be assumed by the callee to not to trap. This
958 may only be applied to the first parameter. This is not a valid attribute
959 for return values. </dd>
961 <dt><tt>noalias</tt></dt>
962 <dd>This indicates that the pointer does not alias any global or any other
963 parameter. The caller is responsible for ensuring that this is the
964 case. On a function return value, <tt>noalias</tt> additionally indicates
965 that the pointer does not alias any other pointers visible to the
966 caller. For further details, please see the discussion of the NoAlias
968 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
971 <dt><tt>nocapture</tt></dt>
972 <dd>This indicates that the callee does not make any copies of the pointer
973 that outlive the callee itself. This is not a valid attribute for return
976 <dt><tt>nest</tt></dt>
977 <dd>This indicates that the pointer parameter can be excised using the
978 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
979 attribute for return values.</dd>
984 <!-- ======================================================================= -->
985 <div class="doc_subsection">
986 <a name="gc">Garbage Collector Names</a>
989 <div class="doc_text">
991 <p>Each function may specify a garbage collector name, which is simply a
994 <div class="doc_code">
996 define void @f() gc "name" { ...
1000 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1001 collector which will cause the compiler to alter its output in order to
1002 support the named garbage collection algorithm.</p>
1006 <!-- ======================================================================= -->
1007 <div class="doc_subsection">
1008 <a name="fnattrs">Function Attributes</a>
1011 <div class="doc_text">
1013 <p>Function attributes are set to communicate additional information about a
1014 function. Function attributes are considered to be part of the function, not
1015 of the function type, so functions with different parameter attributes can
1016 have the same function type.</p>
1018 <p>Function attributes are simple keywords that follow the type specified. If
1019 multiple attributes are needed, they are space separated. For example:</p>
1021 <div class="doc_code">
1023 define void @f() noinline { ... }
1024 define void @f() alwaysinline { ... }
1025 define void @f() alwaysinline optsize { ... }
1026 define void @f() optsize
1031 <dt><tt>alwaysinline</tt></dt>
1032 <dd>This attribute indicates that the inliner should attempt to inline this
1033 function into callers whenever possible, ignoring any active inlining size
1034 threshold for this caller.</dd>
1036 <dt><tt>noinline</tt></dt>
1037 <dd>This attribute indicates that the inliner should never inline this
1038 function in any situation. This attribute may not be used together with
1039 the <tt>alwaysinline</tt> attribute.</dd>
1041 <dt><tt>optsize</tt></dt>
1042 <dd>This attribute suggests that optimization passes and code generator passes
1043 make choices that keep the code size of this function low, and otherwise
1044 do optimizations specifically to reduce code size.</dd>
1046 <dt><tt>noreturn</tt></dt>
1047 <dd>This function attribute indicates that the function never returns
1048 normally. This produces undefined behavior at runtime if the function
1049 ever does dynamically return.</dd>
1051 <dt><tt>nounwind</tt></dt>
1052 <dd>This function attribute indicates that the function never returns with an
1053 unwind or exceptional control flow. If the function does unwind, its
1054 runtime behavior is undefined.</dd>
1056 <dt><tt>readnone</tt></dt>
1057 <dd>This attribute indicates that the function computes its result (or decides
1058 to unwind an exception) based strictly on its arguments, without
1059 dereferencing any pointer arguments or otherwise accessing any mutable
1060 state (e.g. memory, control registers, etc) visible to caller functions.
1061 It does not write through any pointer arguments
1062 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1063 changes any state visible to callers. This means that it cannot unwind
1064 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1065 could use the <tt>unwind</tt> instruction.</dd>
1067 <dt><tt><a name="readonly">readonly</a></tt></dt>
1068 <dd>This attribute indicates that the function does not write through any
1069 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1070 arguments) or otherwise modify any state (e.g. memory, control registers,
1071 etc) visible to caller functions. It may dereference pointer arguments
1072 and read state that may be set in the caller. A readonly function always
1073 returns the same value (or unwinds an exception identically) when called
1074 with the same set of arguments and global state. It cannot unwind an
1075 exception by calling the <tt>C++</tt> exception throwing methods, but may
1076 use the <tt>unwind</tt> instruction.</dd>
1078 <dt><tt><a name="ssp">ssp</a></tt></dt>
1079 <dd>This attribute indicates that the function should emit a stack smashing
1080 protector. It is in the form of a "canary"—a random value placed on
1081 the stack before the local variables that's checked upon return from the
1082 function to see if it has been overwritten. A heuristic is used to
1083 determine if a function needs stack protectors or not.<br>
1085 If a function that has an <tt>ssp</tt> attribute is inlined into a
1086 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1087 function will have an <tt>ssp</tt> attribute.</dd>
1089 <dt><tt>sspreq</tt></dt>
1090 <dd>This attribute indicates that the function should <em>always</em> emit a
1091 stack smashing protector. This overrides
1092 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1094 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1095 function that doesn't have an <tt>sspreq</tt> attribute or which has
1096 an <tt>ssp</tt> attribute, then the resulting function will have
1097 an <tt>sspreq</tt> attribute.</dd>
1099 <dt><tt>noredzone</tt></dt>
1100 <dd>This attribute indicates that the code generator should not use a red
1101 zone, even if the target-specific ABI normally permits it.</dd>
1103 <dt><tt>noimplicitfloat</tt></dt>
1104 <dd>This attributes disables implicit floating point instructions.</dd>
1106 <dt><tt>naked</tt></dt>
1107 <dd>This attribute disables prologue / epilogue emission for the function.
1108 This can have very system-specific consequences.</dd>
1113 <!-- ======================================================================= -->
1114 <div class="doc_subsection">
1115 <a name="moduleasm">Module-Level Inline Assembly</a>
1118 <div class="doc_text">
1120 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1121 the GCC "file scope inline asm" blocks. These blocks are internally
1122 concatenated by LLVM and treated as a single unit, but may be separated in
1123 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1125 <div class="doc_code">
1127 module asm "inline asm code goes here"
1128 module asm "more can go here"
1132 <p>The strings can contain any character by escaping non-printable characters.
1133 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1136 <p>The inline asm code is simply printed to the machine code .s file when
1137 assembly code is generated.</p>
1141 <!-- ======================================================================= -->
1142 <div class="doc_subsection">
1143 <a name="datalayout">Data Layout</a>
1146 <div class="doc_text">
1148 <p>A module may specify a target specific data layout string that specifies how
1149 data is to be laid out in memory. The syntax for the data layout is
1152 <div class="doc_code">
1154 target datalayout = "<i>layout specification</i>"
1158 <p>The <i>layout specification</i> consists of a list of specifications
1159 separated by the minus sign character ('-'). Each specification starts with
1160 a letter and may include other information after the letter to define some
1161 aspect of the data layout. The specifications accepted are as follows:</p>
1165 <dd>Specifies that the target lays out data in big-endian form. That is, the
1166 bits with the most significance have the lowest address location.</dd>
1169 <dd>Specifies that the target lays out data in little-endian form. That is,
1170 the bits with the least significance have the lowest address
1173 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1174 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1175 <i>preferred</i> alignments. All sizes are in bits. Specifying
1176 the <i>pref</i> alignment is optional. If omitted, the
1177 preceding <tt>:</tt> should be omitted too.</dd>
1179 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1180 <dd>This specifies the alignment for an integer type of a given bit
1181 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1183 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1184 <dd>This specifies the alignment for a vector type of a given bit
1187 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1188 <dd>This specifies the alignment for a floating point type of a given bit
1189 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1192 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1193 <dd>This specifies the alignment for an aggregate type of a given bit
1196 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1197 <dd>This specifies the alignment for a stack object of a given bit
1201 <p>When constructing the data layout for a given target, LLVM starts with a
1202 default set of specifications which are then (possibly) overriden by the
1203 specifications in the <tt>datalayout</tt> keyword. The default specifications
1204 are given in this list:</p>
1207 <li><tt>E</tt> - big endian</li>
1208 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1209 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1210 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1211 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1212 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1213 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1214 alignment of 64-bits</li>
1215 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1216 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1217 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1218 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1219 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1220 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1223 <p>When LLVM is determining the alignment for a given type, it uses the
1224 following rules:</p>
1227 <li>If the type sought is an exact match for one of the specifications, that
1228 specification is used.</li>
1230 <li>If no match is found, and the type sought is an integer type, then the
1231 smallest integer type that is larger than the bitwidth of the sought type
1232 is used. If none of the specifications are larger than the bitwidth then
1233 the the largest integer type is used. For example, given the default
1234 specifications above, the i7 type will use the alignment of i8 (next
1235 largest) while both i65 and i256 will use the alignment of i64 (largest
1238 <li>If no match is found, and the type sought is a vector type, then the
1239 largest vector type that is smaller than the sought vector type will be
1240 used as a fall back. This happens because <128 x double> can be
1241 implemented in terms of 64 <2 x double>, for example.</li>
1246 <!-- *********************************************************************** -->
1247 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1248 <!-- *********************************************************************** -->
1250 <div class="doc_text">
1252 <p>The LLVM type system is one of the most important features of the
1253 intermediate representation. Being typed enables a number of optimizations
1254 to be performed on the intermediate representation directly, without having
1255 to do extra analyses on the side before the transformation. A strong type
1256 system makes it easier to read the generated code and enables novel analyses
1257 and transformations that are not feasible to perform on normal three address
1258 code representations.</p>
1262 <!-- ======================================================================= -->
1263 <div class="doc_subsection"> <a name="t_classifications">Type
1264 Classifications</a> </div>
1266 <div class="doc_text">
1268 <p>The types fall into a few useful classifications:</p>
1270 <table border="1" cellspacing="0" cellpadding="4">
1272 <tr><th>Classification</th><th>Types</th></tr>
1274 <td><a href="#t_integer">integer</a></td>
1275 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1278 <td><a href="#t_floating">floating point</a></td>
1279 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1282 <td><a name="t_firstclass">first class</a></td>
1283 <td><a href="#t_integer">integer</a>,
1284 <a href="#t_floating">floating point</a>,
1285 <a href="#t_pointer">pointer</a>,
1286 <a href="#t_vector">vector</a>,
1287 <a href="#t_struct">structure</a>,
1288 <a href="#t_array">array</a>,
1289 <a href="#t_label">label</a>,
1290 <a href="#t_metadata">metadata</a>.
1294 <td><a href="#t_primitive">primitive</a></td>
1295 <td><a href="#t_label">label</a>,
1296 <a href="#t_void">void</a>,
1297 <a href="#t_floating">floating point</a>,
1298 <a href="#t_metadata">metadata</a>.</td>
1301 <td><a href="#t_derived">derived</a></td>
1302 <td><a href="#t_integer">integer</a>,
1303 <a href="#t_array">array</a>,
1304 <a href="#t_function">function</a>,
1305 <a href="#t_pointer">pointer</a>,
1306 <a href="#t_struct">structure</a>,
1307 <a href="#t_pstruct">packed structure</a>,
1308 <a href="#t_vector">vector</a>,
1309 <a href="#t_opaque">opaque</a>.
1315 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1316 important. Values of these types are the only ones which can be produced by
1317 instructions, passed as arguments, or used as operands to instructions.</p>
1321 <!-- ======================================================================= -->
1322 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1324 <div class="doc_text">
1326 <p>The primitive types are the fundamental building blocks of the LLVM
1331 <!-- _______________________________________________________________________ -->
1332 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1334 <div class="doc_text">
1338 <tr><th>Type</th><th>Description</th></tr>
1339 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1340 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1341 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1342 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1343 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1349 <!-- _______________________________________________________________________ -->
1350 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1352 <div class="doc_text">
1355 <p>The void type does not represent any value and has no size.</p>
1364 <!-- _______________________________________________________________________ -->
1365 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1367 <div class="doc_text">
1370 <p>The label type represents code labels.</p>
1379 <!-- _______________________________________________________________________ -->
1380 <div class="doc_subsubsection"> <a name="t_metadata">Metadata Type</a> </div>
1382 <div class="doc_text">
1385 <p>The metadata type represents embedded metadata. The only derived type that
1386 may contain metadata is <tt>metadata*</tt> or a function type that returns or
1387 takes metadata typed parameters, but not pointer to metadata types.</p>
1397 <!-- ======================================================================= -->
1398 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1400 <div class="doc_text">
1402 <p>The real power in LLVM comes from the derived types in the system. This is
1403 what allows a programmer to represent arrays, functions, pointers, and other
1404 useful types. Note that these derived types may be recursive: For example,
1405 it is possible to have a two dimensional array.</p>
1409 <!-- _______________________________________________________________________ -->
1410 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1412 <div class="doc_text">
1415 <p>The integer type is a very simple derived type that simply specifies an
1416 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1417 2^23-1 (about 8 million) can be specified.</p>
1424 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1428 <table class="layout">
1430 <td class="left"><tt>i1</tt></td>
1431 <td class="left">a single-bit integer.</td>
1434 <td class="left"><tt>i32</tt></td>
1435 <td class="left">a 32-bit integer.</td>
1438 <td class="left"><tt>i1942652</tt></td>
1439 <td class="left">a really big integer of over 1 million bits.</td>
1443 <p>Note that the code generator does not yet support large integer types to be
1444 used as function return types. The specific limit on how large a return type
1445 the code generator can currently handle is target-dependent; currently it's
1446 often 64 bits for 32-bit targets and 128 bits for 64-bit targets.</p>
1450 <!-- _______________________________________________________________________ -->
1451 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1453 <div class="doc_text">
1456 <p>The array type is a very simple derived type that arranges elements
1457 sequentially in memory. The array type requires a size (number of elements)
1458 and an underlying data type.</p>
1462 [<# elements> x <elementtype>]
1465 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1466 be any type with a size.</p>
1469 <table class="layout">
1471 <td class="left"><tt>[40 x i32]</tt></td>
1472 <td class="left">Array of 40 32-bit integer values.</td>
1475 <td class="left"><tt>[41 x i32]</tt></td>
1476 <td class="left">Array of 41 32-bit integer values.</td>
1479 <td class="left"><tt>[4 x i8]</tt></td>
1480 <td class="left">Array of 4 8-bit integer values.</td>
1483 <p>Here are some examples of multidimensional arrays:</p>
1484 <table class="layout">
1486 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1487 <td class="left">3x4 array of 32-bit integer values.</td>
1490 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1491 <td class="left">12x10 array of single precision floating point values.</td>
1494 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1495 <td class="left">2x3x4 array of 16-bit integer values.</td>
1499 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1500 length array. Normally, accesses past the end of an array are undefined in
1501 LLVM (e.g. it is illegal to access the 5th element of a 3 element array). As
1502 a special case, however, zero length arrays are recognized to be variable
1503 length. This allows implementation of 'pascal style arrays' with the LLVM
1504 type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1506 <p>Note that the code generator does not yet support large aggregate types to be
1507 used as function return types. The specific limit on how large an aggregate
1508 return type the code generator can currently handle is target-dependent, and
1509 also dependent on the aggregate element types.</p>
1513 <!-- _______________________________________________________________________ -->
1514 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1516 <div class="doc_text">
1519 <p>The function type can be thought of as a function signature. It consists of
1520 a return type and a list of formal parameter types. The return type of a
1521 function type is a scalar type, a void type, or a struct type. If the return
1522 type is a struct type then all struct elements must be of first class types,
1523 and the struct must have at least one element.</p>
1527 <returntype list> (<parameter list>)
1530 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1531 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1532 which indicates that the function takes a variable number of arguments.
1533 Variable argument functions can access their arguments with
1534 the <a href="#int_varargs">variable argument handling intrinsic</a>
1535 functions. '<tt><returntype list></tt>' is a comma-separated list of
1536 <a href="#t_firstclass">first class</a> type specifiers.</p>
1539 <table class="layout">
1541 <td class="left"><tt>i32 (i32)</tt></td>
1542 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1544 </tr><tr class="layout">
1545 <td class="left"><tt>float (i16 signext, i32 *) *
1547 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1548 an <tt>i16</tt> that should be sign extended and a
1549 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1552 </tr><tr class="layout">
1553 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1554 <td class="left">A vararg function that takes at least one
1555 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1556 which returns an integer. This is the signature for <tt>printf</tt> in
1559 </tr><tr class="layout">
1560 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1561 <td class="left">A function taking an <tt>i32</tt>, returning two
1562 <tt>i32</tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1569 <!-- _______________________________________________________________________ -->
1570 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1572 <div class="doc_text">
1575 <p>The structure type is used to represent a collection of data members together
1576 in memory. The packing of the field types is defined to match the ABI of the
1577 underlying processor. The elements of a structure may be any type that has a
1580 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1581 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1582 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1586 { <type list> }
1590 <table class="layout">
1592 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1593 <td class="left">A triple of three <tt>i32</tt> values</td>
1594 </tr><tr class="layout">
1595 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1596 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1597 second element is a <a href="#t_pointer">pointer</a> to a
1598 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1599 an <tt>i32</tt>.</td>
1603 <p>Note that the code generator does not yet support large aggregate types to be
1604 used as function return types. The specific limit on how large an aggregate
1605 return type the code generator can currently handle is target-dependent, and
1606 also dependent on the aggregate element types.</p>
1610 <!-- _______________________________________________________________________ -->
1611 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1614 <div class="doc_text">
1617 <p>The packed structure type is used to represent a collection of data members
1618 together in memory. There is no padding between fields. Further, the
1619 alignment of a packed structure is 1 byte. The elements of a packed
1620 structure may be any type that has a size.</p>
1622 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1623 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1624 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1628 < { <type list> } >
1632 <table class="layout">
1634 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1635 <td class="left">A triple of three <tt>i32</tt> values</td>
1636 </tr><tr class="layout">
1638 <tt>< { float, i32 (i32)* } ></tt></td>
1639 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1640 second element is a <a href="#t_pointer">pointer</a> to a
1641 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1642 an <tt>i32</tt>.</td>
1648 <!-- _______________________________________________________________________ -->
1649 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1651 <div class="doc_text">
1654 <p>As in many languages, the pointer type represents a pointer or reference to
1655 another object, which must live in memory. Pointer types may have an optional
1656 address space attribute defining the target-specific numbered address space
1657 where the pointed-to object resides. The default address space is zero.</p>
1659 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
1660 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1668 <table class="layout">
1670 <td class="left"><tt>[4 x i32]*</tt></td>
1671 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1672 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1675 <td class="left"><tt>i32 (i32 *) *</tt></td>
1676 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1677 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1681 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1682 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1683 that resides in address space #5.</td>
1689 <!-- _______________________________________________________________________ -->
1690 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1692 <div class="doc_text">
1695 <p>A vector type is a simple derived type that represents a vector of elements.
1696 Vector types are used when multiple primitive data are operated in parallel
1697 using a single instruction (SIMD). A vector type requires a size (number of
1698 elements) and an underlying primitive data type. Vectors must have a power
1699 of two length (1, 2, 4, 8, 16 ...). Vector types are considered
1700 <a href="#t_firstclass">first class</a>.</p>
1704 < <# elements> x <elementtype> >
1707 <p>The number of elements is a constant integer value; elementtype may be any
1708 integer or floating point type.</p>
1711 <table class="layout">
1713 <td class="left"><tt><4 x i32></tt></td>
1714 <td class="left">Vector of 4 32-bit integer values.</td>
1717 <td class="left"><tt><8 x float></tt></td>
1718 <td class="left">Vector of 8 32-bit floating-point values.</td>
1721 <td class="left"><tt><2 x i64></tt></td>
1722 <td class="left">Vector of 2 64-bit integer values.</td>
1726 <p>Note that the code generator does not yet support large vector types to be
1727 used as function return types. The specific limit on how large a vector
1728 return type codegen can currently handle is target-dependent; currently it's
1729 often a few times longer than a hardware vector register.</p>
1733 <!-- _______________________________________________________________________ -->
1734 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1735 <div class="doc_text">
1738 <p>Opaque types are used to represent unknown types in the system. This
1739 corresponds (for example) to the C notion of a forward declared structure
1740 type. In LLVM, opaque types can eventually be resolved to any type (not just
1741 a structure type).</p>
1749 <table class="layout">
1751 <td class="left"><tt>opaque</tt></td>
1752 <td class="left">An opaque type.</td>
1758 <!-- ======================================================================= -->
1759 <div class="doc_subsection">
1760 <a name="t_uprefs">Type Up-references</a>
1763 <div class="doc_text">
1766 <p>An "up reference" allows you to refer to a lexically enclosing type without
1767 requiring it to have a name. For instance, a structure declaration may
1768 contain a pointer to any of the types it is lexically a member of. Example
1769 of up references (with their equivalent as named type declarations)
1773 { \2 * } %x = type { %x* }
1774 { \2 }* %y = type { %y }*
1778 <p>An up reference is needed by the asmprinter for printing out cyclic types
1779 when there is no declared name for a type in the cycle. Because the
1780 asmprinter does not want to print out an infinite type string, it needs a
1781 syntax to handle recursive types that have no names (all names are optional
1789 <p>The level is the count of the lexical type that is being referred to.</p>
1792 <table class="layout">
1794 <td class="left"><tt>\1*</tt></td>
1795 <td class="left">Self-referential pointer.</td>
1798 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1799 <td class="left">Recursive structure where the upref refers to the out-most
1806 <!-- *********************************************************************** -->
1807 <div class="doc_section"> <a name="constants">Constants</a> </div>
1808 <!-- *********************************************************************** -->
1810 <div class="doc_text">
1812 <p>LLVM has several different basic types of constants. This section describes
1813 them all and their syntax.</p>
1817 <!-- ======================================================================= -->
1818 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1820 <div class="doc_text">
1823 <dt><b>Boolean constants</b></dt>
1824 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1825 constants of the <tt><a href="#t_primitive">i1</a></tt> type.</dd>
1827 <dt><b>Integer constants</b></dt>
1828 <dd>Standard integers (such as '4') are constants of
1829 the <a href="#t_integer">integer</a> type. Negative numbers may be used
1830 with integer types.</dd>
1832 <dt><b>Floating point constants</b></dt>
1833 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1834 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1835 notation (see below). The assembler requires the exact decimal value of a
1836 floating-point constant. For example, the assembler accepts 1.25 but
1837 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1838 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1840 <dt><b>Null pointer constants</b></dt>
1841 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1842 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1845 <p>The one non-intuitive notation for constants is the hexadecimal form of
1846 floating point constants. For example, the form '<tt>double
1847 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
1848 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
1849 constants are required (and the only time that they are generated by the
1850 disassembler) is when a floating point constant must be emitted but it cannot
1851 be represented as a decimal floating point number in a reasonable number of
1852 digits. For example, NaN's, infinities, and other special values are
1853 represented in their IEEE hexadecimal format so that assembly and disassembly
1854 do not cause any bits to change in the constants.</p>
1856 <p>When using the hexadecimal form, constants of types float and double are
1857 represented using the 16-digit form shown above (which matches the IEEE754
1858 representation for double); float values must, however, be exactly
1859 representable as IEE754 single precision. Hexadecimal format is always used
1860 for long double, and there are three forms of long double. The 80-bit format
1861 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
1862 The 128-bit format used by PowerPC (two adjacent doubles) is represented
1863 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
1864 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
1865 currently supported target uses this format. Long doubles will only work if
1866 they match the long double format on your target. All hexadecimal formats
1867 are big-endian (sign bit at the left).</p>
1871 <!-- ======================================================================= -->
1872 <div class="doc_subsection">
1873 <a name="aggregateconstants"></a> <!-- old anchor -->
1874 <a name="complexconstants">Complex Constants</a>
1877 <div class="doc_text">
1879 <p>Complex constants are a (potentially recursive) combination of simple
1880 constants and smaller complex constants.</p>
1883 <dt><b>Structure constants</b></dt>
1884 <dd>Structure constants are represented with notation similar to structure
1885 type definitions (a comma separated list of elements, surrounded by braces
1886 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1887 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
1888 Structure constants must have <a href="#t_struct">structure type</a>, and
1889 the number and types of elements must match those specified by the
1892 <dt><b>Array constants</b></dt>
1893 <dd>Array constants are represented with notation similar to array type
1894 definitions (a comma separated list of elements, surrounded by square
1895 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
1896 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
1897 the number and types of elements must match those specified by the
1900 <dt><b>Vector constants</b></dt>
1901 <dd>Vector constants are represented with notation similar to vector type
1902 definitions (a comma separated list of elements, surrounded by
1903 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
1904 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
1905 have <a href="#t_vector">vector type</a>, and the number and types of
1906 elements must match those specified by the type.</dd>
1908 <dt><b>Zero initialization</b></dt>
1909 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1910 value to zero of <em>any</em> type, including scalar and aggregate types.
1911 This is often used to avoid having to print large zero initializers
1912 (e.g. for large arrays) and is always exactly equivalent to using explicit
1913 zero initializers.</dd>
1915 <dt><b>Metadata node</b></dt>
1916 <dd>A metadata node is a structure-like constant with
1917 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
1918 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
1919 be interpreted as part of the instruction stream, metadata is a place to
1920 attach additional information such as debug info.</dd>
1925 <!-- ======================================================================= -->
1926 <div class="doc_subsection">
1927 <a name="globalconstants">Global Variable and Function Addresses</a>
1930 <div class="doc_text">
1932 <p>The addresses of <a href="#globalvars">global variables</a>
1933 and <a href="#functionstructure">functions</a> are always implicitly valid
1934 (link-time) constants. These constants are explicitly referenced when
1935 the <a href="#identifiers">identifier for the global</a> is used and always
1936 have <a href="#t_pointer">pointer</a> type. For example, the following is a
1937 legal LLVM file:</p>
1939 <div class="doc_code">
1943 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1949 <!-- ======================================================================= -->
1950 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1951 <div class="doc_text">
1953 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has no
1954 specific value. Undefined values may be of any type and be used anywhere a
1955 constant is permitted.</p>
1957 <p>Undefined values indicate to the compiler that the program is well defined no
1958 matter what value is used, giving the compiler more freedom to optimize.</p>
1962 <!-- ======================================================================= -->
1963 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1966 <div class="doc_text">
1968 <p>Constant expressions are used to allow expressions involving other constants
1969 to be used as constants. Constant expressions may be of
1970 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
1971 operation that does not have side effects (e.g. load and call are not
1972 supported). The following is the syntax for constant expressions:</p>
1975 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1976 <dd>Truncate a constant to another type. The bit size of CST must be larger
1977 than the bit size of TYPE. Both types must be integers.</dd>
1979 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1980 <dd>Zero extend a constant to another type. The bit size of CST must be
1981 smaller or equal to the bit size of TYPE. Both types must be
1984 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1985 <dd>Sign extend a constant to another type. The bit size of CST must be
1986 smaller or equal to the bit size of TYPE. Both types must be
1989 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1990 <dd>Truncate a floating point constant to another floating point type. The
1991 size of CST must be larger than the size of TYPE. Both types must be
1992 floating point.</dd>
1994 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1995 <dd>Floating point extend a constant to another type. The size of CST must be
1996 smaller or equal to the size of TYPE. Both types must be floating
1999 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
2000 <dd>Convert a floating point constant to the corresponding unsigned integer
2001 constant. TYPE must be a scalar or vector integer type. CST must be of
2002 scalar or vector floating point type. Both CST and TYPE must be scalars,
2003 or vectors of the same number of elements. If the value won't fit in the
2004 integer type, the results are undefined.</dd>
2006 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
2007 <dd>Convert a floating point constant to the corresponding signed integer
2008 constant. TYPE must be a scalar or vector integer type. CST must be of
2009 scalar or vector floating point type. Both CST and TYPE must be scalars,
2010 or vectors of the same number of elements. If the value won't fit in the
2011 integer type, the results are undefined.</dd>
2013 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
2014 <dd>Convert an unsigned integer constant to the corresponding floating point
2015 constant. TYPE must be a scalar or vector floating point type. CST must be
2016 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2017 vectors of the same number of elements. If the value won't fit in the
2018 floating point type, the results are undefined.</dd>
2020 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
2021 <dd>Convert a signed integer constant to the corresponding floating point
2022 constant. TYPE must be a scalar or vector floating point type. CST must be
2023 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2024 vectors of the same number of elements. If the value won't fit in the
2025 floating point type, the results are undefined.</dd>
2027 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
2028 <dd>Convert a pointer typed constant to the corresponding integer constant
2029 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2030 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2031 make it fit in <tt>TYPE</tt>.</dd>
2033 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
2034 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2035 type. CST must be of integer type. The CST value is zero extended,
2036 truncated, or unchanged to make it fit in a pointer size. This one is
2037 <i>really</i> dangerous!</dd>
2039 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
2040 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2041 are the same as those for the <a href="#i_bitcast">bitcast
2042 instruction</a>.</dd>
2044 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2045 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2046 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2047 instruction, the index list may have zero or more indexes, which are
2048 required to make sense for the type of "CSTPTR".</dd>
2050 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
2051 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2053 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
2054 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2056 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
2057 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2059 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2060 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2063 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2064 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2067 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2068 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2071 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2072 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2073 be any of the <a href="#binaryops">binary</a>
2074 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2075 on operands are the same as those for the corresponding instruction
2076 (e.g. no bitwise operations on floating point values are allowed).</dd>
2081 <!-- ======================================================================= -->
2082 <div class="doc_subsection"><a name="metadata">Embedded Metadata</a>
2085 <div class="doc_text">
2087 <p>Embedded metadata provides a way to attach arbitrary data to the instruction
2088 stream without affecting the behaviour of the program. There are two
2089 metadata primitives, strings and nodes. All metadata has the
2090 <tt>metadata</tt> type and is identified in syntax by a preceding exclamation
2091 point ('<tt>!</tt>').</p>
2093 <p>A metadata string is a string surrounded by double quotes. It can contain
2094 any character by escaping non-printable characters with "\xx" where "xx" is
2095 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2097 <p>Metadata nodes are represented with notation similar to structure constants
2098 (a comma separated list of elements, surrounded by braces and preceeded by an
2099 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2102 <p>A metadata node will attempt to track changes to the values it holds. In the
2103 event that a value is deleted, it will be replaced with a typeless
2104 "<tt>null</tt>", such as "<tt>metadata !{null, i32 10}</tt>".</p>
2106 <p>Optimizations may rely on metadata to provide additional information about
2107 the program that isn't available in the instructions, or that isn't easily
2108 computable. Similarly, the code generator may expect a certain metadata
2109 format to be used to express debugging information.</p>
2113 <!-- *********************************************************************** -->
2114 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2115 <!-- *********************************************************************** -->
2117 <!-- ======================================================================= -->
2118 <div class="doc_subsection">
2119 <a name="inlineasm">Inline Assembler Expressions</a>
2122 <div class="doc_text">
2124 <p>LLVM supports inline assembler expressions (as opposed
2125 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2126 a special value. This value represents the inline assembler as a string
2127 (containing the instructions to emit), a list of operand constraints (stored
2128 as a string), and a flag that indicates whether or not the inline asm
2129 expression has side effects. An example inline assembler expression is:</p>
2131 <div class="doc_code">
2133 i32 (i32) asm "bswap $0", "=r,r"
2137 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2138 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2141 <div class="doc_code">
2143 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2147 <p>Inline asms with side effects not visible in the constraint list must be
2148 marked as having side effects. This is done through the use of the
2149 '<tt>sideeffect</tt>' keyword, like so:</p>
2151 <div class="doc_code">
2153 call void asm sideeffect "eieio", ""()
2157 <p>TODO: The format of the asm and constraints string still need to be
2158 documented here. Constraints on what can be done (e.g. duplication, moving,
2159 etc need to be documented). This is probably best done by reference to
2160 another document that covers inline asm from a holistic perspective.</p>
2165 <!-- *********************************************************************** -->
2166 <div class="doc_section">
2167 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2169 <!-- *********************************************************************** -->
2171 <p>LLVM has a number of "magic" global variables that contain data that affect
2172 code generation or other IR semantics. These are documented here. All globals
2173 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2174 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2177 <!-- ======================================================================= -->
2178 <div class="doc_subsection">
2179 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2182 <div class="doc_text">
2184 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2185 href="#linkage_appending">appending linkage</a>. This array contains a list of
2186 pointers to global variables and functions which may optionally have a pointer
2187 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2193 @llvm.used = appending global [2 x i8*] [
2195 i8* bitcast (i32* @Y to i8*)
2196 ], section "llvm.metadata"
2199 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2200 compiler, assembler, and linker are required to treat the symbol as if there is
2201 a reference to the global that it cannot see. For example, if a variable has
2202 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2203 list, it cannot be deleted. This is commonly used to represent references from
2204 inline asms and other things the compiler cannot "see", and corresponds to
2205 "attribute((used))" in GNU C.</p>
2207 <p>On some targets, the code generator must emit a directive to the assembler or
2208 object file to prevent the assembler and linker from molesting the symbol.</p>
2212 <!-- ======================================================================= -->
2213 <div class="doc_subsection">
2214 <a name="intg_compiler_used">The '<tt>llvm.compiler.used</tt>' Global Variable</a>
2217 <div class="doc_text">
2219 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2220 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2221 touching the symbol. On targets that support it, this allows an intelligent
2222 linker to optimize references to the symbol without being impeded as it would be
2223 by <tt>@llvm.used</tt>.</p>
2225 <p>This is a rare construct that should only be used in rare circumstances, and
2226 should not be exposed to source languages.</p>
2230 <!-- ======================================================================= -->
2231 <div class="doc_subsection">
2232 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2235 <div class="doc_text">
2237 <p>TODO: Describe this.</p>
2241 <!-- ======================================================================= -->
2242 <div class="doc_subsection">
2243 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2246 <div class="doc_text">
2248 <p>TODO: Describe this.</p>
2253 <!-- *********************************************************************** -->
2254 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2255 <!-- *********************************************************************** -->
2257 <div class="doc_text">
2259 <p>The LLVM instruction set consists of several different classifications of
2260 instructions: <a href="#terminators">terminator
2261 instructions</a>, <a href="#binaryops">binary instructions</a>,
2262 <a href="#bitwiseops">bitwise binary instructions</a>,
2263 <a href="#memoryops">memory instructions</a>, and
2264 <a href="#otherops">other instructions</a>.</p>
2268 <!-- ======================================================================= -->
2269 <div class="doc_subsection"> <a name="terminators">Terminator
2270 Instructions</a> </div>
2272 <div class="doc_text">
2274 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
2275 in a program ends with a "Terminator" instruction, which indicates which
2276 block should be executed after the current block is finished. These
2277 terminator instructions typically yield a '<tt>void</tt>' value: they produce
2278 control flow, not values (the one exception being the
2279 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2281 <p>There are six different terminator instructions: the
2282 '<a href="#i_ret"><tt>ret</tt></a>' instruction, the
2283 '<a href="#i_br"><tt>br</tt></a>' instruction, the
2284 '<a href="#i_switch"><tt>switch</tt></a>' instruction, the
2285 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the
2286 '<a href="#i_unwind"><tt>unwind</tt></a>' instruction, and the
2287 '<a href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2291 <!-- _______________________________________________________________________ -->
2292 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2293 Instruction</a> </div>
2295 <div class="doc_text">
2299 ret <type> <value> <i>; Return a value from a non-void function</i>
2300 ret void <i>; Return from void function</i>
2304 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
2305 a value) from a function back to the caller.</p>
2307 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
2308 value and then causes control flow, and one that just causes control flow to
2312 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
2313 return value. The type of the return value must be a
2314 '<a href="#t_firstclass">first class</a>' type.</p>
2316 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
2317 non-void return type and contains a '<tt>ret</tt>' instruction with no return
2318 value or a return value with a type that does not match its type, or if it
2319 has a void return type and contains a '<tt>ret</tt>' instruction with a
2323 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
2324 the calling function's context. If the caller is a
2325 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
2326 instruction after the call. If the caller was an
2327 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
2328 the beginning of the "normal" destination block. If the instruction returns
2329 a value, that value shall set the call or invoke instruction's return
2334 ret i32 5 <i>; Return an integer value of 5</i>
2335 ret void <i>; Return from a void function</i>
2336 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2339 <p>Note that the code generator does not yet fully support large
2340 return values. The specific sizes that are currently supported are
2341 dependent on the target. For integers, on 32-bit targets the limit
2342 is often 64 bits, and on 64-bit targets the limit is often 128 bits.
2343 For aggregate types, the current limits are dependent on the element
2344 types; for example targets are often limited to 2 total integer
2345 elements and 2 total floating-point elements.</p>
2348 <!-- _______________________________________________________________________ -->
2349 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2351 <div class="doc_text">
2355 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2359 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
2360 different basic block in the current function. There are two forms of this
2361 instruction, corresponding to a conditional branch and an unconditional
2365 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
2366 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
2367 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
2371 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2372 argument is evaluated. If the value is <tt>true</tt>, control flows to the
2373 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2374 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2379 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
2380 br i1 %cond, label %IfEqual, label %IfUnequal
2382 <a href="#i_ret">ret</a> i32 1
2384 <a href="#i_ret">ret</a> i32 0
2389 <!-- _______________________________________________________________________ -->
2390 <div class="doc_subsubsection">
2391 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2394 <div class="doc_text">
2398 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2402 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2403 several different places. It is a generalization of the '<tt>br</tt>'
2404 instruction, allowing a branch to occur to one of many possible
2408 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2409 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
2410 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
2411 The table is not allowed to contain duplicate constant entries.</p>
2414 <p>The <tt>switch</tt> instruction specifies a table of values and
2415 destinations. When the '<tt>switch</tt>' instruction is executed, this table
2416 is searched for the given value. If the value is found, control flow is
2417 transfered to the corresponding destination; otherwise, control flow is
2418 transfered to the default destination.</p>
2420 <h5>Implementation:</h5>
2421 <p>Depending on properties of the target machine and the particular
2422 <tt>switch</tt> instruction, this instruction may be code generated in
2423 different ways. For example, it could be generated as a series of chained
2424 conditional branches or with a lookup table.</p>
2428 <i>; Emulate a conditional br instruction</i>
2429 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2430 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2432 <i>; Emulate an unconditional br instruction</i>
2433 switch i32 0, label %dest [ ]
2435 <i>; Implement a jump table:</i>
2436 switch i32 %val, label %otherwise [ i32 0, label %onzero
2438 i32 2, label %ontwo ]
2443 <!-- _______________________________________________________________________ -->
2444 <div class="doc_subsubsection">
2445 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2448 <div class="doc_text">
2452 <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>]
2453 to label <normal label> unwind label <exception label>
2457 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2458 function, with the possibility of control flow transfer to either the
2459 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
2460 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
2461 control flow will return to the "normal" label. If the callee (or any
2462 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
2463 instruction, control is interrupted and continued at the dynamically nearest
2464 "exception" label.</p>
2467 <p>This instruction requires several arguments:</p>
2470 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
2471 convention</a> the call should use. If none is specified, the call
2472 defaults to using C calling conventions.</li>
2474 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2475 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
2476 '<tt>inreg</tt>' attributes are valid here.</li>
2478 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2479 function value being invoked. In most cases, this is a direct function
2480 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
2481 off an arbitrary pointer to function value.</li>
2483 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2484 function to be invoked. </li>
2486 <li>'<tt>function args</tt>': argument list whose types match the function
2487 signature argument types. If the function signature indicates the
2488 function accepts a variable number of arguments, the extra arguments can
2491 <li>'<tt>normal label</tt>': the label reached when the called function
2492 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2494 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2495 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2497 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2498 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2499 '<tt>readnone</tt>' attributes are valid here.</li>
2503 <p>This instruction is designed to operate as a standard
2504 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
2505 primary difference is that it establishes an association with a label, which
2506 is used by the runtime library to unwind the stack.</p>
2508 <p>This instruction is used in languages with destructors to ensure that proper
2509 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2510 exception. Additionally, this is important for implementation of
2511 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2513 <p>For the purposes of the SSA form, the definition of the value returned by the
2514 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
2515 block to the "normal" label. If the callee unwinds then no return value is
2520 %retval = invoke i32 @Test(i32 15) to label %Continue
2521 unwind label %TestCleanup <i>; {i32}:retval set</i>
2522 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2523 unwind label %TestCleanup <i>; {i32}:retval set</i>
2528 <!-- _______________________________________________________________________ -->
2530 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2531 Instruction</a> </div>
2533 <div class="doc_text">
2541 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2542 at the first callee in the dynamic call stack which used
2543 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
2544 This is primarily used to implement exception handling.</p>
2547 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2548 immediately halt. The dynamic call stack is then searched for the
2549 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
2550 Once found, execution continues at the "exceptional" destination block
2551 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
2552 instruction in the dynamic call chain, undefined behavior results.</p>
2556 <!-- _______________________________________________________________________ -->
2558 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2559 Instruction</a> </div>
2561 <div class="doc_text">
2569 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2570 instruction is used to inform the optimizer that a particular portion of the
2571 code is not reachable. This can be used to indicate that the code after a
2572 no-return function cannot be reached, and other facts.</p>
2575 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2579 <!-- ======================================================================= -->
2580 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2582 <div class="doc_text">
2584 <p>Binary operators are used to do most of the computation in a program. They
2585 require two operands of the same type, execute an operation on them, and
2586 produce a single value. The operands might represent multiple data, as is
2587 the case with the <a href="#t_vector">vector</a> data type. The result value
2588 has the same type as its operands.</p>
2590 <p>There are several different binary operators:</p>
2594 <!-- _______________________________________________________________________ -->
2595 <div class="doc_subsubsection">
2596 <a name="i_add">'<tt>add</tt>' Instruction</a>
2599 <div class="doc_text">
2603 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2607 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2610 <p>The two arguments to the '<tt>add</tt>' instruction must
2611 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2612 integer values. Both arguments must have identical types.</p>
2615 <p>The value produced is the integer sum of the two operands.</p>
2617 <p>If the sum has unsigned overflow, the result returned is the mathematical
2618 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
2620 <p>Because LLVM integers use a two's complement representation, this instruction
2621 is appropriate for both signed and unsigned integers.</p>
2625 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2630 <!-- _______________________________________________________________________ -->
2631 <div class="doc_subsubsection">
2632 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
2635 <div class="doc_text">
2639 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2643 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
2646 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
2647 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2648 floating point values. Both arguments must have identical types.</p>
2651 <p>The value produced is the floating point sum of the two operands.</p>
2655 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
2660 <!-- _______________________________________________________________________ -->
2661 <div class="doc_subsubsection">
2662 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2665 <div class="doc_text">
2669 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2673 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2676 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2677 '<tt>neg</tt>' instruction present in most other intermediate
2678 representations.</p>
2681 <p>The two arguments to the '<tt>sub</tt>' instruction must
2682 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2683 integer values. Both arguments must have identical types.</p>
2686 <p>The value produced is the integer difference of the two operands.</p>
2688 <p>If the difference has unsigned overflow, the result returned is the
2689 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
2692 <p>Because LLVM integers use a two's complement representation, this instruction
2693 is appropriate for both signed and unsigned integers.</p>
2697 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2698 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2703 <!-- _______________________________________________________________________ -->
2704 <div class="doc_subsubsection">
2705 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
2708 <div class="doc_text">
2712 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2716 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
2719 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
2720 '<tt>fneg</tt>' instruction present in most other intermediate
2721 representations.</p>
2724 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
2725 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2726 floating point values. Both arguments must have identical types.</p>
2729 <p>The value produced is the floating point difference of the two operands.</p>
2733 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
2734 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
2739 <!-- _______________________________________________________________________ -->
2740 <div class="doc_subsubsection">
2741 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2744 <div class="doc_text">
2748 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2752 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
2755 <p>The two arguments to the '<tt>mul</tt>' instruction must
2756 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2757 integer values. Both arguments must have identical types.</p>
2760 <p>The value produced is the integer product of the two operands.</p>
2762 <p>If the result of the multiplication has unsigned overflow, the result
2763 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
2764 width of the result.</p>
2766 <p>Because LLVM integers use a two's complement representation, and the result
2767 is the same width as the operands, this instruction returns the correct
2768 result for both signed and unsigned integers. If a full product
2769 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
2770 be sign-extended or zero-extended as appropriate to the width of the full
2775 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2780 <!-- _______________________________________________________________________ -->
2781 <div class="doc_subsubsection">
2782 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
2785 <div class="doc_text">
2789 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2793 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
2796 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
2797 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2798 floating point values. Both arguments must have identical types.</p>
2801 <p>The value produced is the floating point product of the two operands.</p>
2805 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
2810 <!-- _______________________________________________________________________ -->
2811 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2814 <div class="doc_text">
2818 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2822 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
2825 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2826 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2827 values. Both arguments must have identical types.</p>
2830 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2832 <p>Note that unsigned integer division and signed integer division are distinct
2833 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2835 <p>Division by zero leads to undefined behavior.</p>
2839 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2844 <!-- _______________________________________________________________________ -->
2845 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2848 <div class="doc_text">
2852 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2856 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
2859 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2860 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2861 values. Both arguments must have identical types.</p>
2864 <p>The value produced is the signed integer quotient of the two operands rounded
2867 <p>Note that signed integer division and unsigned integer division are distinct
2868 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2870 <p>Division by zero leads to undefined behavior. Overflow also leads to
2871 undefined behavior; this is a rare case, but can occur, for example, by doing
2872 a 32-bit division of -2147483648 by -1.</p>
2876 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2881 <!-- _______________________________________________________________________ -->
2882 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2883 Instruction</a> </div>
2885 <div class="doc_text">
2889 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2893 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
2896 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2897 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2898 floating point values. Both arguments must have identical types.</p>
2901 <p>The value produced is the floating point quotient of the two operands.</p>
2905 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2910 <!-- _______________________________________________________________________ -->
2911 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2914 <div class="doc_text">
2918 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2922 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
2923 division of its two arguments.</p>
2926 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2927 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2928 values. Both arguments must have identical types.</p>
2931 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2932 This instruction always performs an unsigned division to get the
2935 <p>Note that unsigned integer remainder and signed integer remainder are
2936 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2938 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2942 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2947 <!-- _______________________________________________________________________ -->
2948 <div class="doc_subsubsection">
2949 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2952 <div class="doc_text">
2956 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2960 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
2961 division of its two operands. This instruction can also take
2962 <a href="#t_vector">vector</a> versions of the values in which case the
2963 elements must be integers.</p>
2966 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2967 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2968 values. Both arguments must have identical types.</p>
2971 <p>This instruction returns the <i>remainder</i> of a division (where the result
2972 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2973 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
2974 a value. For more information about the difference,
2975 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2976 Math Forum</a>. For a table of how this is implemented in various languages,
2977 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2978 Wikipedia: modulo operation</a>.</p>
2980 <p>Note that signed integer remainder and unsigned integer remainder are
2981 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2983 <p>Taking the remainder of a division by zero leads to undefined behavior.
2984 Overflow also leads to undefined behavior; this is a rare case, but can
2985 occur, for example, by taking the remainder of a 32-bit division of
2986 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
2987 lets srem be implemented using instructions that return both the result of
2988 the division and the remainder.)</p>
2992 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2997 <!-- _______________________________________________________________________ -->
2998 <div class="doc_subsubsection">
2999 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
3001 <div class="doc_text">
3005 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3009 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3010 its two operands.</p>
3013 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3014 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3015 floating point values. Both arguments must have identical types.</p>
3018 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3019 has the same sign as the dividend.</p>
3023 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3028 <!-- ======================================================================= -->
3029 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
3030 Operations</a> </div>
3032 <div class="doc_text">
3034 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3035 program. They are generally very efficient instructions and can commonly be
3036 strength reduced from other instructions. They require two operands of the
3037 same type, execute an operation on them, and produce a single value. The
3038 resulting value is the same type as its operands.</p>
3042 <!-- _______________________________________________________________________ -->
3043 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
3044 Instruction</a> </div>
3046 <div class="doc_text">
3050 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3054 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3055 a specified number of bits.</p>
3058 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3059 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3060 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3063 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3064 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3065 is (statically or dynamically) negative or equal to or larger than the number
3066 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3067 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3068 shift amount in <tt>op2</tt>.</p>
3072 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3073 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3074 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3075 <result> = shl i32 1, 32 <i>; undefined</i>
3076 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3081 <!-- _______________________________________________________________________ -->
3082 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
3083 Instruction</a> </div>
3085 <div class="doc_text">
3089 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3093 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
3094 operand shifted to the right a specified number of bits with zero fill.</p>
3097 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
3098 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3099 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3102 <p>This instruction always performs a logical shift right operation. The most
3103 significant bits of the result will be filled with zero bits after the shift.
3104 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
3105 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3106 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3107 shift amount in <tt>op2</tt>.</p>
3111 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
3112 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
3113 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
3114 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
3115 <result> = lshr i32 1, 32 <i>; undefined</i>
3116 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3121 <!-- _______________________________________________________________________ -->
3122 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
3123 Instruction</a> </div>
3124 <div class="doc_text">
3128 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3132 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
3133 operand shifted to the right a specified number of bits with sign
3137 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
3138 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3139 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3142 <p>This instruction always performs an arithmetic shift right operation, The
3143 most significant bits of the result will be filled with the sign bit
3144 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
3145 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
3146 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
3147 the corresponding shift amount in <tt>op2</tt>.</p>
3151 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
3152 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
3153 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
3154 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
3155 <result> = ashr i32 1, 32 <i>; undefined</i>
3156 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3161 <!-- _______________________________________________________________________ -->
3162 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
3163 Instruction</a> </div>
3165 <div class="doc_text">
3169 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3173 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
3177 <p>The two arguments to the '<tt>and</tt>' instruction must be
3178 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3179 values. Both arguments must have identical types.</p>
3182 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
3184 <table border="1" cellspacing="0" cellpadding="4">
3216 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
3217 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
3218 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
3221 <!-- _______________________________________________________________________ -->
3222 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
3224 <div class="doc_text">
3228 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3232 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
3236 <p>The two arguments to the '<tt>or</tt>' instruction must be
3237 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3238 values. Both arguments must have identical types.</p>
3241 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
3243 <table border="1" cellspacing="0" cellpadding="4">
3275 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3276 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3277 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3282 <!-- _______________________________________________________________________ -->
3283 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3284 Instruction</a> </div>
3286 <div class="doc_text">
3290 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3294 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
3295 its two operands. The <tt>xor</tt> is used to implement the "one's
3296 complement" operation, which is the "~" operator in C.</p>
3299 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3300 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3301 values. Both arguments must have identical types.</p>
3304 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3306 <table border="1" cellspacing="0" cellpadding="4">
3338 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3339 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3340 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3341 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3346 <!-- ======================================================================= -->
3347 <div class="doc_subsection">
3348 <a name="vectorops">Vector Operations</a>
3351 <div class="doc_text">
3353 <p>LLVM supports several instructions to represent vector operations in a
3354 target-independent manner. These instructions cover the element-access and
3355 vector-specific operations needed to process vectors effectively. While LLVM
3356 does directly support these vector operations, many sophisticated algorithms
3357 will want to use target-specific intrinsics to take full advantage of a
3358 specific target.</p>
3362 <!-- _______________________________________________________________________ -->
3363 <div class="doc_subsubsection">
3364 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3367 <div class="doc_text">
3371 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3375 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
3376 from a vector at a specified index.</p>
3380 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
3381 of <a href="#t_vector">vector</a> type. The second operand is an index
3382 indicating the position from which to extract the element. The index may be
3386 <p>The result is a scalar of the same type as the element type of
3387 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3388 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3389 results are undefined.</p>
3393 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3398 <!-- _______________________________________________________________________ -->
3399 <div class="doc_subsubsection">
3400 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3403 <div class="doc_text">
3407 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3411 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
3412 vector at a specified index.</p>
3415 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
3416 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
3417 whose type must equal the element type of the first operand. The third
3418 operand is an index indicating the position at which to insert the value.
3419 The index may be a variable.</p>
3422 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
3423 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
3424 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3425 results are undefined.</p>
3429 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3434 <!-- _______________________________________________________________________ -->
3435 <div class="doc_subsubsection">
3436 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3439 <div class="doc_text">
3443 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3447 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3448 from two input vectors, returning a vector with the same element type as the
3449 input and length that is the same as the shuffle mask.</p>
3452 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3453 with types that match each other. The third argument is a shuffle mask whose
3454 element type is always 'i32'. The result of the instruction is a vector
3455 whose length is the same as the shuffle mask and whose element type is the
3456 same as the element type of the first two operands.</p>
3458 <p>The shuffle mask operand is required to be a constant vector with either
3459 constant integer or undef values.</p>
3462 <p>The elements of the two input vectors are numbered from left to right across
3463 both of the vectors. The shuffle mask operand specifies, for each element of
3464 the result vector, which element of the two input vectors the result element
3465 gets. The element selector may be undef (meaning "don't care") and the
3466 second operand may be undef if performing a shuffle from only one vector.</p>
3470 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3471 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3472 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
3473 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3474 %result = shufflevector <8 x i32> %v1, <8 x i32> undef,
3475 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3476 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3477 <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>
3482 <!-- ======================================================================= -->
3483 <div class="doc_subsection">
3484 <a name="aggregateops">Aggregate Operations</a>
3487 <div class="doc_text">
3489 <p>LLVM supports several instructions for working with aggregate values.</p>
3493 <!-- _______________________________________________________________________ -->
3494 <div class="doc_subsubsection">
3495 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3498 <div class="doc_text">
3502 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3506 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3507 or array element from an aggregate value.</p>
3510 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
3511 of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type. The
3512 operands are constant indices to specify which value to extract in a similar
3513 manner as indices in a
3514 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
3517 <p>The result is the value at the position in the aggregate specified by the
3522 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3527 <!-- _______________________________________________________________________ -->
3528 <div class="doc_subsubsection">
3529 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3532 <div class="doc_text">
3536 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3540 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a struct field or
3541 array element in an aggregate.</p>
3545 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
3546 of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type. The
3547 second operand is a first-class value to insert. The following operands are
3548 constant indices indicating the position at which to insert the value in a
3549 similar manner as indices in a
3550 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction. The
3551 value to insert must have the same type as the value identified by the
3555 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
3556 that of <tt>val</tt> except that the value at the position specified by the
3557 indices is that of <tt>elt</tt>.</p>
3561 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3567 <!-- ======================================================================= -->
3568 <div class="doc_subsection">
3569 <a name="memoryops">Memory Access and Addressing Operations</a>
3572 <div class="doc_text">
3574 <p>A key design point of an SSA-based representation is how it represents
3575 memory. In LLVM, no memory locations are in SSA form, which makes things
3576 very simple. This section describes how to read, write, allocate, and free
3581 <!-- _______________________________________________________________________ -->
3582 <div class="doc_subsubsection">
3583 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3586 <div class="doc_text">
3590 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3594 <p>The '<tt>malloc</tt>' instruction allocates memory from the system heap and
3595 returns a pointer to it. The object is always allocated in the generic
3596 address space (address space zero).</p>
3599 <p>The '<tt>malloc</tt>' instruction allocates
3600 <tt>sizeof(<type>)*NumElements</tt> bytes of memory from the operating
3601 system and returns a pointer of the appropriate type to the program. If
3602 "NumElements" is specified, it is the number of elements allocated, otherwise
3603 "NumElements" is defaulted to be one. If a constant alignment is specified,
3604 the value result of the allocation is guaranteed to be aligned to at least
3605 that boundary. If not specified, or if zero, the target can choose to align
3606 the allocation on any convenient boundary compatible with the type.</p>
3608 <p>'<tt>type</tt>' must be a sized type.</p>
3611 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and a
3612 pointer is returned. The result of a zero byte allocation is undefined. The
3613 result is null if there is insufficient memory available.</p>
3617 %array = malloc [4 x i8] <i>; yields {[%4 x i8]*}:array</i>
3619 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3620 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3621 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3622 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3623 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3626 <p>Note that the code generator does not yet respect the alignment value.</p>
3630 <!-- _______________________________________________________________________ -->
3631 <div class="doc_subsubsection">
3632 <a name="i_free">'<tt>free</tt>' Instruction</a>
3635 <div class="doc_text">
3639 free <type> <value> <i>; yields {void}</i>
3643 <p>The '<tt>free</tt>' instruction returns memory back to the unused memory heap
3644 to be reallocated in the future.</p>
3647 <p>'<tt>value</tt>' shall be a pointer value that points to a value that was
3648 allocated with the '<tt><a href="#i_malloc">malloc</a></tt>' instruction.</p>
3651 <p>Access to the memory pointed to by the pointer is no longer defined after
3652 this instruction executes. If the pointer is null, the operation is a
3657 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3658 free [4 x i8]* %array
3663 <!-- _______________________________________________________________________ -->
3664 <div class="doc_subsubsection">
3665 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3668 <div class="doc_text">
3672 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3676 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3677 currently executing function, to be automatically released when this function
3678 returns to its caller. The object is always allocated in the generic address
3679 space (address space zero).</p>
3682 <p>The '<tt>alloca</tt>' instruction
3683 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
3684 runtime stack, returning a pointer of the appropriate type to the program.
3685 If "NumElements" is specified, it is the number of elements allocated,
3686 otherwise "NumElements" is defaulted to be one. If a constant alignment is
3687 specified, the value result of the allocation is guaranteed to be aligned to
3688 at least that boundary. If not specified, or if zero, the target can choose
3689 to align the allocation on any convenient boundary compatible with the
3692 <p>'<tt>type</tt>' may be any sized type.</p>
3695 <p>Memory is allocated; a pointer is returned. The operation is undefined if
3696 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3697 memory is automatically released when the function returns. The
3698 '<tt>alloca</tt>' instruction is commonly used to represent automatic
3699 variables that must have an address available. When the function returns
3700 (either with the <tt><a href="#i_ret">ret</a></tt>
3701 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
3702 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
3706 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3707 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3708 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3709 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3714 <!-- _______________________________________________________________________ -->
3715 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3716 Instruction</a> </div>
3718 <div class="doc_text">
3722 <result> = load <ty>* <pointer>[, align <alignment>]
3723 <result> = volatile load <ty>* <pointer>[, align <alignment>]
3727 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3730 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
3731 from which to load. The pointer must point to
3732 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3733 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
3734 number or order of execution of this <tt>load</tt> with other
3735 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3738 <p>The optional constant "align" argument specifies the alignment of the
3739 operation (that is, the alignment of the memory address). A value of 0 or an
3740 omitted "align" argument means that the operation has the preferential
3741 alignment for the target. It is the responsibility of the code emitter to
3742 ensure that the alignment information is correct. Overestimating the
3743 alignment results in an undefined behavior. Underestimating the alignment may
3744 produce less efficient code. An alignment of 1 is always safe.</p>
3747 <p>The location of memory pointed to is loaded. If the value being loaded is of
3748 scalar type then the number of bytes read does not exceed the minimum number
3749 of bytes needed to hold all bits of the type. For example, loading an
3750 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
3751 <tt>i20</tt> with a size that is not an integral number of bytes, the result
3752 is undefined if the value was not originally written using a store of the
3757 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3758 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3759 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3764 <!-- _______________________________________________________________________ -->
3765 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3766 Instruction</a> </div>
3768 <div class="doc_text">
3772 store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3773 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3777 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3780 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
3781 and an address at which to store it. The type of the
3782 '<tt><pointer></tt>' operand must be a pointer to
3783 the <a href="#t_firstclass">first class</a> type of the
3784 '<tt><value></tt>' operand. If the <tt>store</tt> is marked
3785 as <tt>volatile</tt>, then the optimizer is not allowed to modify the number
3786 or order of execution of this <tt>store</tt> with other
3787 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3790 <p>The optional constant "align" argument specifies the alignment of the
3791 operation (that is, the alignment of the memory address). A value of 0 or an
3792 omitted "align" argument means that the operation has the preferential
3793 alignment for the target. It is the responsibility of the code emitter to
3794 ensure that the alignment information is correct. Overestimating the
3795 alignment results in an undefined behavior. Underestimating the alignment may
3796 produce less efficient code. An alignment of 1 is always safe.</p>
3799 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
3800 location specified by the '<tt><pointer></tt>' operand. If
3801 '<tt><value></tt>' is of scalar type then the number of bytes written
3802 does not exceed the minimum number of bytes needed to hold all bits of the
3803 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
3804 writing a value of a type like <tt>i20</tt> with a size that is not an
3805 integral number of bytes, it is unspecified what happens to the extra bits
3806 that do not belong to the type, but they will typically be overwritten.</p>
3810 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3811 store i32 3, i32* %ptr <i>; yields {void}</i>
3812 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3817 <!-- _______________________________________________________________________ -->
3818 <div class="doc_subsubsection">
3819 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3822 <div class="doc_text">
3826 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
3830 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
3831 subelement of an aggregate data structure. It performs address calculation
3832 only and does not access memory.</p>
3835 <p>The first argument is always a pointer, and forms the basis of the
3836 calculation. The remaining arguments are indices, that indicate which of the
3837 elements of the aggregate object are indexed. The interpretation of each
3838 index is dependent on the type being indexed into. The first index always
3839 indexes the pointer value given as the first argument, the second index
3840 indexes a value of the type pointed to (not necessarily the value directly
3841 pointed to, since the first index can be non-zero), etc. The first type
3842 indexed into must be a pointer value, subsequent types can be arrays, vectors
3843 and structs. Note that subsequent types being indexed into can never be
3844 pointers, since that would require loading the pointer before continuing
3847 <p>The type of each index argument depends on the type it is indexing into.
3848 When indexing into a (packed) structure, only <tt>i32</tt> integer
3849 <b>constants</b> are allowed. When indexing into an array, pointer or
3850 vector, integers of any width are allowed (also non-constants).</p>
3852 <p>For example, let's consider a C code fragment and how it gets compiled to
3855 <div class="doc_code">
3868 int *foo(struct ST *s) {
3869 return &s[1].Z.B[5][13];
3874 <p>The LLVM code generated by the GCC frontend is:</p>
3876 <div class="doc_code">
3878 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
3879 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
3881 define i32* %foo(%ST* %s) {
3883 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3890 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3891 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3892 }</tt>' type, a structure. The second index indexes into the third element
3893 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3894 i8 }</tt>' type, another structure. The third index indexes into the second
3895 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3896 array. The two dimensions of the array are subscripted into, yielding an
3897 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
3898 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3900 <p>Note that it is perfectly legal to index partially through a structure,
3901 returning a pointer to an inner element. Because of this, the LLVM code for
3902 the given testcase is equivalent to:</p>
3905 define i32* %foo(%ST* %s) {
3906 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3907 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3908 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3909 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3910 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3915 <p>Note that it is undefined to access an array out of bounds: array and pointer
3916 indexes must always be within the defined bounds of the array type when
3917 accessed with an instruction that dereferences the pointer (e.g. a load or
3918 store instruction). The one exception for this rule is zero length arrays.
3919 These arrays are defined to be accessible as variable length arrays, which
3920 requires access beyond the zero'th element.</p>
3922 <p>The getelementptr instruction is often confusing. For some more insight into
3923 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
3927 <i>; yields [12 x i8]*:aptr</i>
3928 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
3929 <i>; yields i8*:vptr</i>
3930 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
3931 <i>; yields i8*:eptr</i>
3932 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
3933 <i>; yields i32*:iptr</i>
3934 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
3939 <!-- ======================================================================= -->
3940 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3943 <div class="doc_text">
3945 <p>The instructions in this category are the conversion instructions (casting)
3946 which all take a single operand and a type. They perform various bit
3947 conversions on the operand.</p>
3951 <!-- _______________________________________________________________________ -->
3952 <div class="doc_subsubsection">
3953 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3955 <div class="doc_text">
3959 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3963 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
3964 type <tt>ty2</tt>.</p>
3967 <p>The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3968 be an <a href="#t_integer">integer</a> type, and a type that specifies the
3969 size and type of the result, which must be
3970 an <a href="#t_integer">integer</a> type. The bit size of <tt>value</tt> must
3971 be larger than the bit size of <tt>ty2</tt>. Equal sized types are not
3975 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
3976 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
3977 source size must be larger than the destination size, <tt>trunc</tt> cannot
3978 be a <i>no-op cast</i>. It will always truncate bits.</p>
3982 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3983 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3984 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3989 <!-- _______________________________________________________________________ -->
3990 <div class="doc_subsubsection">
3991 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3993 <div class="doc_text">
3997 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
4001 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
4006 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
4007 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4008 also be of <a href="#t_integer">integer</a> type. The bit size of the
4009 <tt>value</tt> must be smaller than the bit size of the destination type,
4013 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
4014 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
4016 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
4020 %X = zext i32 257 to i64 <i>; yields i64:257</i>
4021 %Y = zext i1 true to i32 <i>; yields i32:1</i>
4026 <!-- _______________________________________________________________________ -->
4027 <div class="doc_subsubsection">
4028 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
4030 <div class="doc_text">
4034 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
4038 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
4041 <p>The '<tt>sext</tt>' instruction takes a value to cast, which must be of
4042 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4043 also be of <a href="#t_integer">integer</a> type. The bit size of the
4044 <tt>value</tt> must be smaller than the bit size of the destination type,
4048 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
4049 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
4050 of the type <tt>ty2</tt>.</p>
4052 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
4056 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
4057 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
4062 <!-- _______________________________________________________________________ -->
4063 <div class="doc_subsubsection">
4064 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
4067 <div class="doc_text">
4071 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
4075 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
4079 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
4080 point</a> value to cast and a <a href="#t_floating">floating point</a> type
4081 to cast it to. The size of <tt>value</tt> must be larger than the size of
4082 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
4083 <i>no-op cast</i>.</p>
4086 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
4087 <a href="#t_floating">floating point</a> type to a smaller
4088 <a href="#t_floating">floating point</a> type. If the value cannot fit
4089 within the destination type, <tt>ty2</tt>, then the results are
4094 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
4095 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
4100 <!-- _______________________________________________________________________ -->
4101 <div class="doc_subsubsection">
4102 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
4104 <div class="doc_text">
4108 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
4112 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
4113 floating point value.</p>
4116 <p>The '<tt>fpext</tt>' instruction takes a
4117 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
4118 a <a href="#t_floating">floating point</a> type to cast it to. The source
4119 type must be smaller than the destination type.</p>
4122 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
4123 <a href="#t_floating">floating point</a> type to a larger
4124 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
4125 used to make a <i>no-op cast</i> because it always changes bits. Use
4126 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
4130 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
4131 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
4136 <!-- _______________________________________________________________________ -->
4137 <div class="doc_subsubsection">
4138 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
4140 <div class="doc_text">
4144 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
4148 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
4149 unsigned integer equivalent of type <tt>ty2</tt>.</p>
4152 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
4153 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4154 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4155 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4156 vector integer type with the same number of elements as <tt>ty</tt></p>
4159 <p>The '<tt>fptoui</tt>' instruction converts its
4160 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4161 towards zero) unsigned integer value. If the value cannot fit
4162 in <tt>ty2</tt>, the results are undefined.</p>
4166 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
4167 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
4168 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
4173 <!-- _______________________________________________________________________ -->
4174 <div class="doc_subsubsection">
4175 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
4177 <div class="doc_text">
4181 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
4185 <p>The '<tt>fptosi</tt>' instruction converts
4186 <a href="#t_floating">floating point</a> <tt>value</tt> to
4187 type <tt>ty2</tt>.</p>
4190 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
4191 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4192 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4193 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4194 vector integer type with the same number of elements as <tt>ty</tt></p>
4197 <p>The '<tt>fptosi</tt>' instruction converts its
4198 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4199 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
4200 the results are undefined.</p>
4204 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
4205 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
4206 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
4211 <!-- _______________________________________________________________________ -->
4212 <div class="doc_subsubsection">
4213 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
4215 <div class="doc_text">
4219 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4223 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4224 integer and converts that value to the <tt>ty2</tt> type.</p>
4227 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4228 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4229 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4230 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4231 floating point type with the same number of elements as <tt>ty</tt></p>
4234 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4235 integer quantity and converts it to the corresponding floating point
4236 value. If the value cannot fit in the floating point value, the results are
4241 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4242 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4247 <!-- _______________________________________________________________________ -->
4248 <div class="doc_subsubsection">
4249 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4251 <div class="doc_text">
4255 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4259 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
4260 and converts that value to the <tt>ty2</tt> type.</p>
4263 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4264 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4265 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4266 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4267 floating point type with the same number of elements as <tt>ty</tt></p>
4270 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
4271 quantity and converts it to the corresponding floating point value. If the
4272 value cannot fit in the floating point value, the results are undefined.</p>
4276 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4277 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4282 <!-- _______________________________________________________________________ -->
4283 <div class="doc_subsubsection">
4284 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4286 <div class="doc_text">
4290 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4294 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4295 the integer type <tt>ty2</tt>.</p>
4298 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4299 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4300 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4303 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4304 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4305 truncating or zero extending that value to the size of the integer type. If
4306 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4307 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4308 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4313 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4314 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4319 <!-- _______________________________________________________________________ -->
4320 <div class="doc_subsubsection">
4321 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4323 <div class="doc_text">
4327 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4331 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
4332 pointer type, <tt>ty2</tt>.</p>
4335 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4336 value to cast, and a type to cast it to, which must be a
4337 <a href="#t_pointer">pointer</a> type.</p>
4340 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4341 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4342 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4343 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
4344 than the size of a pointer then a zero extension is done. If they are the
4345 same size, nothing is done (<i>no-op cast</i>).</p>
4349 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4350 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4351 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4356 <!-- _______________________________________________________________________ -->
4357 <div class="doc_subsubsection">
4358 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4360 <div class="doc_text">
4364 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4368 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4369 <tt>ty2</tt> without changing any bits.</p>
4372 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
4373 non-aggregate first class value, and a type to cast it to, which must also be
4374 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
4375 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
4376 identical. If the source type is a pointer, the destination type must also be
4377 a pointer. This instruction supports bitwise conversion of vectors to
4378 integers and to vectors of other types (as long as they have the same
4382 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4383 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4384 this conversion. The conversion is done as if the <tt>value</tt> had been
4385 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
4386 be converted to other pointer types with this instruction. To convert
4387 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
4388 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4392 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4393 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4394 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4399 <!-- ======================================================================= -->
4400 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4402 <div class="doc_text">
4404 <p>The instructions in this category are the "miscellaneous" instructions, which
4405 defy better classification.</p>
4409 <!-- _______________________________________________________________________ -->
4410 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4413 <div class="doc_text">
4417 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4421 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
4422 boolean values based on comparison of its two integer, integer vector, or
4423 pointer operands.</p>
4426 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4427 the condition code indicating the kind of comparison to perform. It is not a
4428 value, just a keyword. The possible condition code are:</p>
4431 <li><tt>eq</tt>: equal</li>
4432 <li><tt>ne</tt>: not equal </li>
4433 <li><tt>ugt</tt>: unsigned greater than</li>
4434 <li><tt>uge</tt>: unsigned greater or equal</li>
4435 <li><tt>ult</tt>: unsigned less than</li>
4436 <li><tt>ule</tt>: unsigned less or equal</li>
4437 <li><tt>sgt</tt>: signed greater than</li>
4438 <li><tt>sge</tt>: signed greater or equal</li>
4439 <li><tt>slt</tt>: signed less than</li>
4440 <li><tt>sle</tt>: signed less or equal</li>
4443 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4444 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
4445 typed. They must also be identical types.</p>
4448 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
4449 condition code given as <tt>cond</tt>. The comparison performed always yields
4450 either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt>
4451 result, as follows:</p>
4454 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4455 <tt>false</tt> otherwise. No sign interpretation is necessary or
4458 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4459 <tt>false</tt> otherwise. No sign interpretation is necessary or
4462 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4463 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4465 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4466 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4467 to <tt>op2</tt>.</li>
4469 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4470 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4472 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4473 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4475 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4476 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4478 <li><tt>sge</tt>: interprets the operands as signed values and yields
4479 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4480 to <tt>op2</tt>.</li>
4482 <li><tt>slt</tt>: interprets the operands as signed values and yields
4483 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4485 <li><tt>sle</tt>: interprets the operands as signed values and yields
4486 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4489 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4490 values are compared as if they were integers.</p>
4492 <p>If the operands are integer vectors, then they are compared element by
4493 element. The result is an <tt>i1</tt> vector with the same number of elements
4494 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
4498 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4499 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4500 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4501 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4502 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4503 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4506 <p>Note that the code generator does not yet support vector types with
4507 the <tt>icmp</tt> instruction.</p>
4511 <!-- _______________________________________________________________________ -->
4512 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4515 <div class="doc_text">
4519 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4523 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
4524 values based on comparison of its operands.</p>
4526 <p>If the operands are floating point scalars, then the result type is a boolean
4527 (<a href="#t_primitive"><tt>i1</tt></a>).</p>
4529 <p>If the operands are floating point vectors, then the result type is a vector
4530 of boolean with the same number of elements as the operands being
4534 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4535 the condition code indicating the kind of comparison to perform. It is not a
4536 value, just a keyword. The possible condition code are:</p>
4539 <li><tt>false</tt>: no comparison, always returns false</li>
4540 <li><tt>oeq</tt>: ordered and equal</li>
4541 <li><tt>ogt</tt>: ordered and greater than </li>
4542 <li><tt>oge</tt>: ordered and greater than or equal</li>
4543 <li><tt>olt</tt>: ordered and less than </li>
4544 <li><tt>ole</tt>: ordered and less than or equal</li>
4545 <li><tt>one</tt>: ordered and not equal</li>
4546 <li><tt>ord</tt>: ordered (no nans)</li>
4547 <li><tt>ueq</tt>: unordered or equal</li>
4548 <li><tt>ugt</tt>: unordered or greater than </li>
4549 <li><tt>uge</tt>: unordered or greater than or equal</li>
4550 <li><tt>ult</tt>: unordered or less than </li>
4551 <li><tt>ule</tt>: unordered or less than or equal</li>
4552 <li><tt>une</tt>: unordered or not equal</li>
4553 <li><tt>uno</tt>: unordered (either nans)</li>
4554 <li><tt>true</tt>: no comparison, always returns true</li>
4557 <p><i>Ordered</i> means that neither operand is a QNAN while
4558 <i>unordered</i> means that either operand may be a QNAN.</p>
4560 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
4561 a <a href="#t_floating">floating point</a> type or
4562 a <a href="#t_vector">vector</a> of floating point type. They must have
4563 identical types.</p>
4566 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4567 according to the condition code given as <tt>cond</tt>. If the operands are
4568 vectors, then the vectors are compared element by element. Each comparison
4569 performed always yields an <a href="#t_primitive">i1</a> result, as
4573 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4575 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4576 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4578 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4579 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4581 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4582 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4584 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4585 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4587 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4588 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4590 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4591 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4593 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4595 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4596 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4598 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4599 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4601 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4602 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4604 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4605 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4607 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4608 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4610 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4611 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4613 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4615 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4620 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4621 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4622 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4623 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4626 <p>Note that the code generator does not yet support vector types with
4627 the <tt>fcmp</tt> instruction.</p>
4631 <!-- _______________________________________________________________________ -->
4632 <div class="doc_subsubsection">
4633 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4636 <div class="doc_text">
4640 <result> = phi <ty> [ <val0>, <label0>], ...
4644 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
4645 SSA graph representing the function.</p>
4648 <p>The type of the incoming values is specified with the first type field. After
4649 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
4650 one pair for each predecessor basic block of the current block. Only values
4651 of <a href="#t_firstclass">first class</a> type may be used as the value
4652 arguments to the PHI node. Only labels may be used as the label
4655 <p>There must be no non-phi instructions between the start of a basic block and
4656 the PHI instructions: i.e. PHI instructions must be first in a basic
4659 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
4660 occur on the edge from the corresponding predecessor block to the current
4661 block (but after any definition of an '<tt>invoke</tt>' instruction's return
4662 value on the same edge).</p>
4665 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4666 specified by the pair corresponding to the predecessor basic block that
4667 executed just prior to the current block.</p>
4671 Loop: ; Infinite loop that counts from 0 on up...
4672 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4673 %nextindvar = add i32 %indvar, 1
4679 <!-- _______________________________________________________________________ -->
4680 <div class="doc_subsubsection">
4681 <a name="i_select">'<tt>select</tt>' Instruction</a>
4684 <div class="doc_text">
4688 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4690 <i>selty</i> is either i1 or {<N x i1>}
4694 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
4695 condition, without branching.</p>
4699 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
4700 values indicating the condition, and two values of the
4701 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
4702 vectors and the condition is a scalar, then entire vectors are selected, not
4703 individual elements.</p>
4706 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
4707 first value argument; otherwise, it returns the second value argument.</p>
4709 <p>If the condition is a vector of i1, then the value arguments must be vectors
4710 of the same size, and the selection is done element by element.</p>
4714 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4717 <p>Note that the code generator does not yet support conditions
4718 with vector type.</p>
4722 <!-- _______________________________________________________________________ -->
4723 <div class="doc_subsubsection">
4724 <a name="i_call">'<tt>call</tt>' Instruction</a>
4727 <div class="doc_text">
4731 <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>]
4735 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4738 <p>This instruction requires several arguments:</p>
4741 <li>The optional "tail" marker indicates whether the callee function accesses
4742 any allocas or varargs in the caller. If the "tail" marker is present,
4743 the function call is eligible for tail call optimization. Note that calls
4744 may be marked "tail" even if they do not occur before
4745 a <a href="#i_ret"><tt>ret</tt></a> instruction.</li>
4747 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
4748 convention</a> the call should use. If none is specified, the call
4749 defaults to using C calling conventions.</li>
4751 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4752 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
4753 '<tt>inreg</tt>' attributes are valid here.</li>
4755 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
4756 type of the return value. Functions that return no value are marked
4757 <tt><a href="#t_void">void</a></tt>.</li>
4759 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
4760 being invoked. The argument types must match the types implied by this
4761 signature. This type can be omitted if the function is not varargs and if
4762 the function type does not return a pointer to a function.</li>
4764 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4765 be invoked. In most cases, this is a direct function invocation, but
4766 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4767 to function value.</li>
4769 <li>'<tt>function args</tt>': argument list whose types match the function
4770 signature argument types. All arguments must be of
4771 <a href="#t_firstclass">first class</a> type. If the function signature
4772 indicates the function accepts a variable number of arguments, the extra
4773 arguments can be specified.</li>
4775 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
4776 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4777 '<tt>readnone</tt>' attributes are valid here.</li>
4781 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
4782 a specified function, with its incoming arguments bound to the specified
4783 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
4784 function, control flow continues with the instruction after the function
4785 call, and the return value of the function is bound to the result
4790 %retval = call i32 @test(i32 %argc)
4791 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4792 %X = tail call i32 @foo() <i>; yields i32</i>
4793 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4794 call void %foo(i8 97 signext)
4796 %struct.A = type { i32, i8 }
4797 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4798 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4799 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
4800 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
4801 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
4806 <!-- _______________________________________________________________________ -->
4807 <div class="doc_subsubsection">
4808 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4811 <div class="doc_text">
4815 <resultval> = va_arg <va_list*> <arglist>, <argty>
4819 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4820 the "variable argument" area of a function call. It is used to implement the
4821 <tt>va_arg</tt> macro in C.</p>
4824 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
4825 argument. It returns a value of the specified argument type and increments
4826 the <tt>va_list</tt> to point to the next argument. The actual type
4827 of <tt>va_list</tt> is target specific.</p>
4830 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
4831 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
4832 to the next argument. For more information, see the variable argument
4833 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
4835 <p>It is legal for this instruction to be called in a function which does not
4836 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4839 <p><tt>va_arg</tt> is an LLVM instruction instead of
4840 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
4844 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4846 <p>Note that the code generator does not yet fully support va_arg on many
4847 targets. Also, it does not currently support va_arg with aggregate types on
4852 <!-- *********************************************************************** -->
4853 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4854 <!-- *********************************************************************** -->
4856 <div class="doc_text">
4858 <p>LLVM supports the notion of an "intrinsic function". These functions have
4859 well known names and semantics and are required to follow certain
4860 restrictions. Overall, these intrinsics represent an extension mechanism for
4861 the LLVM language that does not require changing all of the transformations
4862 in LLVM when adding to the language (or the bitcode reader/writer, the
4863 parser, etc...).</p>
4865 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4866 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4867 begin with this prefix. Intrinsic functions must always be external
4868 functions: you cannot define the body of intrinsic functions. Intrinsic
4869 functions may only be used in call or invoke instructions: it is illegal to
4870 take the address of an intrinsic function. Additionally, because intrinsic
4871 functions are part of the LLVM language, it is required if any are added that
4872 they be documented here.</p>
4874 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
4875 family of functions that perform the same operation but on different data
4876 types. Because LLVM can represent over 8 million different integer types,
4877 overloading is used commonly to allow an intrinsic function to operate on any
4878 integer type. One or more of the argument types or the result type can be
4879 overloaded to accept any integer type. Argument types may also be defined as
4880 exactly matching a previous argument's type or the result type. This allows
4881 an intrinsic function which accepts multiple arguments, but needs all of them
4882 to be of the same type, to only be overloaded with respect to a single
4883 argument or the result.</p>
4885 <p>Overloaded intrinsics will have the names of its overloaded argument types
4886 encoded into its function name, each preceded by a period. Only those types
4887 which are overloaded result in a name suffix. Arguments whose type is matched
4888 against another type do not. For example, the <tt>llvm.ctpop</tt> function
4889 can take an integer of any width and returns an integer of exactly the same
4890 integer width. This leads to a family of functions such as
4891 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
4892 %val)</tt>. Only one type, the return type, is overloaded, and only one type
4893 suffix is required. Because the argument's type is matched against the return
4894 type, it does not require its own name suffix.</p>
4896 <p>To learn how to add an intrinsic function, please see the
4897 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
4901 <!-- ======================================================================= -->
4902 <div class="doc_subsection">
4903 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4906 <div class="doc_text">
4908 <p>Variable argument support is defined in LLVM with
4909 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4910 intrinsic functions. These functions are related to the similarly named
4911 macros defined in the <tt><stdarg.h></tt> header file.</p>
4913 <p>All of these functions operate on arguments that use a target-specific value
4914 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
4915 not define what this type is, so all transformations should be prepared to
4916 handle these functions regardless of the type used.</p>
4918 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4919 instruction and the variable argument handling intrinsic functions are
4922 <div class="doc_code">
4924 define i32 @test(i32 %X, ...) {
4925 ; Initialize variable argument processing
4927 %ap2 = bitcast i8** %ap to i8*
4928 call void @llvm.va_start(i8* %ap2)
4930 ; Read a single integer argument
4931 %tmp = va_arg i8** %ap, i32
4933 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4935 %aq2 = bitcast i8** %aq to i8*
4936 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4937 call void @llvm.va_end(i8* %aq2)
4939 ; Stop processing of arguments.
4940 call void @llvm.va_end(i8* %ap2)
4944 declare void @llvm.va_start(i8*)
4945 declare void @llvm.va_copy(i8*, i8*)
4946 declare void @llvm.va_end(i8*)
4952 <!-- _______________________________________________________________________ -->
4953 <div class="doc_subsubsection">
4954 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4958 <div class="doc_text">
4962 declare void %llvm.va_start(i8* <arglist>)
4966 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
4967 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
4970 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4973 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4974 macro available in C. In a target-dependent way, it initializes
4975 the <tt>va_list</tt> element to which the argument points, so that the next
4976 call to <tt>va_arg</tt> will produce the first variable argument passed to
4977 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
4978 need to know the last argument of the function as the compiler can figure
4983 <!-- _______________________________________________________________________ -->
4984 <div class="doc_subsubsection">
4985 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4988 <div class="doc_text">
4992 declare void @llvm.va_end(i8* <arglist>)
4996 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4997 which has been initialized previously
4998 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4999 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
5002 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
5005 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
5006 macro available in C. In a target-dependent way, it destroys
5007 the <tt>va_list</tt> element to which the argument points. Calls
5008 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
5009 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
5010 with calls to <tt>llvm.va_end</tt>.</p>
5014 <!-- _______________________________________________________________________ -->
5015 <div class="doc_subsubsection">
5016 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
5019 <div class="doc_text">
5023 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
5027 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
5028 from the source argument list to the destination argument list.</p>
5031 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
5032 The second argument is a pointer to a <tt>va_list</tt> element to copy
5036 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
5037 macro available in C. In a target-dependent way, it copies the
5038 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
5039 element. This intrinsic is necessary because
5040 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
5041 arbitrarily complex and require, for example, memory allocation.</p>
5045 <!-- ======================================================================= -->
5046 <div class="doc_subsection">
5047 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
5050 <div class="doc_text">
5052 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
5053 Collection</a> (GC) requires the implementation and generation of these
5054 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
5055 roots on the stack</a>, as well as garbage collector implementations that
5056 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
5057 barriers. Front-ends for type-safe garbage collected languages should generate
5058 these intrinsics to make use of the LLVM garbage collectors. For more details,
5059 see <a href="GarbageCollection.html">Accurate Garbage Collection with
5062 <p>The garbage collection intrinsics only operate on objects in the generic
5063 address space (address space zero).</p>
5067 <!-- _______________________________________________________________________ -->
5068 <div class="doc_subsubsection">
5069 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
5072 <div class="doc_text">
5076 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5080 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
5081 the code generator, and allows some metadata to be associated with it.</p>
5084 <p>The first argument specifies the address of a stack object that contains the
5085 root pointer. The second pointer (which must be either a constant or a
5086 global value address) contains the meta-data to be associated with the
5090 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
5091 location. At compile-time, the code generator generates information to allow
5092 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
5093 intrinsic may only be used in a function which <a href="#gc">specifies a GC
5098 <!-- _______________________________________________________________________ -->
5099 <div class="doc_subsubsection">
5100 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
5103 <div class="doc_text">
5107 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5111 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
5112 locations, allowing garbage collector implementations that require read
5116 <p>The second argument is the address to read from, which should be an address
5117 allocated from the garbage collector. The first object is a pointer to the
5118 start of the referenced object, if needed by the language runtime (otherwise
5122 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
5123 instruction, but may be replaced with substantially more complex code by the
5124 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
5125 may only be used in a function which <a href="#gc">specifies a GC
5130 <!-- _______________________________________________________________________ -->
5131 <div class="doc_subsubsection">
5132 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5135 <div class="doc_text">
5139 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5143 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5144 locations, allowing garbage collector implementations that require write
5145 barriers (such as generational or reference counting collectors).</p>
5148 <p>The first argument is the reference to store, the second is the start of the
5149 object to store it to, and the third is the address of the field of Obj to
5150 store to. If the runtime does not require a pointer to the object, Obj may
5154 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5155 instruction, but may be replaced with substantially more complex code by the
5156 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5157 may only be used in a function which <a href="#gc">specifies a GC
5162 <!-- ======================================================================= -->
5163 <div class="doc_subsection">
5164 <a name="int_codegen">Code Generator Intrinsics</a>
5167 <div class="doc_text">
5169 <p>These intrinsics are provided by LLVM to expose special features that may
5170 only be implemented with code generator support.</p>
5174 <!-- _______________________________________________________________________ -->
5175 <div class="doc_subsubsection">
5176 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5179 <div class="doc_text">
5183 declare i8 *@llvm.returnaddress(i32 <level>)
5187 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5188 target-specific value indicating the return address of the current function
5189 or one of its callers.</p>
5192 <p>The argument to this intrinsic indicates which function to return the address
5193 for. Zero indicates the calling function, one indicates its caller, etc.
5194 The argument is <b>required</b> to be a constant integer value.</p>
5197 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
5198 indicating the return address of the specified call frame, or zero if it
5199 cannot be identified. The value returned by this intrinsic is likely to be
5200 incorrect or 0 for arguments other than zero, so it should only be used for
5201 debugging purposes.</p>
5203 <p>Note that calling this intrinsic does not prevent function inlining or other
5204 aggressive transformations, so the value returned may not be that of the
5205 obvious source-language caller.</p>
5209 <!-- _______________________________________________________________________ -->
5210 <div class="doc_subsubsection">
5211 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5214 <div class="doc_text">
5218 declare i8 *@llvm.frameaddress(i32 <level>)
5222 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5223 target-specific frame pointer value for the specified stack frame.</p>
5226 <p>The argument to this intrinsic indicates which function to return the frame
5227 pointer for. Zero indicates the calling function, one indicates its caller,
5228 etc. The argument is <b>required</b> to be a constant integer value.</p>
5231 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
5232 indicating the frame address of the specified call frame, or zero if it
5233 cannot be identified. The value returned by this intrinsic is likely to be
5234 incorrect or 0 for arguments other than zero, so it should only be used for
5235 debugging purposes.</p>
5237 <p>Note that calling this intrinsic does not prevent function inlining or other
5238 aggressive transformations, so the value returned may not be that of the
5239 obvious source-language caller.</p>
5243 <!-- _______________________________________________________________________ -->
5244 <div class="doc_subsubsection">
5245 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5248 <div class="doc_text">
5252 declare i8 *@llvm.stacksave()
5256 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
5257 of the function stack, for use
5258 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
5259 useful for implementing language features like scoped automatic variable
5260 sized arrays in C99.</p>
5263 <p>This intrinsic returns a opaque pointer value that can be passed
5264 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
5265 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
5266 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
5267 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
5268 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
5269 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
5273 <!-- _______________________________________________________________________ -->
5274 <div class="doc_subsubsection">
5275 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5278 <div class="doc_text">
5282 declare void @llvm.stackrestore(i8 * %ptr)
5286 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5287 the function stack to the state it was in when the
5288 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
5289 executed. This is useful for implementing language features like scoped
5290 automatic variable sized arrays in C99.</p>
5293 <p>See the description
5294 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
5298 <!-- _______________________________________________________________________ -->
5299 <div class="doc_subsubsection">
5300 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5303 <div class="doc_text">
5307 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5311 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
5312 insert a prefetch instruction if supported; otherwise, it is a noop.
5313 Prefetches have no effect on the behavior of the program but can change its
5314 performance characteristics.</p>
5317 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
5318 specifier determining if the fetch should be for a read (0) or write (1),
5319 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5320 locality, to (3) - extremely local keep in cache. The <tt>rw</tt>
5321 and <tt>locality</tt> arguments must be constant integers.</p>
5324 <p>This intrinsic does not modify the behavior of the program. In particular,
5325 prefetches cannot trap and do not produce a value. On targets that support
5326 this intrinsic, the prefetch can provide hints to the processor cache for
5327 better performance.</p>
5331 <!-- _______________________________________________________________________ -->
5332 <div class="doc_subsubsection">
5333 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5336 <div class="doc_text">
5340 declare void @llvm.pcmarker(i32 <id>)
5344 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
5345 Counter (PC) in a region of code to simulators and other tools. The method
5346 is target specific, but it is expected that the marker will use exported
5347 symbols to transmit the PC of the marker. The marker makes no guarantees
5348 that it will remain with any specific instruction after optimizations. It is
5349 possible that the presence of a marker will inhibit optimizations. The
5350 intended use is to be inserted after optimizations to allow correlations of
5351 simulation runs.</p>
5354 <p><tt>id</tt> is a numerical id identifying the marker.</p>
5357 <p>This intrinsic does not modify the behavior of the program. Backends that do
5358 not support this intrinisic may ignore it.</p>
5362 <!-- _______________________________________________________________________ -->
5363 <div class="doc_subsubsection">
5364 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5367 <div class="doc_text">
5371 declare i64 @llvm.readcyclecounter( )
5375 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5376 counter register (or similar low latency, high accuracy clocks) on those
5377 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5378 should map to RPCC. As the backing counters overflow quickly (on the order
5379 of 9 seconds on alpha), this should only be used for small timings.</p>
5382 <p>When directly supported, reading the cycle counter should not modify any
5383 memory. Implementations are allowed to either return a application specific
5384 value or a system wide value. On backends without support, this is lowered
5385 to a constant 0.</p>
5389 <!-- ======================================================================= -->
5390 <div class="doc_subsection">
5391 <a name="int_libc">Standard C Library Intrinsics</a>
5394 <div class="doc_text">
5396 <p>LLVM provides intrinsics for a few important standard C library functions.
5397 These intrinsics allow source-language front-ends to pass information about
5398 the alignment of the pointer arguments to the code generator, providing
5399 opportunity for more efficient code generation.</p>
5403 <!-- _______________________________________________________________________ -->
5404 <div class="doc_subsubsection">
5405 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5408 <div class="doc_text">
5411 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
5412 integer bit width. Not all targets support all bit widths however.</p>
5415 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5416 i8 <len>, i32 <align>)
5417 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5418 i16 <len>, i32 <align>)
5419 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5420 i32 <len>, i32 <align>)
5421 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5422 i64 <len>, i32 <align>)
5426 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5427 source location to the destination location.</p>
5429 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5430 intrinsics do not return a value, and takes an extra alignment argument.</p>
5433 <p>The first argument is a pointer to the destination, the second is a pointer
5434 to the source. The third argument is an integer argument specifying the
5435 number of bytes to copy, and the fourth argument is the alignment of the
5436 source and destination locations.</p>
5438 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5439 then the caller guarantees that both the source and destination pointers are
5440 aligned to that boundary.</p>
5443 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5444 source location to the destination location, which are not allowed to
5445 overlap. It copies "len" bytes of memory over. If the argument is known to
5446 be aligned to some boundary, this can be specified as the fourth argument,
5447 otherwise it should be set to 0 or 1.</p>
5451 <!-- _______________________________________________________________________ -->
5452 <div class="doc_subsubsection">
5453 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5456 <div class="doc_text">
5459 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5460 width. Not all targets support all bit widths however.</p>
5463 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5464 i8 <len>, i32 <align>)
5465 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5466 i16 <len>, i32 <align>)
5467 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5468 i32 <len>, i32 <align>)
5469 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5470 i64 <len>, i32 <align>)
5474 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
5475 source location to the destination location. It is similar to the
5476 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
5479 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5480 intrinsics do not return a value, and takes an extra alignment argument.</p>
5483 <p>The first argument is a pointer to the destination, the second is a pointer
5484 to the source. The third argument is an integer argument specifying the
5485 number of bytes to copy, and the fourth argument is the alignment of the
5486 source and destination locations.</p>
5488 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5489 then the caller guarantees that the source and destination pointers are
5490 aligned to that boundary.</p>
5493 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
5494 source location to the destination location, which may overlap. It copies
5495 "len" bytes of memory over. If the argument is known to be aligned to some
5496 boundary, this can be specified as the fourth argument, otherwise it should
5497 be set to 0 or 1.</p>
5501 <!-- _______________________________________________________________________ -->
5502 <div class="doc_subsubsection">
5503 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5506 <div class="doc_text">
5509 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5510 width. Not all targets support all bit widths however.</p>
5513 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5514 i8 <len>, i32 <align>)
5515 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5516 i16 <len>, i32 <align>)
5517 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5518 i32 <len>, i32 <align>)
5519 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5520 i64 <len>, i32 <align>)
5524 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
5525 particular byte value.</p>
5527 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
5528 intrinsic does not return a value, and takes an extra alignment argument.</p>
5531 <p>The first argument is a pointer to the destination to fill, the second is the
5532 byte value to fill it with, the third argument is an integer argument
5533 specifying the number of bytes to fill, and the fourth argument is the known
5534 alignment of destination location.</p>
5536 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5537 then the caller guarantees that the destination pointer is aligned to that
5541 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
5542 at the destination location. If the argument is known to be aligned to some
5543 boundary, this can be specified as the fourth argument, otherwise it should
5544 be set to 0 or 1.</p>
5548 <!-- _______________________________________________________________________ -->
5549 <div class="doc_subsubsection">
5550 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5553 <div class="doc_text">
5556 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5557 floating point or vector of floating point type. Not all targets support all
5561 declare float @llvm.sqrt.f32(float %Val)
5562 declare double @llvm.sqrt.f64(double %Val)
5563 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5564 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5565 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5569 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5570 returning the same value as the libm '<tt>sqrt</tt>' functions would.
5571 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
5572 behavior for negative numbers other than -0.0 (which allows for better
5573 optimization, because there is no need to worry about errno being
5574 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
5577 <p>The argument and return value are floating point numbers of the same
5581 <p>This function returns the sqrt of the specified operand if it is a
5582 nonnegative floating point number.</p>
5586 <!-- _______________________________________________________________________ -->
5587 <div class="doc_subsubsection">
5588 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5591 <div class="doc_text">
5594 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5595 floating point or vector of floating point type. Not all targets support all
5599 declare float @llvm.powi.f32(float %Val, i32 %power)
5600 declare double @llvm.powi.f64(double %Val, i32 %power)
5601 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5602 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5603 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5607 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5608 specified (positive or negative) power. The order of evaluation of
5609 multiplications is not defined. When a vector of floating point type is
5610 used, the second argument remains a scalar integer value.</p>
5613 <p>The second argument is an integer power, and the first is a value to raise to
5617 <p>This function returns the first value raised to the second power with an
5618 unspecified sequence of rounding operations.</p>
5622 <!-- _______________________________________________________________________ -->
5623 <div class="doc_subsubsection">
5624 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5627 <div class="doc_text">
5630 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5631 floating point or vector of floating point type. Not all targets support all
5635 declare float @llvm.sin.f32(float %Val)
5636 declare double @llvm.sin.f64(double %Val)
5637 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5638 declare fp128 @llvm.sin.f128(fp128 %Val)
5639 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5643 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
5646 <p>The argument and return value are floating point numbers of the same
5650 <p>This function returns the sine of the specified operand, returning the same
5651 values as the libm <tt>sin</tt> functions would, and handles error conditions
5652 in the same way.</p>
5656 <!-- _______________________________________________________________________ -->
5657 <div class="doc_subsubsection">
5658 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5661 <div class="doc_text">
5664 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5665 floating point or vector of floating point type. Not all targets support all
5669 declare float @llvm.cos.f32(float %Val)
5670 declare double @llvm.cos.f64(double %Val)
5671 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5672 declare fp128 @llvm.cos.f128(fp128 %Val)
5673 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5677 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
5680 <p>The argument and return value are floating point numbers of the same
5684 <p>This function returns the cosine of the specified operand, returning the same
5685 values as the libm <tt>cos</tt> functions would, and handles error conditions
5686 in the same way.</p>
5690 <!-- _______________________________________________________________________ -->
5691 <div class="doc_subsubsection">
5692 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5695 <div class="doc_text">
5698 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5699 floating point or vector of floating point type. Not all targets support all
5703 declare float @llvm.pow.f32(float %Val, float %Power)
5704 declare double @llvm.pow.f64(double %Val, double %Power)
5705 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5706 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5707 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5711 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5712 specified (positive or negative) power.</p>
5715 <p>The second argument is a floating point power, and the first is a value to
5716 raise to that power.</p>
5719 <p>This function returns the first value raised to the second power, returning
5720 the same values as the libm <tt>pow</tt> functions would, and handles error
5721 conditions in the same way.</p>
5725 <!-- ======================================================================= -->
5726 <div class="doc_subsection">
5727 <a name="int_manip">Bit Manipulation Intrinsics</a>
5730 <div class="doc_text">
5732 <p>LLVM provides intrinsics for a few important bit manipulation operations.
5733 These allow efficient code generation for some algorithms.</p>
5737 <!-- _______________________________________________________________________ -->
5738 <div class="doc_subsubsection">
5739 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5742 <div class="doc_text">
5745 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5746 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
5749 declare i16 @llvm.bswap.i16(i16 <id>)
5750 declare i32 @llvm.bswap.i32(i32 <id>)
5751 declare i64 @llvm.bswap.i64(i64 <id>)
5755 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5756 values with an even number of bytes (positive multiple of 16 bits). These
5757 are useful for performing operations on data that is not in the target's
5758 native byte order.</p>
5761 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5762 and low byte of the input i16 swapped. Similarly,
5763 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
5764 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
5765 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
5766 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
5767 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
5768 more, respectively).</p>
5772 <!-- _______________________________________________________________________ -->
5773 <div class="doc_subsubsection">
5774 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5777 <div class="doc_text">
5780 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5781 width. Not all targets support all bit widths however.</p>
5784 declare i8 @llvm.ctpop.i8(i8 <src>)
5785 declare i16 @llvm.ctpop.i16(i16 <src>)
5786 declare i32 @llvm.ctpop.i32(i32 <src>)
5787 declare i64 @llvm.ctpop.i64(i64 <src>)
5788 declare i256 @llvm.ctpop.i256(i256 <src>)
5792 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
5796 <p>The only argument is the value to be counted. The argument may be of any
5797 integer type. The return type must match the argument type.</p>
5800 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.</p>
5804 <!-- _______________________________________________________________________ -->
5805 <div class="doc_subsubsection">
5806 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5809 <div class="doc_text">
5812 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5813 integer bit width. Not all targets support all bit widths however.</p>
5816 declare i8 @llvm.ctlz.i8 (i8 <src>)
5817 declare i16 @llvm.ctlz.i16(i16 <src>)
5818 declare i32 @llvm.ctlz.i32(i32 <src>)
5819 declare i64 @llvm.ctlz.i64(i64 <src>)
5820 declare i256 @llvm.ctlz.i256(i256 <src>)
5824 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5825 leading zeros in a variable.</p>
5828 <p>The only argument is the value to be counted. The argument may be of any
5829 integer type. The return type must match the argument type.</p>
5832 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
5833 zeros in a variable. If the src == 0 then the result is the size in bits of
5834 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
5838 <!-- _______________________________________________________________________ -->
5839 <div class="doc_subsubsection">
5840 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5843 <div class="doc_text">
5846 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5847 integer bit width. Not all targets support all bit widths however.</p>
5850 declare i8 @llvm.cttz.i8 (i8 <src>)
5851 declare i16 @llvm.cttz.i16(i16 <src>)
5852 declare i32 @llvm.cttz.i32(i32 <src>)
5853 declare i64 @llvm.cttz.i64(i64 <src>)
5854 declare i256 @llvm.cttz.i256(i256 <src>)
5858 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5862 <p>The only argument is the value to be counted. The argument may be of any
5863 integer type. The return type must match the argument type.</p>
5866 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
5867 zeros in a variable. If the src == 0 then the result is the size in bits of
5868 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
5872 <!-- ======================================================================= -->
5873 <div class="doc_subsection">
5874 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
5877 <div class="doc_text">
5879 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
5883 <!-- _______________________________________________________________________ -->
5884 <div class="doc_subsubsection">
5885 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
5888 <div class="doc_text">
5891 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
5892 on any integer bit width.</p>
5895 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
5896 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
5897 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
5901 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
5902 a signed addition of the two arguments, and indicate whether an overflow
5903 occurred during the signed summation.</p>
5906 <p>The arguments (%a and %b) and the first element of the result structure may
5907 be of integer types of any bit width, but they must have the same bit
5908 width. The second element of the result structure must be of
5909 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
5910 undergo signed addition.</p>
5913 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
5914 a signed addition of the two variables. They return a structure — the
5915 first element of which is the signed summation, and the second element of
5916 which is a bit specifying if the signed summation resulted in an
5921 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
5922 %sum = extractvalue {i32, i1} %res, 0
5923 %obit = extractvalue {i32, i1} %res, 1
5924 br i1 %obit, label %overflow, label %normal
5929 <!-- _______________________________________________________________________ -->
5930 <div class="doc_subsubsection">
5931 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
5934 <div class="doc_text">
5937 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
5938 on any integer bit width.</p>
5941 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
5942 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
5943 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
5947 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
5948 an unsigned addition of the two arguments, and indicate whether a carry
5949 occurred during the unsigned summation.</p>
5952 <p>The arguments (%a and %b) and the first element of the result structure may
5953 be of integer types of any bit width, but they must have the same bit
5954 width. The second element of the result structure must be of
5955 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
5956 undergo unsigned addition.</p>
5959 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
5960 an unsigned addition of the two arguments. They return a structure —
5961 the first element of which is the sum, and the second element of which is a
5962 bit specifying if the unsigned summation resulted in a carry.</p>
5966 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
5967 %sum = extractvalue {i32, i1} %res, 0
5968 %obit = extractvalue {i32, i1} %res, 1
5969 br i1 %obit, label %carry, label %normal
5974 <!-- _______________________________________________________________________ -->
5975 <div class="doc_subsubsection">
5976 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
5979 <div class="doc_text">
5982 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
5983 on any integer bit width.</p>
5986 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
5987 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
5988 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
5992 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
5993 a signed subtraction of the two arguments, and indicate whether an overflow
5994 occurred during the signed subtraction.</p>
5997 <p>The arguments (%a and %b) and the first element of the result structure may
5998 be of integer types of any bit width, but they must have the same bit
5999 width. The second element of the result structure must be of
6000 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6001 undergo signed subtraction.</p>
6004 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6005 a signed subtraction of the two arguments. They return a structure —
6006 the first element of which is the subtraction, and the second element of
6007 which is a bit specifying if the signed subtraction resulted in an
6012 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6013 %sum = extractvalue {i32, i1} %res, 0
6014 %obit = extractvalue {i32, i1} %res, 1
6015 br i1 %obit, label %overflow, label %normal
6020 <!-- _______________________________________________________________________ -->
6021 <div class="doc_subsubsection">
6022 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6025 <div class="doc_text">
6028 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6029 on any integer bit width.</p>
6032 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6033 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6034 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6038 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6039 an unsigned subtraction of the two arguments, and indicate whether an
6040 overflow occurred during the unsigned subtraction.</p>
6043 <p>The arguments (%a and %b) and the first element of the result structure may
6044 be of integer types of any bit width, but they must have the same bit
6045 width. The second element of the result structure must be of
6046 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6047 undergo unsigned subtraction.</p>
6050 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6051 an unsigned subtraction of the two arguments. They return a structure —
6052 the first element of which is the subtraction, and the second element of
6053 which is a bit specifying if the unsigned subtraction resulted in an
6058 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6059 %sum = extractvalue {i32, i1} %res, 0
6060 %obit = extractvalue {i32, i1} %res, 1
6061 br i1 %obit, label %overflow, label %normal
6066 <!-- _______________________________________________________________________ -->
6067 <div class="doc_subsubsection">
6068 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6071 <div class="doc_text">
6074 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6075 on any integer bit width.</p>
6078 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6079 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6080 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6085 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6086 a signed multiplication of the two arguments, and indicate whether an
6087 overflow occurred during the signed multiplication.</p>
6090 <p>The arguments (%a and %b) and the first element of the result structure may
6091 be of integer types of any bit width, but they must have the same bit
6092 width. The second element of the result structure must be of
6093 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6094 undergo signed multiplication.</p>
6097 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6098 a signed multiplication of the two arguments. They return a structure —
6099 the first element of which is the multiplication, and the second element of
6100 which is a bit specifying if the signed multiplication resulted in an
6105 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6106 %sum = extractvalue {i32, i1} %res, 0
6107 %obit = extractvalue {i32, i1} %res, 1
6108 br i1 %obit, label %overflow, label %normal
6113 <!-- _______________________________________________________________________ -->
6114 <div class="doc_subsubsection">
6115 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6118 <div class="doc_text">
6121 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6122 on any integer bit width.</p>
6125 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6126 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6127 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6131 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6132 a unsigned multiplication of the two arguments, and indicate whether an
6133 overflow occurred during the unsigned multiplication.</p>
6136 <p>The arguments (%a and %b) and the first element of the result structure may
6137 be of integer types of any bit width, but they must have the same bit
6138 width. The second element of the result structure must be of
6139 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6140 undergo unsigned multiplication.</p>
6143 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6144 an unsigned multiplication of the two arguments. They return a structure
6145 — the first element of which is the multiplication, and the second
6146 element of which is a bit specifying if the unsigned multiplication resulted
6151 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6152 %sum = extractvalue {i32, i1} %res, 0
6153 %obit = extractvalue {i32, i1} %res, 1
6154 br i1 %obit, label %overflow, label %normal
6159 <!-- ======================================================================= -->
6160 <div class="doc_subsection">
6161 <a name="int_debugger">Debugger Intrinsics</a>
6164 <div class="doc_text">
6166 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
6167 prefix), are described in
6168 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
6169 Level Debugging</a> document.</p>
6173 <!-- ======================================================================= -->
6174 <div class="doc_subsection">
6175 <a name="int_eh">Exception Handling Intrinsics</a>
6178 <div class="doc_text">
6180 <p>The LLVM exception handling intrinsics (which all start with
6181 <tt>llvm.eh.</tt> prefix), are described in
6182 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6183 Handling</a> document.</p>
6187 <!-- ======================================================================= -->
6188 <div class="doc_subsection">
6189 <a name="int_trampoline">Trampoline Intrinsic</a>
6192 <div class="doc_text">
6194 <p>This intrinsic makes it possible to excise one parameter, marked with
6195 the <tt>nest</tt> attribute, from a function. The result is a callable
6196 function pointer lacking the nest parameter - the caller does not need to
6197 provide a value for it. Instead, the value to use is stored in advance in a
6198 "trampoline", a block of memory usually allocated on the stack, which also
6199 contains code to splice the nest value into the argument list. This is used
6200 to implement the GCC nested function address extension.</p>
6202 <p>For example, if the function is
6203 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6204 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
6207 <div class="doc_code">
6209 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6210 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6211 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6212 %fp = bitcast i8* %p to i32 (i32, i32)*
6216 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6217 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6221 <!-- _______________________________________________________________________ -->
6222 <div class="doc_subsubsection">
6223 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6226 <div class="doc_text">
6230 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6234 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
6235 function pointer suitable for executing it.</p>
6238 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6239 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
6240 sufficiently aligned block of memory; this memory is written to by the
6241 intrinsic. Note that the size and the alignment are target-specific - LLVM
6242 currently provides no portable way of determining them, so a front-end that
6243 generates this intrinsic needs to have some target-specific knowledge.
6244 The <tt>func</tt> argument must hold a function bitcast to
6245 an <tt>i8*</tt>.</p>
6248 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
6249 dependent code, turning it into a function. A pointer to this function is
6250 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
6251 function pointer type</a> before being called. The new function's signature
6252 is the same as that of <tt>func</tt> with any arguments marked with
6253 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
6254 is allowed, and it must be of pointer type. Calling the new function is
6255 equivalent to calling <tt>func</tt> with the same argument list, but
6256 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
6257 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
6258 by <tt>tramp</tt> is modified, then the effect of any later call to the
6259 returned function pointer is undefined.</p>
6263 <!-- ======================================================================= -->
6264 <div class="doc_subsection">
6265 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6268 <div class="doc_text">
6270 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
6271 hardware constructs for atomic operations and memory synchronization. This
6272 provides an interface to the hardware, not an interface to the programmer. It
6273 is aimed at a low enough level to allow any programming models or APIs
6274 (Application Programming Interfaces) which need atomic behaviors to map
6275 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6276 hardware provides a "universal IR" for source languages, it also provides a
6277 starting point for developing a "universal" atomic operation and
6278 synchronization IR.</p>
6280 <p>These do <em>not</em> form an API such as high-level threading libraries,
6281 software transaction memory systems, atomic primitives, and intrinsic
6282 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6283 application libraries. The hardware interface provided by LLVM should allow
6284 a clean implementation of all of these APIs and parallel programming models.
6285 No one model or paradigm should be selected above others unless the hardware
6286 itself ubiquitously does so.</p>
6290 <!-- _______________________________________________________________________ -->
6291 <div class="doc_subsubsection">
6292 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6294 <div class="doc_text">
6297 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device> )
6301 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6302 specific pairs of memory access types.</p>
6305 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6306 The first four arguments enables a specific barrier as listed below. The
6307 fith argument specifies that the barrier applies to io or device or uncached
6311 <li><tt>ll</tt>: load-load barrier</li>
6312 <li><tt>ls</tt>: load-store barrier</li>
6313 <li><tt>sl</tt>: store-load barrier</li>
6314 <li><tt>ss</tt>: store-store barrier</li>
6315 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6319 <p>This intrinsic causes the system to enforce some ordering constraints upon
6320 the loads and stores of the program. This barrier does not
6321 indicate <em>when</em> any events will occur, it only enforces
6322 an <em>order</em> in which they occur. For any of the specified pairs of load
6323 and store operations (f.ex. load-load, or store-load), all of the first
6324 operations preceding the barrier will complete before any of the second
6325 operations succeeding the barrier begin. Specifically the semantics for each
6326 pairing is as follows:</p>
6329 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6330 after the barrier begins.</li>
6331 <li><tt>ls</tt>: All loads before the barrier must complete before any
6332 store after the barrier begins.</li>
6333 <li><tt>ss</tt>: All stores before the barrier must complete before any
6334 store after the barrier begins.</li>
6335 <li><tt>sl</tt>: All stores before the barrier must complete before any
6336 load after the barrier begins.</li>
6339 <p>These semantics are applied with a logical "and" behavior when more than one
6340 is enabled in a single memory barrier intrinsic.</p>
6342 <p>Backends may implement stronger barriers than those requested when they do
6343 not support as fine grained a barrier as requested. Some architectures do
6344 not need all types of barriers and on such architectures, these become
6352 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6353 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6354 <i>; guarantee the above finishes</i>
6355 store i32 8, %ptr <i>; before this begins</i>
6360 <!-- _______________________________________________________________________ -->
6361 <div class="doc_subsubsection">
6362 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6365 <div class="doc_text">
6368 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6369 any integer bit width and for different address spaces. Not all targets
6370 support all bit widths however.</p>
6373 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6374 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6375 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6376 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6380 <p>This loads a value in memory and compares it to a given value. If they are
6381 equal, it stores a new value into the memory.</p>
6384 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
6385 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6386 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6387 this integer type. While any bit width integer may be used, targets may only
6388 lower representations they support in hardware.</p>
6391 <p>This entire intrinsic must be executed atomically. It first loads the value
6392 in memory pointed to by <tt>ptr</tt> and compares it with the
6393 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
6394 memory. The loaded value is yielded in all cases. This provides the
6395 equivalent of an atomic compare-and-swap operation within the SSA
6403 %val1 = add i32 4, 4
6404 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6405 <i>; yields {i32}:result1 = 4</i>
6406 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6407 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6409 %val2 = add i32 1, 1
6410 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6411 <i>; yields {i32}:result2 = 8</i>
6412 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6414 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6419 <!-- _______________________________________________________________________ -->
6420 <div class="doc_subsubsection">
6421 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6423 <div class="doc_text">
6426 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6427 integer bit width. Not all targets support all bit widths however.</p>
6430 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6431 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6432 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6433 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6437 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6438 the value from memory. It then stores the value in <tt>val</tt> in the memory
6439 at <tt>ptr</tt>.</p>
6442 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
6443 the <tt>val</tt> argument and the result must be integers of the same bit
6444 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6445 integer type. The targets may only lower integer representations they
6449 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6450 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6451 equivalent of an atomic swap operation within the SSA framework.</p>
6458 %val1 = add i32 4, 4
6459 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6460 <i>; yields {i32}:result1 = 4</i>
6461 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6462 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6464 %val2 = add i32 1, 1
6465 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6466 <i>; yields {i32}:result2 = 8</i>
6468 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6469 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6474 <!-- _______________________________________________________________________ -->
6475 <div class="doc_subsubsection">
6476 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6480 <div class="doc_text">
6483 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
6484 any integer bit width. Not all targets support all bit widths however.</p>
6487 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6488 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6489 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6490 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6494 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
6495 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6498 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6499 and the second an integer value. The result is also an integer value. These
6500 integer types can have any bit width, but they must all have the same bit
6501 width. The targets may only lower integer representations they support.</p>
6504 <p>This intrinsic does a series of operations atomically. It first loads the
6505 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6506 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
6512 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6513 <i>; yields {i32}:result1 = 4</i>
6514 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6515 <i>; yields {i32}:result2 = 8</i>
6516 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6517 <i>; yields {i32}:result3 = 10</i>
6518 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6523 <!-- _______________________________________________________________________ -->
6524 <div class="doc_subsubsection">
6525 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6529 <div class="doc_text">
6532 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6533 any integer bit width and for different address spaces. Not all targets
6534 support all bit widths however.</p>
6537 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6538 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6539 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6540 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6544 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6545 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6548 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6549 and the second an integer value. The result is also an integer value. These
6550 integer types can have any bit width, but they must all have the same bit
6551 width. The targets may only lower integer representations they support.</p>
6554 <p>This intrinsic does a series of operations atomically. It first loads the
6555 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6556 result to <tt>ptr</tt>. It yields the original value stored
6557 at <tt>ptr</tt>.</p>
6563 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6564 <i>; yields {i32}:result1 = 8</i>
6565 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6566 <i>; yields {i32}:result2 = 4</i>
6567 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6568 <i>; yields {i32}:result3 = 2</i>
6569 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6574 <!-- _______________________________________________________________________ -->
6575 <div class="doc_subsubsection">
6576 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6577 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6578 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6579 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6582 <div class="doc_text">
6585 <p>These are overloaded intrinsics. You can
6586 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
6587 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
6588 bit width and for different address spaces. Not all targets support all bit
6592 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6593 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6594 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6595 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6599 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6600 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6601 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6602 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6606 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6607 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6608 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6609 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6613 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6614 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6615 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6616 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6620 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6621 the value stored in memory at <tt>ptr</tt>. It yields the original value
6622 at <tt>ptr</tt>.</p>
6625 <p>These intrinsics take two arguments, the first a pointer to an integer value
6626 and the second an integer value. The result is also an integer value. These
6627 integer types can have any bit width, but they must all have the same bit
6628 width. The targets may only lower integer representations they support.</p>
6631 <p>These intrinsics does a series of operations atomically. They first load the
6632 value stored at <tt>ptr</tt>. They then do the bitwise
6633 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
6634 original value stored at <tt>ptr</tt>.</p>
6639 store i32 0x0F0F, %ptr
6640 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6641 <i>; yields {i32}:result0 = 0x0F0F</i>
6642 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6643 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6644 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6645 <i>; yields {i32}:result2 = 0xF0</i>
6646 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6647 <i>; yields {i32}:result3 = FF</i>
6648 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6653 <!-- _______________________________________________________________________ -->
6654 <div class="doc_subsubsection">
6655 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6656 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6657 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6658 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6661 <div class="doc_text">
6664 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6665 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6666 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6667 address spaces. Not all targets support all bit widths however.</p>
6670 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6671 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6672 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6673 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6677 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6678 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6679 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6680 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6684 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6685 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6686 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6687 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6691 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6692 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6693 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6694 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6698 <p>These intrinsics takes the signed or unsigned minimum or maximum of
6699 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6700 original value at <tt>ptr</tt>.</p>
6703 <p>These intrinsics take two arguments, the first a pointer to an integer value
6704 and the second an integer value. The result is also an integer value. These
6705 integer types can have any bit width, but they must all have the same bit
6706 width. The targets may only lower integer representations they support.</p>
6709 <p>These intrinsics does a series of operations atomically. They first load the
6710 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
6711 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
6712 yield the original value stored at <tt>ptr</tt>.</p>
6718 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6719 <i>; yields {i32}:result0 = 7</i>
6720 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6721 <i>; yields {i32}:result1 = -2</i>
6722 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6723 <i>; yields {i32}:result2 = 8</i>
6724 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6725 <i>; yields {i32}:result3 = 8</i>
6726 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6731 <!-- ======================================================================= -->
6732 <div class="doc_subsection">
6733 <a name="int_general">General Intrinsics</a>
6736 <div class="doc_text">
6738 <p>This class of intrinsics is designed to be generic and has no specific
6743 <!-- _______________________________________________________________________ -->
6744 <div class="doc_subsubsection">
6745 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6748 <div class="doc_text">
6752 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6756 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
6759 <p>The first argument is a pointer to a value, the second is a pointer to a
6760 global string, the third is a pointer to a global string which is the source
6761 file name, and the last argument is the line number.</p>
6764 <p>This intrinsic allows annotation of local variables with arbitrary strings.
6765 This can be useful for special purpose optimizations that want to look for
6766 these annotations. These have no other defined use, they are ignored by code
6767 generation and optimization.</p>
6771 <!-- _______________________________________________________________________ -->
6772 <div class="doc_subsubsection">
6773 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6776 <div class="doc_text">
6779 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6780 any integer bit width.</p>
6783 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6784 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6785 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6786 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6787 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6791 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
6794 <p>The first argument is an integer value (result of some expression), the
6795 second is a pointer to a global string, the third is a pointer to a global
6796 string which is the source file name, and the last argument is the line
6797 number. It returns the value of the first argument.</p>
6800 <p>This intrinsic allows annotations to be put on arbitrary expressions with
6801 arbitrary strings. This can be useful for special purpose optimizations that
6802 want to look for these annotations. These have no other defined use, they
6803 are ignored by code generation and optimization.</p>
6807 <!-- _______________________________________________________________________ -->
6808 <div class="doc_subsubsection">
6809 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6812 <div class="doc_text">
6816 declare void @llvm.trap()
6820 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
6826 <p>This intrinsics is lowered to the target dependent trap instruction. If the
6827 target does not have a trap instruction, this intrinsic will be lowered to
6828 the call of the <tt>abort()</tt> function.</p>
6832 <!-- _______________________________________________________________________ -->
6833 <div class="doc_subsubsection">
6834 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
6837 <div class="doc_text">
6841 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
6845 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
6846 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
6847 ensure that it is placed on the stack before local variables.</p>
6850 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
6851 arguments. The first argument is the value loaded from the stack
6852 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
6853 that has enough space to hold the value of the guard.</p>
6856 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
6857 the <tt>AllocaInst</tt> stack slot to be before local variables on the
6858 stack. This is to ensure that if a local variable on the stack is
6859 overwritten, it will destroy the value of the guard. When the function exits,
6860 the guard on the stack is checked against the original guard. If they're
6861 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
6866 <!-- *********************************************************************** -->
6869 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6870 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
6871 <a href="http://validator.w3.org/check/referer"><img
6872 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
6874 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6875 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6876 Last modified: $Date$