1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a>
25 <li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
26 <li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
27 <li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
28 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
29 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
30 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
31 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
32 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
33 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
34 <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
35 <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
36 <li><a href="#linkage_external">'<tt>externally visible</tt>' Linkage</a></li>
37 <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
38 <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
41 <li><a href="#callingconv">Calling Conventions</a></li>
42 <li><a href="#namedtypes">Named Types</a></li>
43 <li><a href="#globalvars">Global Variables</a></li>
44 <li><a href="#functionstructure">Functions</a></li>
45 <li><a href="#aliasstructure">Aliases</a></li>
46 <li><a href="#namedmetadatastructure">Named Metadata</a></li>
47 <li><a href="#paramattrs">Parameter Attributes</a></li>
48 <li><a href="#fnattrs">Function Attributes</a></li>
49 <li><a href="#gc">Garbage Collector Names</a></li>
50 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
51 <li><a href="#datalayout">Data Layout</a></li>
52 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
53 <li><a href="#volatile">Volatile Memory Accesses</a></li>
56 <li><a href="#typesystem">Type System</a>
58 <li><a href="#t_classifications">Type Classifications</a></li>
59 <li><a href="#t_primitive">Primitive Types</a>
61 <li><a href="#t_integer">Integer Type</a></li>
62 <li><a href="#t_floating">Floating Point Types</a></li>
63 <li><a href="#t_void">Void Type</a></li>
64 <li><a href="#t_label">Label Type</a></li>
65 <li><a href="#t_metadata">Metadata Type</a></li>
68 <li><a href="#t_derived">Derived Types</a>
70 <li><a href="#t_aggregate">Aggregate Types</a>
72 <li><a href="#t_array">Array Type</a></li>
73 <li><a href="#t_struct">Structure Type</a></li>
74 <li><a href="#t_pstruct">Packed Structure Type</a></li>
75 <li><a href="#t_union">Union Type</a></li>
76 <li><a href="#t_vector">Vector Type</a></li>
79 <li><a href="#t_function">Function Type</a></li>
80 <li><a href="#t_pointer">Pointer Type</a></li>
81 <li><a href="#t_opaque">Opaque Type</a></li>
84 <li><a href="#t_uprefs">Type Up-references</a></li>
87 <li><a href="#constants">Constants</a>
89 <li><a href="#simpleconstants">Simple Constants</a></li>
90 <li><a href="#complexconstants">Complex Constants</a></li>
91 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
92 <li><a href="#undefvalues">Undefined Values</a></li>
93 <li><a href="#trapvalues">Trap Values</a></li>
94 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
95 <li><a href="#constantexprs">Constant Expressions</a></li>
98 <li><a href="#othervalues">Other Values</a>
100 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
101 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a></li>
104 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
106 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
107 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
108 Global Variable</a></li>
109 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
110 Global Variable</a></li>
111 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
112 Global Variable</a></li>
115 <li><a href="#instref">Instruction Reference</a>
117 <li><a href="#terminators">Terminator Instructions</a>
119 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
120 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
121 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
122 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
123 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
124 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
125 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
128 <li><a href="#binaryops">Binary Operations</a>
130 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
131 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
132 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
133 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
134 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
135 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
136 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
137 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
138 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
139 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
140 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
141 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
144 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
146 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
147 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
148 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
149 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
150 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
151 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
154 <li><a href="#vectorops">Vector Operations</a>
156 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
157 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
158 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
161 <li><a href="#aggregateops">Aggregate Operations</a>
163 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
164 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
167 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
169 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
170 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
171 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
172 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
175 <li><a href="#convertops">Conversion Operations</a>
177 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
178 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
179 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
180 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
181 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
182 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
183 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
184 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
185 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
186 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
187 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
188 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
191 <li><a href="#otherops">Other Operations</a>
193 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
194 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
195 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
196 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
197 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
198 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
203 <li><a href="#intrinsics">Intrinsic Functions</a>
205 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
207 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
208 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
209 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
212 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
214 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
215 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
216 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
219 <li><a href="#int_codegen">Code Generator Intrinsics</a>
221 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
222 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
223 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
224 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
225 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
226 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
227 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
230 <li><a href="#int_libc">Standard C Library Intrinsics</a>
232 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
233 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
234 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
235 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
236 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
237 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
238 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
239 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
242 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
244 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
245 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
246 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
247 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
250 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
252 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
253 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
254 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
255 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
256 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
257 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
260 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
262 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
263 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
266 <li><a href="#int_debugger">Debugger intrinsics</a></li>
267 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
268 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
270 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
273 <li><a href="#int_atomics">Atomic intrinsics</a>
275 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
276 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
277 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
278 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
279 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
280 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
281 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
282 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
283 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
284 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
285 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
286 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
287 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
290 <li><a href="#int_memorymarkers">Memory Use Markers</a>
292 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
293 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
294 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
295 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
298 <li><a href="#int_general">General intrinsics</a>
300 <li><a href="#int_var_annotation">
301 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
302 <li><a href="#int_annotation">
303 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
304 <li><a href="#int_trap">
305 '<tt>llvm.trap</tt>' Intrinsic</a></li>
306 <li><a href="#int_stackprotector">
307 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
308 <li><a href="#int_objectsize">
309 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
316 <div class="doc_author">
317 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
318 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
321 <!-- *********************************************************************** -->
322 <div class="doc_section"> <a name="abstract">Abstract </a></div>
323 <!-- *********************************************************************** -->
325 <div class="doc_text">
327 <p>This document is a reference manual for the LLVM assembly language. LLVM is
328 a Static Single Assignment (SSA) based representation that provides type
329 safety, low-level operations, flexibility, and the capability of representing
330 'all' high-level languages cleanly. It is the common code representation
331 used throughout all phases of the LLVM compilation strategy.</p>
335 <!-- *********************************************************************** -->
336 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
337 <!-- *********************************************************************** -->
339 <div class="doc_text">
341 <p>The LLVM code representation is designed to be used in three different forms:
342 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
343 for fast loading by a Just-In-Time compiler), and as a human readable
344 assembly language representation. This allows LLVM to provide a powerful
345 intermediate representation for efficient compiler transformations and
346 analysis, while providing a natural means to debug and visualize the
347 transformations. The three different forms of LLVM are all equivalent. This
348 document describes the human readable representation and notation.</p>
350 <p>The LLVM representation aims to be light-weight and low-level while being
351 expressive, typed, and extensible at the same time. It aims to be a
352 "universal IR" of sorts, by being at a low enough level that high-level ideas
353 may be cleanly mapped to it (similar to how microprocessors are "universal
354 IR's", allowing many source languages to be mapped to them). By providing
355 type information, LLVM can be used as the target of optimizations: for
356 example, through pointer analysis, it can be proven that a C automatic
357 variable is never accessed outside of the current function, allowing it to
358 be promoted to a simple SSA value instead of a memory location.</p>
362 <!-- _______________________________________________________________________ -->
363 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
365 <div class="doc_text">
367 <p>It is important to note that this document describes 'well formed' LLVM
368 assembly language. There is a difference between what the parser accepts and
369 what is considered 'well formed'. For example, the following instruction is
370 syntactically okay, but not well formed:</p>
372 <div class="doc_code">
374 %x = <a href="#i_add">add</a> i32 1, %x
378 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
379 LLVM infrastructure provides a verification pass that may be used to verify
380 that an LLVM module is well formed. This pass is automatically run by the
381 parser after parsing input assembly and by the optimizer before it outputs
382 bitcode. The violations pointed out by the verifier pass indicate bugs in
383 transformation passes or input to the parser.</p>
387 <!-- Describe the typesetting conventions here. -->
389 <!-- *********************************************************************** -->
390 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
391 <!-- *********************************************************************** -->
393 <div class="doc_text">
395 <p>LLVM identifiers come in two basic types: global and local. Global
396 identifiers (functions, global variables) begin with the <tt>'@'</tt>
397 character. Local identifiers (register names, types) begin with
398 the <tt>'%'</tt> character. Additionally, there are three different formats
399 for identifiers, for different purposes:</p>
402 <li>Named values are represented as a string of characters with their prefix.
403 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
404 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
405 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
406 other characters in their names can be surrounded with quotes. Special
407 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
408 ASCII code for the character in hexadecimal. In this way, any character
409 can be used in a name value, even quotes themselves.</li>
411 <li>Unnamed values are represented as an unsigned numeric value with their
412 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
414 <li>Constants, which are described in a <a href="#constants">section about
415 constants</a>, below.</li>
418 <p>LLVM requires that values start with a prefix for two reasons: Compilers
419 don't need to worry about name clashes with reserved words, and the set of
420 reserved words may be expanded in the future without penalty. Additionally,
421 unnamed identifiers allow a compiler to quickly come up with a temporary
422 variable without having to avoid symbol table conflicts.</p>
424 <p>Reserved words in LLVM are very similar to reserved words in other
425 languages. There are keywords for different opcodes
426 ('<tt><a href="#i_add">add</a></tt>',
427 '<tt><a href="#i_bitcast">bitcast</a></tt>',
428 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
429 ('<tt><a href="#t_void">void</a></tt>',
430 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
431 reserved words cannot conflict with variable names, because none of them
432 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
434 <p>Here is an example of LLVM code to multiply the integer variable
435 '<tt>%X</tt>' by 8:</p>
439 <div class="doc_code">
441 %result = <a href="#i_mul">mul</a> i32 %X, 8
445 <p>After strength reduction:</p>
447 <div class="doc_code">
449 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
453 <p>And the hard way:</p>
455 <div class="doc_code">
457 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
458 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
459 %result = <a href="#i_add">add</a> i32 %1, %1
463 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
464 lexical features of LLVM:</p>
467 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
470 <li>Unnamed temporaries are created when the result of a computation is not
471 assigned to a named value.</li>
473 <li>Unnamed temporaries are numbered sequentially</li>
476 <p>It also shows a convention that we follow in this document. When
477 demonstrating instructions, we will follow an instruction with a comment that
478 defines the type and name of value produced. Comments are shown in italic
483 <!-- *********************************************************************** -->
484 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
485 <!-- *********************************************************************** -->
487 <!-- ======================================================================= -->
488 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
491 <div class="doc_text">
493 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
494 of the input programs. Each module consists of functions, global variables,
495 and symbol table entries. Modules may be combined together with the LLVM
496 linker, which merges function (and global variable) definitions, resolves
497 forward declarations, and merges symbol table entries. Here is an example of
498 the "hello world" module:</p>
500 <div class="doc_code">
502 <i>; Declare the string constant as a global constant.</i>
503 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
505 <i>; External declaration of the puts function</i>
506 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
508 <i>; Definition of main function</i>
509 define i32 @main() { <i>; i32()* </i>
510 <i>; Convert [13 x i8]* to i8 *...</i>
511 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
513 <i>; Call puts function to write out the string to stdout.</i>
514 <a href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
515 <a href="#i_ret">ret</a> i32 0<br>}
517 <i>; Named metadata</i>
518 !1 = metadata !{i32 41}
523 <p>This example is made up of a <a href="#globalvars">global variable</a> named
524 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
525 a <a href="#functionstructure">function definition</a> for
526 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
529 <p>In general, a module is made up of a list of global values, where both
530 functions and global variables are global values. Global values are
531 represented by a pointer to a memory location (in this case, a pointer to an
532 array of char, and a pointer to a function), and have one of the
533 following <a href="#linkage">linkage types</a>.</p>
537 <!-- ======================================================================= -->
538 <div class="doc_subsection">
539 <a name="linkage">Linkage Types</a>
542 <div class="doc_text">
544 <p>All Global Variables and Functions have one of the following types of
548 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
549 <dd>Global values with private linkage are only directly accessible by objects
550 in the current module. In particular, linking code into a module with an
551 private global value may cause the private to be renamed as necessary to
552 avoid collisions. Because the symbol is private to the module, all
553 references can be updated. This doesn't show up in any symbol table in the
556 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
557 <dd>Similar to private, but the symbol is passed through the assembler and
558 removed by the linker after evaluation. Note that (unlike private
559 symbols) linker_private symbols are subject to coalescing by the linker:
560 weak symbols get merged and redefinitions are rejected. However, unlike
561 normal strong symbols, they are removed by the linker from the final
562 linked image (executable or dynamic library).</dd>
564 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
565 <dd>Similar to private, but the value shows as a local symbol
566 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
567 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
569 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
570 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
571 into the object file corresponding to the LLVM module. They exist to
572 allow inlining and other optimizations to take place given knowledge of
573 the definition of the global, which is known to be somewhere outside the
574 module. Globals with <tt>available_externally</tt> linkage are allowed to
575 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
576 This linkage type is only allowed on definitions, not declarations.</dd>
578 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
579 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
580 the same name when linkage occurs. This can be used to implement
581 some forms of inline functions, templates, or other code which must be
582 generated in each translation unit that uses it, but where the body may
583 be overridden with a more definitive definition later. Unreferenced
584 <tt>linkonce</tt> globals are allowed to be discarded. Note that
585 <tt>linkonce</tt> linkage does not actually allow the optimizer to
586 inline the body of this function into callers because it doesn't know if
587 this definition of the function is the definitive definition within the
588 program or whether it will be overridden by a stronger definition.
589 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
592 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
593 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
594 <tt>linkonce</tt> linkage, except that unreferenced globals with
595 <tt>weak</tt> linkage may not be discarded. This is used for globals that
596 are declared "weak" in C source code.</dd>
598 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
599 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
600 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
602 Symbols with "<tt>common</tt>" linkage are merged in the same way as
603 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
604 <tt>common</tt> symbols may not have an explicit section,
605 must have a zero initializer, and may not be marked '<a
606 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
607 have common linkage.</dd>
610 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
611 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
612 pointer to array type. When two global variables with appending linkage
613 are linked together, the two global arrays are appended together. This is
614 the LLVM, typesafe, equivalent of having the system linker append together
615 "sections" with identical names when .o files are linked.</dd>
617 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
618 <dd>The semantics of this linkage follow the ELF object file model: the symbol
619 is weak until linked, if not linked, the symbol becomes null instead of
620 being an undefined reference.</dd>
622 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
623 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
624 <dd>Some languages allow differing globals to be merged, such as two functions
625 with different semantics. Other languages, such as <tt>C++</tt>, ensure
626 that only equivalent globals are ever merged (the "one definition rule" -
627 "ODR"). Such languages can use the <tt>linkonce_odr</tt>
628 and <tt>weak_odr</tt> linkage types to indicate that the global will only
629 be merged with equivalent globals. These linkage types are otherwise the
630 same as their non-<tt>odr</tt> versions.</dd>
632 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
633 <dd>If none of the above identifiers are used, the global is externally
634 visible, meaning that it participates in linkage and can be used to
635 resolve external symbol references.</dd>
638 <p>The next two types of linkage are targeted for Microsoft Windows platform
639 only. They are designed to support importing (exporting) symbols from (to)
640 DLLs (Dynamic Link Libraries).</p>
643 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
644 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
645 or variable via a global pointer to a pointer that is set up by the DLL
646 exporting the symbol. On Microsoft Windows targets, the pointer name is
647 formed by combining <code>__imp_</code> and the function or variable
650 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
651 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
652 pointer to a pointer in a DLL, so that it can be referenced with the
653 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
654 name is formed by combining <code>__imp_</code> and the function or
658 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
659 another module defined a "<tt>.LC0</tt>" variable and was linked with this
660 one, one of the two would be renamed, preventing a collision. Since
661 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
662 declarations), they are accessible outside of the current module.</p>
664 <p>It is illegal for a function <i>declaration</i> to have any linkage type
665 other than "externally visible", <tt>dllimport</tt>
666 or <tt>extern_weak</tt>.</p>
668 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
669 or <tt>weak_odr</tt> linkages.</p>
673 <!-- ======================================================================= -->
674 <div class="doc_subsection">
675 <a name="callingconv">Calling Conventions</a>
678 <div class="doc_text">
680 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
681 and <a href="#i_invoke">invokes</a> can all have an optional calling
682 convention specified for the call. The calling convention of any pair of
683 dynamic caller/callee must match, or the behavior of the program is
684 undefined. The following calling conventions are supported by LLVM, and more
685 may be added in the future:</p>
688 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
689 <dd>This calling convention (the default if no other calling convention is
690 specified) matches the target C calling conventions. This calling
691 convention supports varargs function calls and tolerates some mismatch in
692 the declared prototype and implemented declaration of the function (as
695 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
696 <dd>This calling convention attempts to make calls as fast as possible
697 (e.g. by passing things in registers). This calling convention allows the
698 target to use whatever tricks it wants to produce fast code for the
699 target, without having to conform to an externally specified ABI
700 (Application Binary Interface).
701 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
702 when this or the GHC convention is used.</a> This calling convention
703 does not support varargs and requires the prototype of all callees to
704 exactly match the prototype of the function definition.</dd>
706 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
707 <dd>This calling convention attempts to make code in the caller as efficient
708 as possible under the assumption that the call is not commonly executed.
709 As such, these calls often preserve all registers so that the call does
710 not break any live ranges in the caller side. This calling convention
711 does not support varargs and requires the prototype of all callees to
712 exactly match the prototype of the function definition.</dd>
714 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
715 <dd>This calling convention has been implemented specifically for use by the
716 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
717 It passes everything in registers, going to extremes to achieve this by
718 disabling callee save registers. This calling convention should not be
719 used lightly but only for specific situations such as an alternative to
720 the <em>register pinning</em> performance technique often used when
721 implementing functional programming languages.At the moment only X86
722 supports this convention and it has the following limitations:
724 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
725 floating point types are supported.</li>
726 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
727 6 floating point parameters.</li>
729 This calling convention supports
730 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
731 requires both the caller and callee are using it.
734 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
735 <dd>Any calling convention may be specified by number, allowing
736 target-specific calling conventions to be used. Target specific calling
737 conventions start at 64.</dd>
740 <p>More calling conventions can be added/defined on an as-needed basis, to
741 support Pascal conventions or any other well-known target-independent
746 <!-- ======================================================================= -->
747 <div class="doc_subsection">
748 <a name="visibility">Visibility Styles</a>
751 <div class="doc_text">
753 <p>All Global Variables and Functions have one of the following visibility
757 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
758 <dd>On targets that use the ELF object file format, default visibility means
759 that the declaration is visible to other modules and, in shared libraries,
760 means that the declared entity may be overridden. On Darwin, default
761 visibility means that the declaration is visible to other modules. Default
762 visibility corresponds to "external linkage" in the language.</dd>
764 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
765 <dd>Two declarations of an object with hidden visibility refer to the same
766 object if they are in the same shared object. Usually, hidden visibility
767 indicates that the symbol will not be placed into the dynamic symbol
768 table, so no other module (executable or shared library) can reference it
771 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
772 <dd>On ELF, protected visibility indicates that the symbol will be placed in
773 the dynamic symbol table, but that references within the defining module
774 will bind to the local symbol. That is, the symbol cannot be overridden by
780 <!-- ======================================================================= -->
781 <div class="doc_subsection">
782 <a name="namedtypes">Named Types</a>
785 <div class="doc_text">
787 <p>LLVM IR allows you to specify name aliases for certain types. This can make
788 it easier to read the IR and make the IR more condensed (particularly when
789 recursive types are involved). An example of a name specification is:</p>
791 <div class="doc_code">
793 %mytype = type { %mytype*, i32 }
797 <p>You may give a name to any <a href="#typesystem">type</a> except
798 "<a href="t_void">void</a>". Type name aliases may be used anywhere a type
799 is expected with the syntax "%mytype".</p>
801 <p>Note that type names are aliases for the structural type that they indicate,
802 and that you can therefore specify multiple names for the same type. This
803 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
804 uses structural typing, the name is not part of the type. When printing out
805 LLVM IR, the printer will pick <em>one name</em> to render all types of a
806 particular shape. This means that if you have code where two different
807 source types end up having the same LLVM type, that the dumper will sometimes
808 print the "wrong" or unexpected type. This is an important design point and
809 isn't going to change.</p>
813 <!-- ======================================================================= -->
814 <div class="doc_subsection">
815 <a name="globalvars">Global Variables</a>
818 <div class="doc_text">
820 <p>Global variables define regions of memory allocated at compilation time
821 instead of run-time. Global variables may optionally be initialized, may
822 have an explicit section to be placed in, and may have an optional explicit
823 alignment specified. A variable may be defined as "thread_local", which
824 means that it will not be shared by threads (each thread will have a
825 separated copy of the variable). A variable may be defined as a global
826 "constant," which indicates that the contents of the variable
827 will <b>never</b> be modified (enabling better optimization, allowing the
828 global data to be placed in the read-only section of an executable, etc).
829 Note that variables that need runtime initialization cannot be marked
830 "constant" as there is a store to the variable.</p>
832 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
833 constant, even if the final definition of the global is not. This capability
834 can be used to enable slightly better optimization of the program, but
835 requires the language definition to guarantee that optimizations based on the
836 'constantness' are valid for the translation units that do not include the
839 <p>As SSA values, global variables define pointer values that are in scope
840 (i.e. they dominate) all basic blocks in the program. Global variables
841 always define a pointer to their "content" type because they describe a
842 region of memory, and all memory objects in LLVM are accessed through
845 <p>A global variable may be declared to reside in a target-specific numbered
846 address space. For targets that support them, address spaces may affect how
847 optimizations are performed and/or what target instructions are used to
848 access the variable. The default address space is zero. The address space
849 qualifier must precede any other attributes.</p>
851 <p>LLVM allows an explicit section to be specified for globals. If the target
852 supports it, it will emit globals to the section specified.</p>
854 <p>An explicit alignment may be specified for a global. If not present, or if
855 the alignment is set to zero, the alignment of the global is set by the
856 target to whatever it feels convenient. If an explicit alignment is
857 specified, the global is forced to have at least that much alignment. All
858 alignments must be a power of 2.</p>
860 <p>For example, the following defines a global in a numbered address space with
861 an initializer, section, and alignment:</p>
863 <div class="doc_code">
865 @G = addrspace(5) constant float 1.0, section "foo", align 4
872 <!-- ======================================================================= -->
873 <div class="doc_subsection">
874 <a name="functionstructure">Functions</a>
877 <div class="doc_text">
879 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
880 optional <a href="#linkage">linkage type</a>, an optional
881 <a href="#visibility">visibility style</a>, an optional
882 <a href="#callingconv">calling convention</a>, a return type, an optional
883 <a href="#paramattrs">parameter attribute</a> for the return type, a function
884 name, a (possibly empty) argument list (each with optional
885 <a href="#paramattrs">parameter attributes</a>), optional
886 <a href="#fnattrs">function attributes</a>, an optional section, an optional
887 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
888 curly brace, a list of basic blocks, and a closing curly brace.</p>
890 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
891 optional <a href="#linkage">linkage type</a>, an optional
892 <a href="#visibility">visibility style</a>, an optional
893 <a href="#callingconv">calling convention</a>, a return type, an optional
894 <a href="#paramattrs">parameter attribute</a> for the return type, a function
895 name, a possibly empty list of arguments, an optional alignment, and an
896 optional <a href="#gc">garbage collector name</a>.</p>
898 <p>A function definition contains a list of basic blocks, forming the CFG
899 (Control Flow Graph) for the function. Each basic block may optionally start
900 with a label (giving the basic block a symbol table entry), contains a list
901 of instructions, and ends with a <a href="#terminators">terminator</a>
902 instruction (such as a branch or function return).</p>
904 <p>The first basic block in a function is special in two ways: it is immediately
905 executed on entrance to the function, and it is not allowed to have
906 predecessor basic blocks (i.e. there can not be any branches to the entry
907 block of a function). Because the block can have no predecessors, it also
908 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
910 <p>LLVM allows an explicit section to be specified for functions. If the target
911 supports it, it will emit functions to the section specified.</p>
913 <p>An explicit alignment may be specified for a function. If not present, or if
914 the alignment is set to zero, the alignment of the function is set by the
915 target to whatever it feels convenient. If an explicit alignment is
916 specified, the function is forced to have at least that much alignment. All
917 alignments must be a power of 2.</p>
920 <div class="doc_code">
922 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
923 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
924 <ResultType> @<FunctionName> ([argument list])
925 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
926 [<a href="#gc">gc</a>] { ... }
932 <!-- ======================================================================= -->
933 <div class="doc_subsection">
934 <a name="aliasstructure">Aliases</a>
937 <div class="doc_text">
939 <p>Aliases act as "second name" for the aliasee value (which can be either
940 function, global variable, another alias or bitcast of global value). Aliases
941 may have an optional <a href="#linkage">linkage type</a>, and an
942 optional <a href="#visibility">visibility style</a>.</p>
945 <div class="doc_code">
947 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
953 <!-- ======================================================================= -->
954 <div class="doc_subsection">
955 <a name="namedmetadatastructure">Named Metadata</a>
958 <div class="doc_text">
960 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
961 nodes</a> (but not metadata strings) and null are the only valid operands for
962 a named metadata.</p>
965 <div class="doc_code">
967 !1 = metadata !{metadata !"one"}
974 <!-- ======================================================================= -->
975 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
977 <div class="doc_text">
979 <p>The return type and each parameter of a function type may have a set of
980 <i>parameter attributes</i> associated with them. Parameter attributes are
981 used to communicate additional information about the result or parameters of
982 a function. Parameter attributes are considered to be part of the function,
983 not of the function type, so functions with different parameter attributes
984 can have the same function type.</p>
986 <p>Parameter attributes are simple keywords that follow the type specified. If
987 multiple parameter attributes are needed, they are space separated. For
990 <div class="doc_code">
992 declare i32 @printf(i8* noalias nocapture, ...)
993 declare i32 @atoi(i8 zeroext)
994 declare signext i8 @returns_signed_char()
998 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
999 <tt>readonly</tt>) come immediately after the argument list.</p>
1001 <p>Currently, only the following parameter attributes are defined:</p>
1004 <dt><tt><b>zeroext</b></tt></dt>
1005 <dd>This indicates to the code generator that the parameter or return value
1006 should be zero-extended to a 32-bit value by the caller (for a parameter)
1007 or the callee (for a return value).</dd>
1009 <dt><tt><b>signext</b></tt></dt>
1010 <dd>This indicates to the code generator that the parameter or return value
1011 should be sign-extended to a 32-bit value by the caller (for a parameter)
1012 or the callee (for a return value).</dd>
1014 <dt><tt><b>inreg</b></tt></dt>
1015 <dd>This indicates that this parameter or return value should be treated in a
1016 special target-dependent fashion during while emitting code for a function
1017 call or return (usually, by putting it in a register as opposed to memory,
1018 though some targets use it to distinguish between two different kinds of
1019 registers). Use of this attribute is target-specific.</dd>
1021 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1022 <dd>This indicates that the pointer parameter should really be passed by value
1023 to the function. The attribute implies that a hidden copy of the pointee
1024 is made between the caller and the callee, so the callee is unable to
1025 modify the value in the callee. This attribute is only valid on LLVM
1026 pointer arguments. It is generally used to pass structs and arrays by
1027 value, but is also valid on pointers to scalars. The copy is considered
1028 to belong to the caller not the callee (for example,
1029 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1030 <tt>byval</tt> parameters). This is not a valid attribute for return
1031 values. The byval attribute also supports specifying an alignment with
1032 the align attribute. This has a target-specific effect on the code
1033 generator that usually indicates a desired alignment for the synthesized
1036 <dt><tt><b>sret</b></tt></dt>
1037 <dd>This indicates that the pointer parameter specifies the address of a
1038 structure that is the return value of the function in the source program.
1039 This pointer must be guaranteed by the caller to be valid: loads and
1040 stores to the structure may be assumed by the callee to not to trap. This
1041 may only be applied to the first parameter. This is not a valid attribute
1042 for return values. </dd>
1044 <dt><tt><b>noalias</b></tt></dt>
1045 <dd>This indicates that the pointer does not alias any global or any other
1046 parameter. The caller is responsible for ensuring that this is the
1047 case. On a function return value, <tt>noalias</tt> additionally indicates
1048 that the pointer does not alias any other pointers visible to the
1049 caller. For further details, please see the discussion of the NoAlias
1051 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
1054 <dt><tt><b>nocapture</b></tt></dt>
1055 <dd>This indicates that the callee does not make any copies of the pointer
1056 that outlive the callee itself. This is not a valid attribute for return
1059 <dt><tt><b>nest</b></tt></dt>
1060 <dd>This indicates that the pointer parameter can be excised using the
1061 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1062 attribute for return values.</dd>
1067 <!-- ======================================================================= -->
1068 <div class="doc_subsection">
1069 <a name="gc">Garbage Collector Names</a>
1072 <div class="doc_text">
1074 <p>Each function may specify a garbage collector name, which is simply a
1077 <div class="doc_code">
1079 define void @f() gc "name" { ... }
1083 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1084 collector which will cause the compiler to alter its output in order to
1085 support the named garbage collection algorithm.</p>
1089 <!-- ======================================================================= -->
1090 <div class="doc_subsection">
1091 <a name="fnattrs">Function Attributes</a>
1094 <div class="doc_text">
1096 <p>Function attributes are set to communicate additional information about a
1097 function. Function attributes are considered to be part of the function, not
1098 of the function type, so functions with different parameter attributes can
1099 have the same function type.</p>
1101 <p>Function attributes are simple keywords that follow the type specified. If
1102 multiple attributes are needed, they are space separated. For example:</p>
1104 <div class="doc_code">
1106 define void @f() noinline { ... }
1107 define void @f() alwaysinline { ... }
1108 define void @f() alwaysinline optsize { ... }
1109 define void @f() optsize { ... }
1114 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1115 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1116 the backend should forcibly align the stack pointer. Specify the
1117 desired alignment, which must be a power of two, in parentheses.
1119 <dt><tt><b>alwaysinline</b></tt></dt>
1120 <dd>This attribute indicates that the inliner should attempt to inline this
1121 function into callers whenever possible, ignoring any active inlining size
1122 threshold for this caller.</dd>
1124 <dt><tt><b>inlinehint</b></tt></dt>
1125 <dd>This attribute indicates that the source code contained a hint that inlining
1126 this function is desirable (such as the "inline" keyword in C/C++). It
1127 is just a hint; it imposes no requirements on the inliner.</dd>
1129 <dt><tt><b>noinline</b></tt></dt>
1130 <dd>This attribute indicates that the inliner should never inline this
1131 function in any situation. This attribute may not be used together with
1132 the <tt>alwaysinline</tt> attribute.</dd>
1134 <dt><tt><b>optsize</b></tt></dt>
1135 <dd>This attribute suggests that optimization passes and code generator passes
1136 make choices that keep the code size of this function low, and otherwise
1137 do optimizations specifically to reduce code size.</dd>
1139 <dt><tt><b>noreturn</b></tt></dt>
1140 <dd>This function attribute indicates that the function never returns
1141 normally. This produces undefined behavior at runtime if the function
1142 ever does dynamically return.</dd>
1144 <dt><tt><b>nounwind</b></tt></dt>
1145 <dd>This function attribute indicates that the function never returns with an
1146 unwind or exceptional control flow. If the function does unwind, its
1147 runtime behavior is undefined.</dd>
1149 <dt><tt><b>readnone</b></tt></dt>
1150 <dd>This attribute indicates that the function computes its result (or decides
1151 to unwind an exception) based strictly on its arguments, without
1152 dereferencing any pointer arguments or otherwise accessing any mutable
1153 state (e.g. memory, control registers, etc) visible to caller functions.
1154 It does not write through any pointer arguments
1155 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1156 changes any state visible to callers. This means that it cannot unwind
1157 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1158 could use the <tt>unwind</tt> instruction.</dd>
1160 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1161 <dd>This attribute indicates that the function does not write through any
1162 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1163 arguments) or otherwise modify any state (e.g. memory, control registers,
1164 etc) visible to caller functions. It may dereference pointer arguments
1165 and read state that may be set in the caller. A readonly function always
1166 returns the same value (or unwinds an exception identically) when called
1167 with the same set of arguments and global state. It cannot unwind an
1168 exception by calling the <tt>C++</tt> exception throwing methods, but may
1169 use the <tt>unwind</tt> instruction.</dd>
1171 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1172 <dd>This attribute indicates that the function should emit a stack smashing
1173 protector. It is in the form of a "canary"—a random value placed on
1174 the stack before the local variables that's checked upon return from the
1175 function to see if it has been overwritten. A heuristic is used to
1176 determine if a function needs stack protectors or not.<br>
1178 If a function that has an <tt>ssp</tt> attribute is inlined into a
1179 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1180 function will have an <tt>ssp</tt> attribute.</dd>
1182 <dt><tt><b>sspreq</b></tt></dt>
1183 <dd>This attribute indicates that the function should <em>always</em> emit a
1184 stack smashing protector. This overrides
1185 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1187 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1188 function that doesn't have an <tt>sspreq</tt> attribute or which has
1189 an <tt>ssp</tt> attribute, then the resulting function will have
1190 an <tt>sspreq</tt> attribute.</dd>
1192 <dt><tt><b>noredzone</b></tt></dt>
1193 <dd>This attribute indicates that the code generator should not use a red
1194 zone, even if the target-specific ABI normally permits it.</dd>
1196 <dt><tt><b>noimplicitfloat</b></tt></dt>
1197 <dd>This attributes disables implicit floating point instructions.</dd>
1199 <dt><tt><b>naked</b></tt></dt>
1200 <dd>This attribute disables prologue / epilogue emission for the function.
1201 This can have very system-specific consequences.</dd>
1206 <!-- ======================================================================= -->
1207 <div class="doc_subsection">
1208 <a name="moduleasm">Module-Level Inline Assembly</a>
1211 <div class="doc_text">
1213 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1214 the GCC "file scope inline asm" blocks. These blocks are internally
1215 concatenated by LLVM and treated as a single unit, but may be separated in
1216 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1218 <div class="doc_code">
1220 module asm "inline asm code goes here"
1221 module asm "more can go here"
1225 <p>The strings can contain any character by escaping non-printable characters.
1226 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1229 <p>The inline asm code is simply printed to the machine code .s file when
1230 assembly code is generated.</p>
1234 <!-- ======================================================================= -->
1235 <div class="doc_subsection">
1236 <a name="datalayout">Data Layout</a>
1239 <div class="doc_text">
1241 <p>A module may specify a target specific data layout string that specifies how
1242 data is to be laid out in memory. The syntax for the data layout is
1245 <div class="doc_code">
1247 target datalayout = "<i>layout specification</i>"
1251 <p>The <i>layout specification</i> consists of a list of specifications
1252 separated by the minus sign character ('-'). Each specification starts with
1253 a letter and may include other information after the letter to define some
1254 aspect of the data layout. The specifications accepted are as follows:</p>
1258 <dd>Specifies that the target lays out data in big-endian form. That is, the
1259 bits with the most significance have the lowest address location.</dd>
1262 <dd>Specifies that the target lays out data in little-endian form. That is,
1263 the bits with the least significance have the lowest address
1266 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1267 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1268 <i>preferred</i> alignments. All sizes are in bits. Specifying
1269 the <i>pref</i> alignment is optional. If omitted, the
1270 preceding <tt>:</tt> should be omitted too.</dd>
1272 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1273 <dd>This specifies the alignment for an integer type of a given bit
1274 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1276 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1277 <dd>This specifies the alignment for a vector type of a given bit
1280 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1281 <dd>This specifies the alignment for a floating point type of a given bit
1282 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1285 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1286 <dd>This specifies the alignment for an aggregate type of a given bit
1289 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1290 <dd>This specifies the alignment for a stack object of a given bit
1293 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1294 <dd>This specifies a set of native integer widths for the target CPU
1295 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1296 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1297 this set are considered to support most general arithmetic
1298 operations efficiently.</dd>
1301 <p>When constructing the data layout for a given target, LLVM starts with a
1302 default set of specifications which are then (possibly) overriden by the
1303 specifications in the <tt>datalayout</tt> keyword. The default specifications
1304 are given in this list:</p>
1307 <li><tt>E</tt> - big endian</li>
1308 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1309 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1310 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1311 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1312 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1313 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1314 alignment of 64-bits</li>
1315 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1316 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1317 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1318 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1319 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1320 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1323 <p>When LLVM is determining the alignment for a given type, it uses the
1324 following rules:</p>
1327 <li>If the type sought is an exact match for one of the specifications, that
1328 specification is used.</li>
1330 <li>If no match is found, and the type sought is an integer type, then the
1331 smallest integer type that is larger than the bitwidth of the sought type
1332 is used. If none of the specifications are larger than the bitwidth then
1333 the the largest integer type is used. For example, given the default
1334 specifications above, the i7 type will use the alignment of i8 (next
1335 largest) while both i65 and i256 will use the alignment of i64 (largest
1338 <li>If no match is found, and the type sought is a vector type, then the
1339 largest vector type that is smaller than the sought vector type will be
1340 used as a fall back. This happens because <128 x double> can be
1341 implemented in terms of 64 <2 x double>, for example.</li>
1346 <!-- ======================================================================= -->
1347 <div class="doc_subsection">
1348 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1351 <div class="doc_text">
1353 <p>Any memory access must be done through a pointer value associated
1354 with an address range of the memory access, otherwise the behavior
1355 is undefined. Pointer values are associated with address ranges
1356 according to the following rules:</p>
1359 <li>A pointer value formed from a
1360 <tt><a href="#i_getelementptr">getelementptr</a></tt> instruction
1361 is associated with the addresses associated with the first operand
1362 of the <tt>getelementptr</tt>.</li>
1363 <li>An address of a global variable is associated with the address
1364 range of the variable's storage.</li>
1365 <li>The result value of an allocation instruction is associated with
1366 the address range of the allocated storage.</li>
1367 <li>A null pointer in the default address-space is associated with
1369 <li>A pointer value formed by an
1370 <tt><a href="#i_inttoptr">inttoptr</a></tt> is associated with all
1371 address ranges of all pointer values that contribute (directly or
1372 indirectly) to the computation of the pointer's value.</li>
1373 <li>The result value of a
1374 <tt><a href="#i_bitcast">bitcast</a></tt> is associated with all
1375 addresses associated with the operand of the <tt>bitcast</tt>.</li>
1376 <li>An integer constant other than zero or a pointer value returned
1377 from a function not defined within LLVM may be associated with address
1378 ranges allocated through mechanisms other than those provided by
1379 LLVM. Such ranges shall not overlap with any ranges of addresses
1380 allocated by mechanisms provided by LLVM.</li>
1383 <p>LLVM IR does not associate types with memory. The result type of a
1384 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1385 alignment of the memory from which to load, as well as the
1386 interpretation of the value. The first operand of a
1387 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1388 and alignment of the store.</p>
1390 <p>Consequently, type-based alias analysis, aka TBAA, aka
1391 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1392 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1393 additional information which specialized optimization passes may use
1394 to implement type-based alias analysis.</p>
1398 <!-- ======================================================================= -->
1399 <div class="doc_subsection">
1400 <a name="volatile">Volatile Memory Accesses</a>
1403 <div class="doc_text">
1405 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1406 href="#i_store"><tt>store</tt></a>s, and <a
1407 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1408 The optimizers must not change the number of volatile operations or change their
1409 order of execution relative to other volatile operations. The optimizers
1410 <i>may</i> change the order of volatile operations relative to non-volatile
1411 operations. This is not Java's "volatile" and has no cross-thread
1412 synchronization behavior.</p>
1416 <!-- *********************************************************************** -->
1417 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1418 <!-- *********************************************************************** -->
1420 <div class="doc_text">
1422 <p>The LLVM type system is one of the most important features of the
1423 intermediate representation. Being typed enables a number of optimizations
1424 to be performed on the intermediate representation directly, without having
1425 to do extra analyses on the side before the transformation. A strong type
1426 system makes it easier to read the generated code and enables novel analyses
1427 and transformations that are not feasible to perform on normal three address
1428 code representations.</p>
1432 <!-- ======================================================================= -->
1433 <div class="doc_subsection"> <a name="t_classifications">Type
1434 Classifications</a> </div>
1436 <div class="doc_text">
1438 <p>The types fall into a few useful classifications:</p>
1440 <table border="1" cellspacing="0" cellpadding="4">
1442 <tr><th>Classification</th><th>Types</th></tr>
1444 <td><a href="#t_integer">integer</a></td>
1445 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1448 <td><a href="#t_floating">floating point</a></td>
1449 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1452 <td><a name="t_firstclass">first class</a></td>
1453 <td><a href="#t_integer">integer</a>,
1454 <a href="#t_floating">floating point</a>,
1455 <a href="#t_pointer">pointer</a>,
1456 <a href="#t_vector">vector</a>,
1457 <a href="#t_struct">structure</a>,
1458 <a href="#t_union">union</a>,
1459 <a href="#t_array">array</a>,
1460 <a href="#t_label">label</a>,
1461 <a href="#t_metadata">metadata</a>.
1465 <td><a href="#t_primitive">primitive</a></td>
1466 <td><a href="#t_label">label</a>,
1467 <a href="#t_void">void</a>,
1468 <a href="#t_floating">floating point</a>,
1469 <a href="#t_metadata">metadata</a>.</td>
1472 <td><a href="#t_derived">derived</a></td>
1473 <td><a href="#t_array">array</a>,
1474 <a href="#t_function">function</a>,
1475 <a href="#t_pointer">pointer</a>,
1476 <a href="#t_struct">structure</a>,
1477 <a href="#t_pstruct">packed structure</a>,
1478 <a href="#t_union">union</a>,
1479 <a href="#t_vector">vector</a>,
1480 <a href="#t_opaque">opaque</a>.
1486 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1487 important. Values of these types are the only ones which can be produced by
1492 <!-- ======================================================================= -->
1493 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1495 <div class="doc_text">
1497 <p>The primitive types are the fundamental building blocks of the LLVM
1502 <!-- _______________________________________________________________________ -->
1503 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1505 <div class="doc_text">
1508 <p>The integer type is a very simple type that simply specifies an arbitrary
1509 bit width for the integer type desired. Any bit width from 1 bit to
1510 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1517 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1521 <table class="layout">
1523 <td class="left"><tt>i1</tt></td>
1524 <td class="left">a single-bit integer.</td>
1527 <td class="left"><tt>i32</tt></td>
1528 <td class="left">a 32-bit integer.</td>
1531 <td class="left"><tt>i1942652</tt></td>
1532 <td class="left">a really big integer of over 1 million bits.</td>
1538 <!-- _______________________________________________________________________ -->
1539 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1541 <div class="doc_text">
1545 <tr><th>Type</th><th>Description</th></tr>
1546 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1547 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1548 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1549 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1550 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1556 <!-- _______________________________________________________________________ -->
1557 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1559 <div class="doc_text">
1562 <p>The void type does not represent any value and has no size.</p>
1571 <!-- _______________________________________________________________________ -->
1572 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1574 <div class="doc_text">
1577 <p>The label type represents code labels.</p>
1586 <!-- _______________________________________________________________________ -->
1587 <div class="doc_subsubsection"> <a name="t_metadata">Metadata Type</a> </div>
1589 <div class="doc_text">
1592 <p>The metadata type represents embedded metadata. No derived types may be
1593 created from metadata except for <a href="#t_function">function</a>
1604 <!-- ======================================================================= -->
1605 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1607 <div class="doc_text">
1609 <p>The real power in LLVM comes from the derived types in the system. This is
1610 what allows a programmer to represent arrays, functions, pointers, and other
1611 useful types. Each of these types contain one or more element types which
1612 may be a primitive type, or another derived type. For example, it is
1613 possible to have a two dimensional array, using an array as the element type
1614 of another array.</p>
1619 <!-- _______________________________________________________________________ -->
1620 <div class="doc_subsubsection"> <a name="t_aggregate">Aggregate Types</a> </div>
1622 <div class="doc_text">
1624 <p>Aggregate Types are a subset of derived types that can contain multiple
1625 member types. <a href="#t_array">Arrays</a>,
1626 <a href="#t_struct">structs</a>, <a href="#t_vector">vectors</a> and
1627 <a href="#t_union">unions</a> are aggregate types.</p>
1633 <!-- _______________________________________________________________________ -->
1634 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1636 <div class="doc_text">
1639 <p>The array type is a very simple derived type that arranges elements
1640 sequentially in memory. The array type requires a size (number of elements)
1641 and an underlying data type.</p>
1645 [<# elements> x <elementtype>]
1648 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1649 be any type with a size.</p>
1652 <table class="layout">
1654 <td class="left"><tt>[40 x i32]</tt></td>
1655 <td class="left">Array of 40 32-bit integer values.</td>
1658 <td class="left"><tt>[41 x i32]</tt></td>
1659 <td class="left">Array of 41 32-bit integer values.</td>
1662 <td class="left"><tt>[4 x i8]</tt></td>
1663 <td class="left">Array of 4 8-bit integer values.</td>
1666 <p>Here are some examples of multidimensional arrays:</p>
1667 <table class="layout">
1669 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1670 <td class="left">3x4 array of 32-bit integer values.</td>
1673 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1674 <td class="left">12x10 array of single precision floating point values.</td>
1677 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1678 <td class="left">2x3x4 array of 16-bit integer values.</td>
1682 <p>There is no restriction on indexing beyond the end of the array implied by
1683 a static type (though there are restrictions on indexing beyond the bounds
1684 of an allocated object in some cases). This means that single-dimension
1685 'variable sized array' addressing can be implemented in LLVM with a zero
1686 length array type. An implementation of 'pascal style arrays' in LLVM could
1687 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1691 <!-- _______________________________________________________________________ -->
1692 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1694 <div class="doc_text">
1697 <p>The function type can be thought of as a function signature. It consists of
1698 a return type and a list of formal parameter types. The return type of a
1699 function type is a scalar type, a void type, a struct type, or a union
1700 type. If the return type is a struct type then all struct elements must be
1701 of first class types, and the struct must have at least one element.</p>
1705 <returntype> (<parameter list>)
1708 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1709 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1710 which indicates that the function takes a variable number of arguments.
1711 Variable argument functions can access their arguments with
1712 the <a href="#int_varargs">variable argument handling intrinsic</a>
1713 functions. '<tt><returntype></tt>' is any type except
1714 <a href="#t_label">label</a>.</p>
1717 <table class="layout">
1719 <td class="left"><tt>i32 (i32)</tt></td>
1720 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1722 </tr><tr class="layout">
1723 <td class="left"><tt>float (i16, i32 *) *
1725 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1726 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
1727 returning <tt>float</tt>.
1729 </tr><tr class="layout">
1730 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1731 <td class="left">A vararg function that takes at least one
1732 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1733 which returns an integer. This is the signature for <tt>printf</tt> in
1736 </tr><tr class="layout">
1737 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1738 <td class="left">A function taking an <tt>i32</tt>, returning a
1739 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
1746 <!-- _______________________________________________________________________ -->
1747 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1749 <div class="doc_text">
1752 <p>The structure type is used to represent a collection of data members together
1753 in memory. The packing of the field types is defined to match the ABI of the
1754 underlying processor. The elements of a structure may be any type that has a
1757 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
1758 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
1759 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
1760 Structures in registers are accessed using the
1761 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
1762 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
1765 { <type list> }
1769 <table class="layout">
1771 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1772 <td class="left">A triple of three <tt>i32</tt> values</td>
1773 </tr><tr class="layout">
1774 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1775 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1776 second element is a <a href="#t_pointer">pointer</a> to a
1777 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1778 an <tt>i32</tt>.</td>
1784 <!-- _______________________________________________________________________ -->
1785 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1788 <div class="doc_text">
1791 <p>The packed structure type is used to represent a collection of data members
1792 together in memory. There is no padding between fields. Further, the
1793 alignment of a packed structure is 1 byte. The elements of a packed
1794 structure may be any type that has a size.</p>
1796 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1797 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1798 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1802 < { <type list> } >
1806 <table class="layout">
1808 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1809 <td class="left">A triple of three <tt>i32</tt> values</td>
1810 </tr><tr class="layout">
1812 <tt>< { float, i32 (i32)* } ></tt></td>
1813 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1814 second element is a <a href="#t_pointer">pointer</a> to a
1815 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1816 an <tt>i32</tt>.</td>
1822 <!-- _______________________________________________________________________ -->
1823 <div class="doc_subsubsection"> <a name="t_union">Union Type</a> </div>
1825 <div class="doc_text">
1828 <p>A union type describes an object with size and alignment suitable for
1829 an object of any one of a given set of types (also known as an "untagged"
1830 union). It is similar in concept and usage to a
1831 <a href="#t_struct">struct</a>, except that all members of the union
1832 have an offset of zero. The elements of a union may be any type that has a
1833 size. Unions must have at least one member - empty unions are not allowed.
1836 <p>The size of the union as a whole will be the size of its largest member,
1837 and the alignment requirements of the union as a whole will be the largest
1838 alignment requirement of any member.</p>
1840 <p>Union members are accessed using '<tt><a href="#i_load">load</a></tt> and
1841 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1842 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
1843 Since all members are at offset zero, the getelementptr instruction does
1844 not affect the address, only the type of the resulting pointer.</p>
1848 union { <type list> }
1852 <table class="layout">
1854 <td class="left"><tt>union { i32, i32*, float }</tt></td>
1855 <td class="left">A union of three types: an <tt>i32</tt>, a pointer to
1856 an <tt>i32</tt>, and a <tt>float</tt>.</td>
1857 </tr><tr class="layout">
1859 <tt>union { float, i32 (i32) * }</tt></td>
1860 <td class="left">A union, where the first element is a <tt>float</tt> and the
1861 second element is a <a href="#t_pointer">pointer</a> to a
1862 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1863 an <tt>i32</tt>.</td>
1869 <!-- _______________________________________________________________________ -->
1870 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1872 <div class="doc_text">
1875 <p>The pointer type is used to specify memory locations.
1876 Pointers are commonly used to reference objects in memory.</p>
1878 <p>Pointer types may have an optional address space attribute defining the
1879 numbered address space where the pointed-to object resides. The default
1880 address space is number zero. The semantics of non-zero address
1881 spaces are target-specific.</p>
1883 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
1884 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1892 <table class="layout">
1894 <td class="left"><tt>[4 x i32]*</tt></td>
1895 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1896 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1899 <td class="left"><tt>i32 (i32 *) *</tt></td>
1900 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1901 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1905 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1906 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1907 that resides in address space #5.</td>
1913 <!-- _______________________________________________________________________ -->
1914 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1916 <div class="doc_text">
1919 <p>A vector type is a simple derived type that represents a vector of elements.
1920 Vector types are used when multiple primitive data are operated in parallel
1921 using a single instruction (SIMD). A vector type requires a size (number of
1922 elements) and an underlying primitive data type. Vector types are considered
1923 <a href="#t_firstclass">first class</a>.</p>
1927 < <# elements> x <elementtype> >
1930 <p>The number of elements is a constant integer value; elementtype may be any
1931 integer or floating point type.</p>
1934 <table class="layout">
1936 <td class="left"><tt><4 x i32></tt></td>
1937 <td class="left">Vector of 4 32-bit integer values.</td>
1940 <td class="left"><tt><8 x float></tt></td>
1941 <td class="left">Vector of 8 32-bit floating-point values.</td>
1944 <td class="left"><tt><2 x i64></tt></td>
1945 <td class="left">Vector of 2 64-bit integer values.</td>
1951 <!-- _______________________________________________________________________ -->
1952 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1953 <div class="doc_text">
1956 <p>Opaque types are used to represent unknown types in the system. This
1957 corresponds (for example) to the C notion of a forward declared structure
1958 type. In LLVM, opaque types can eventually be resolved to any type (not just
1959 a structure type).</p>
1967 <table class="layout">
1969 <td class="left"><tt>opaque</tt></td>
1970 <td class="left">An opaque type.</td>
1976 <!-- ======================================================================= -->
1977 <div class="doc_subsection">
1978 <a name="t_uprefs">Type Up-references</a>
1981 <div class="doc_text">
1984 <p>An "up reference" allows you to refer to a lexically enclosing type without
1985 requiring it to have a name. For instance, a structure declaration may
1986 contain a pointer to any of the types it is lexically a member of. Example
1987 of up references (with their equivalent as named type declarations)
1991 { \2 * } %x = type { %x* }
1992 { \2 }* %y = type { %y }*
1996 <p>An up reference is needed by the asmprinter for printing out cyclic types
1997 when there is no declared name for a type in the cycle. Because the
1998 asmprinter does not want to print out an infinite type string, it needs a
1999 syntax to handle recursive types that have no names (all names are optional
2007 <p>The level is the count of the lexical type that is being referred to.</p>
2010 <table class="layout">
2012 <td class="left"><tt>\1*</tt></td>
2013 <td class="left">Self-referential pointer.</td>
2016 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
2017 <td class="left">Recursive structure where the upref refers to the out-most
2024 <!-- *********************************************************************** -->
2025 <div class="doc_section"> <a name="constants">Constants</a> </div>
2026 <!-- *********************************************************************** -->
2028 <div class="doc_text">
2030 <p>LLVM has several different basic types of constants. This section describes
2031 them all and their syntax.</p>
2035 <!-- ======================================================================= -->
2036 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
2038 <div class="doc_text">
2041 <dt><b>Boolean constants</b></dt>
2042 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2043 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2045 <dt><b>Integer constants</b></dt>
2046 <dd>Standard integers (such as '4') are constants of
2047 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2048 with integer types.</dd>
2050 <dt><b>Floating point constants</b></dt>
2051 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2052 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2053 notation (see below). The assembler requires the exact decimal value of a
2054 floating-point constant. For example, the assembler accepts 1.25 but
2055 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2056 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2058 <dt><b>Null pointer constants</b></dt>
2059 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2060 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2063 <p>The one non-intuitive notation for constants is the hexadecimal form of
2064 floating point constants. For example, the form '<tt>double
2065 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2066 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2067 constants are required (and the only time that they are generated by the
2068 disassembler) is when a floating point constant must be emitted but it cannot
2069 be represented as a decimal floating point number in a reasonable number of
2070 digits. For example, NaN's, infinities, and other special values are
2071 represented in their IEEE hexadecimal format so that assembly and disassembly
2072 do not cause any bits to change in the constants.</p>
2074 <p>When using the hexadecimal form, constants of types float and double are
2075 represented using the 16-digit form shown above (which matches the IEEE754
2076 representation for double); float values must, however, be exactly
2077 representable as IEE754 single precision. Hexadecimal format is always used
2078 for long double, and there are three forms of long double. The 80-bit format
2079 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2080 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2081 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2082 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2083 currently supported target uses this format. Long doubles will only work if
2084 they match the long double format on your target. All hexadecimal formats
2085 are big-endian (sign bit at the left).</p>
2089 <!-- ======================================================================= -->
2090 <div class="doc_subsection">
2091 <a name="aggregateconstants"></a> <!-- old anchor -->
2092 <a name="complexconstants">Complex Constants</a>
2095 <div class="doc_text">
2097 <p>Complex constants are a (potentially recursive) combination of simple
2098 constants and smaller complex constants.</p>
2101 <dt><b>Structure constants</b></dt>
2102 <dd>Structure constants are represented with notation similar to structure
2103 type definitions (a comma separated list of elements, surrounded by braces
2104 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2105 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2106 Structure constants must have <a href="#t_struct">structure type</a>, and
2107 the number and types of elements must match those specified by the
2110 <dt><b>Union constants</b></dt>
2111 <dd>Union constants are represented with notation similar to a structure with
2112 a single element - that is, a single typed element surrounded
2113 by braces (<tt>{}</tt>)). For example: "<tt>{ i32 4 }</tt>". The
2114 <a href="#t_union">union type</a> can be initialized with a single-element
2115 struct as long as the type of the struct element matches the type of
2116 one of the union members.</dd>
2118 <dt><b>Array constants</b></dt>
2119 <dd>Array constants are represented with notation similar to array type
2120 definitions (a comma separated list of elements, surrounded by square
2121 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2122 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2123 the number and types of elements must match those specified by the
2126 <dt><b>Vector constants</b></dt>
2127 <dd>Vector constants are represented with notation similar to vector type
2128 definitions (a comma separated list of elements, surrounded by
2129 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2130 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2131 have <a href="#t_vector">vector type</a>, and the number and types of
2132 elements must match those specified by the type.</dd>
2134 <dt><b>Zero initialization</b></dt>
2135 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2136 value to zero of <em>any</em> type, including scalar and
2137 <a href="#t_aggregate">aggregate</a> types.
2138 This is often used to avoid having to print large zero initializers
2139 (e.g. for large arrays) and is always exactly equivalent to using explicit
2140 zero initializers.</dd>
2142 <dt><b>Metadata node</b></dt>
2143 <dd>A metadata node is a structure-like constant with
2144 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2145 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2146 be interpreted as part of the instruction stream, metadata is a place to
2147 attach additional information such as debug info.</dd>
2152 <!-- ======================================================================= -->
2153 <div class="doc_subsection">
2154 <a name="globalconstants">Global Variable and Function Addresses</a>
2157 <div class="doc_text">
2159 <p>The addresses of <a href="#globalvars">global variables</a>
2160 and <a href="#functionstructure">functions</a> are always implicitly valid
2161 (link-time) constants. These constants are explicitly referenced when
2162 the <a href="#identifiers">identifier for the global</a> is used and always
2163 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2164 legal LLVM file:</p>
2166 <div class="doc_code">
2170 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2176 <!-- ======================================================================= -->
2177 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
2178 <div class="doc_text">
2180 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2181 indicates that the user of the value may receive an unspecified bit-pattern.
2182 Undefined values may be of any type (other than label or void) and be used
2183 anywhere a constant is permitted.</p>
2185 <p>Undefined values are useful because they indicate to the compiler that the
2186 program is well defined no matter what value is used. This gives the
2187 compiler more freedom to optimize. Here are some examples of (potentially
2188 surprising) transformations that are valid (in pseudo IR):</p>
2191 <div class="doc_code">
2203 <p>This is safe because all of the output bits are affected by the undef bits.
2204 Any output bit can have a zero or one depending on the input bits.</p>
2206 <div class="doc_code">
2219 <p>These logical operations have bits that are not always affected by the input.
2220 For example, if "%X" has a zero bit, then the output of the 'and' operation will
2221 always be a zero, no matter what the corresponding bit from the undef is. As
2222 such, it is unsafe to optimize or assume that the result of the and is undef.
2223 However, it is safe to assume that all bits of the undef could be 0, and
2224 optimize the and to 0. Likewise, it is safe to assume that all the bits of
2225 the undef operand to the or could be set, allowing the or to be folded to
2228 <div class="doc_code">
2230 %A = select undef, %X, %Y
2231 %B = select undef, 42, %Y
2232 %C = select %X, %Y, undef
2244 <p>This set of examples show that undefined select (and conditional branch)
2245 conditions can go "either way" but they have to come from one of the two
2246 operands. In the %A example, if %X and %Y were both known to have a clear low
2247 bit, then %A would have to have a cleared low bit. However, in the %C example,
2248 the optimizer is allowed to assume that the undef operand could be the same as
2249 %Y, allowing the whole select to be eliminated.</p>
2252 <div class="doc_code">
2254 %A = xor undef, undef
2273 <p>This example points out that two undef operands are not necessarily the same.
2274 This can be surprising to people (and also matches C semantics) where they
2275 assume that "X^X" is always zero, even if X is undef. This isn't true for a
2276 number of reasons, but the short answer is that an undef "variable" can
2277 arbitrarily change its value over its "live range". This is true because the
2278 "variable" doesn't actually <em>have a live range</em>. Instead, the value is
2279 logically read from arbitrary registers that happen to be around when needed,
2280 so the value is not necessarily consistent over time. In fact, %A and %C need
2281 to have the same semantics or the core LLVM "replace all uses with" concept
2284 <div class="doc_code">
2294 <p>These examples show the crucial difference between an <em>undefined
2295 value</em> and <em>undefined behavior</em>. An undefined value (like undef) is
2296 allowed to have an arbitrary bit-pattern. This means that the %A operation
2297 can be constant folded to undef because the undef could be an SNaN, and fdiv is
2298 not (currently) defined on SNaN's. However, in the second example, we can make
2299 a more aggressive assumption: because the undef is allowed to be an arbitrary
2300 value, we are allowed to assume that it could be zero. Since a divide by zero
2301 has <em>undefined behavior</em>, we are allowed to assume that the operation
2302 does not execute at all. This allows us to delete the divide and all code after
2303 it: since the undefined operation "can't happen", the optimizer can assume that
2304 it occurs in dead code.
2307 <div class="doc_code">
2309 a: store undef -> %X
2310 b: store %X -> undef
2317 <p>These examples reiterate the fdiv example: a store "of" an undefined value
2318 can be assumed to not have any effect: we can assume that the value is
2319 overwritten with bits that happen to match what was already there. However, a
2320 store "to" an undefined location could clobber arbitrary memory, therefore, it
2321 has undefined behavior.</p>
2325 <!-- ======================================================================= -->
2326 <div class="doc_subsection"><a name="trapvalues">Trap Values</a></div>
2327 <div class="doc_text">
2329 <p>Trap values are similar to <a href="#undefvalues">undef values</a>, however
2330 instead of representing an unspecified bit pattern, they represent the
2331 fact that an instruction or constant expression which cannot evoke side
2332 effects has nevertheless detected a condition which results in undefined
2335 <p>Any value other than a non-intrinsic call, invoke, or phi with a trap
2336 operand has trap as its result value. Any instruction with
2337 a trap operand which may have side effects emits those side effects as
2338 if it had an undef operand instead. If the side effects are externally
2339 visible, the behavior is undefined.</p>
2341 <p>Trap values may be stored to memory; a load from memory including any
2342 part of a trap value results in a (full) trap value.</p>
2346 <!-- FIXME: In the case of multiple threads, this only applies to loads from
2347 the same thread as the store, or loads which are sequenced after the
2348 store by synchronization. -->
2350 <div class="doc_code">
2352 %trap = sub nuw i32 0, 1 ; Results in a trap value.
2353 %still_trap = and i32 %trap, 0 ; Whereas (and i32 undef, 0) would return 0.
2354 %trap_yet_again = getelementptr i32* @h, i32 %still_trap
2355 store i32 0, i32* %trap_yet_again ; undefined behavior
2357 volatile store i32 %trap, i32* @g ; External observation; undefined behavior.
2358 %trap2 = load i32* @g ; Returns a trap value, not just undef.
2359 %narrowaddr = bitcast i32* @g to i16*
2360 %wideaddr = bitcast i32* @g to i64*
2361 %trap3 = load 16* %narrowaddr ; Returns a trap value
2362 %trap4 = load i64* %widaddr ; Returns a trap value, not partial trap.
2366 <p>If a <a href="#i_br"><tt>br</tt></a> or
2367 <a href="#i_switch"><tt>switch</tt></a> instruction has a trap value
2368 operand, all non-phi non-void instructions which control-depend on it
2369 have trap as their result value. A <a href="#i_phi"><tt>phi</tt></a>
2370 node with an incoming value associated with a control edge which is
2371 control-dependent on it has trap as its result value when control is
2372 transferred from that block. If any instruction which control-depends
2373 on the <tt>br</tt> or <tt>switch</tt> invokes externally visible side
2374 effects, the behavior of the program is undefined. For example:</p>
2376 <!-- FIXME: What about exceptions thrown from control-dependent instrs? -->
2378 <div class="doc_code">
2381 %trap = sub nuw i32 0, 1 ; Results in a trap value.
2382 %cmp = icmp i32 slt %trap, 0 ; Still trap.
2383 %br i1 %cmp, %true, %end ; Branch to either destination.
2386 volatile store i32 0, i32* @g ; Externally visible side effects
2387 ; control-dependent on %cmp.
2388 ; Undefined behavior.
2392 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2393 ; Both edges into this PHI are
2394 ; control-dependent on %cmp, so this
2395 ; results in a trap value.
2397 volatile store i32 0, i32* @g ; %end is control-equivalent to %entry
2398 ; so this is defined (ignoring earlier
2399 ; undefined behavior in this example).
2404 <p>There is currently no way of representing a trap constant in the IR; they
2405 only exist when produced by certain instructions, such as an
2406 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag
2407 set, when overflow occurs.</p>
2411 <!-- ======================================================================= -->
2412 <div class="doc_subsection"><a name="blockaddress">Addresses of Basic
2414 <div class="doc_text">
2416 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2418 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2419 basic block in the specified function, and always has an i8* type. Taking
2420 the address of the entry block is illegal.</p>
2422 <p>This value only has defined behavior when used as an operand to the
2423 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction or for comparisons
2424 against null. Pointer equality tests between labels addresses is undefined
2425 behavior - though, again, comparison against null is ok, and no label is
2426 equal to the null pointer. This may also be passed around as an opaque
2427 pointer sized value as long as the bits are not inspected. This allows
2428 <tt>ptrtoint</tt> and arithmetic to be performed on these values so long as
2429 the original value is reconstituted before the <tt>indirectbr</tt>.</p>
2431 <p>Finally, some targets may provide defined semantics when
2432 using the value as the operand to an inline assembly, but that is target
2439 <!-- ======================================================================= -->
2440 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
2443 <div class="doc_text">
2445 <p>Constant expressions are used to allow expressions involving other constants
2446 to be used as constants. Constant expressions may be of
2447 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2448 operation that does not have side effects (e.g. load and call are not
2449 supported). The following is the syntax for constant expressions:</p>
2452 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
2453 <dd>Truncate a constant to another type. The bit size of CST must be larger
2454 than the bit size of TYPE. Both types must be integers.</dd>
2456 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
2457 <dd>Zero extend a constant to another type. The bit size of CST must be
2458 smaller or equal to the bit size of TYPE. Both types must be
2461 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
2462 <dd>Sign extend a constant to another type. The bit size of CST must be
2463 smaller or equal to the bit size of TYPE. Both types must be
2466 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
2467 <dd>Truncate a floating point constant to another floating point type. The
2468 size of CST must be larger than the size of TYPE. Both types must be
2469 floating point.</dd>
2471 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
2472 <dd>Floating point extend a constant to another type. The size of CST must be
2473 smaller or equal to the size of TYPE. Both types must be floating
2476 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
2477 <dd>Convert a floating point constant to the corresponding unsigned integer
2478 constant. TYPE must be a scalar or vector integer type. CST must be of
2479 scalar or vector floating point type. Both CST and TYPE must be scalars,
2480 or vectors of the same number of elements. If the value won't fit in the
2481 integer type, the results are undefined.</dd>
2483 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
2484 <dd>Convert a floating point constant to the corresponding signed integer
2485 constant. TYPE must be a scalar or vector integer type. CST must be of
2486 scalar or vector floating point type. Both CST and TYPE must be scalars,
2487 or vectors of the same number of elements. If the value won't fit in the
2488 integer type, the results are undefined.</dd>
2490 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
2491 <dd>Convert an unsigned integer constant to the corresponding floating point
2492 constant. TYPE must be a scalar or vector floating point type. CST must be
2493 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2494 vectors of the same number of elements. If the value won't fit in the
2495 floating point type, the results are undefined.</dd>
2497 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
2498 <dd>Convert a signed integer constant to the corresponding floating point
2499 constant. TYPE must be a scalar or vector floating point type. CST must be
2500 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2501 vectors of the same number of elements. If the value won't fit in the
2502 floating point type, the results are undefined.</dd>
2504 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
2505 <dd>Convert a pointer typed constant to the corresponding integer constant
2506 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2507 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2508 make it fit in <tt>TYPE</tt>.</dd>
2510 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
2511 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2512 type. CST must be of integer type. The CST value is zero extended,
2513 truncated, or unchanged to make it fit in a pointer size. This one is
2514 <i>really</i> dangerous!</dd>
2516 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
2517 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2518 are the same as those for the <a href="#i_bitcast">bitcast
2519 instruction</a>.</dd>
2521 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2522 <dt><b><tt>getelementptr inbounds ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2523 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2524 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2525 instruction, the index list may have zero or more indexes, which are
2526 required to make sense for the type of "CSTPTR".</dd>
2528 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
2529 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2531 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
2532 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2534 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
2535 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2537 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2538 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2541 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2542 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2545 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2546 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2549 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2550 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2551 be any of the <a href="#binaryops">binary</a>
2552 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2553 on operands are the same as those for the corresponding instruction
2554 (e.g. no bitwise operations on floating point values are allowed).</dd>
2559 <!-- *********************************************************************** -->
2560 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2561 <!-- *********************************************************************** -->
2563 <!-- ======================================================================= -->
2564 <div class="doc_subsection">
2565 <a name="inlineasm">Inline Assembler Expressions</a>
2568 <div class="doc_text">
2570 <p>LLVM supports inline assembler expressions (as opposed
2571 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2572 a special value. This value represents the inline assembler as a string
2573 (containing the instructions to emit), a list of operand constraints (stored
2574 as a string), a flag that indicates whether or not the inline asm
2575 expression has side effects, and a flag indicating whether the function
2576 containing the asm needs to align its stack conservatively. An example
2577 inline assembler expression is:</p>
2579 <div class="doc_code">
2581 i32 (i32) asm "bswap $0", "=r,r"
2585 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2586 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2589 <div class="doc_code">
2591 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2595 <p>Inline asms with side effects not visible in the constraint list must be
2596 marked as having side effects. This is done through the use of the
2597 '<tt>sideeffect</tt>' keyword, like so:</p>
2599 <div class="doc_code">
2601 call void asm sideeffect "eieio", ""()
2605 <p>In some cases inline asms will contain code that will not work unless the
2606 stack is aligned in some way, such as calls or SSE instructions on x86,
2607 yet will not contain code that does that alignment within the asm.
2608 The compiler should make conservative assumptions about what the asm might
2609 contain and should generate its usual stack alignment code in the prologue
2610 if the '<tt>alignstack</tt>' keyword is present:</p>
2612 <div class="doc_code">
2614 call void asm alignstack "eieio", ""()
2618 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2621 <p>TODO: The format of the asm and constraints string still need to be
2622 documented here. Constraints on what can be done (e.g. duplication, moving,
2623 etc need to be documented). This is probably best done by reference to
2624 another document that covers inline asm from a holistic perspective.</p>
2627 <div class="doc_subsubsection">
2628 <a name="inlineasm_md">Inline Asm Metadata</a>
2631 <div class="doc_text">
2633 <p>The call instructions that wrap inline asm nodes may have a "!srcloc" MDNode
2634 attached to it that contains a constant integer. If present, the code
2635 generator will use the integer as the location cookie value when report
2636 errors through the LLVMContext error reporting mechanisms. This allows a
2637 front-end to corrolate backend errors that occur with inline asm back to the
2638 source code that produced it. For example:</p>
2640 <div class="doc_code">
2642 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2644 !42 = !{ i32 1234567 }
2648 <p>It is up to the front-end to make sense of the magic numbers it places in the
2653 <!-- ======================================================================= -->
2654 <div class="doc_subsection"><a name="metadata">Metadata Nodes and Metadata
2658 <div class="doc_text">
2660 <p>LLVM IR allows metadata to be attached to instructions in the program that
2661 can convey extra information about the code to the optimizers and code
2662 generator. One example application of metadata is source-level debug
2663 information. There are two metadata primitives: strings and nodes. All
2664 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2665 preceding exclamation point ('<tt>!</tt>').</p>
2667 <p>A metadata string is a string surrounded by double quotes. It can contain
2668 any character by escaping non-printable characters with "\xx" where "xx" is
2669 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2671 <p>Metadata nodes are represented with notation similar to structure constants
2672 (a comma separated list of elements, surrounded by braces and preceded by an
2673 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2674 10}</tt>". Metadata nodes can have any values as their operand.</p>
2676 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2677 metadata nodes, which can be looked up in the module symbol table. For
2678 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2680 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2681 function is using two metadata arguments.
2683 <div class="doc_code">
2685 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2689 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2690 attached with <tt>add</tt> instruction using <tt>!dbg</tt> identifier.
2692 <div class="doc_code">
2694 %indvar.next = add i64 %indvar, 1, !dbg !21
2700 <!-- *********************************************************************** -->
2701 <div class="doc_section">
2702 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2704 <!-- *********************************************************************** -->
2706 <p>LLVM has a number of "magic" global variables that contain data that affect
2707 code generation or other IR semantics. These are documented here. All globals
2708 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2709 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2712 <!-- ======================================================================= -->
2713 <div class="doc_subsection">
2714 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2717 <div class="doc_text">
2719 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2720 href="#linkage_appending">appending linkage</a>. This array contains a list of
2721 pointers to global variables and functions which may optionally have a pointer
2722 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2728 @llvm.used = appending global [2 x i8*] [
2730 i8* bitcast (i32* @Y to i8*)
2731 ], section "llvm.metadata"
2734 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2735 compiler, assembler, and linker are required to treat the symbol as if there is
2736 a reference to the global that it cannot see. For example, if a variable has
2737 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2738 list, it cannot be deleted. This is commonly used to represent references from
2739 inline asms and other things the compiler cannot "see", and corresponds to
2740 "attribute((used))" in GNU C.</p>
2742 <p>On some targets, the code generator must emit a directive to the assembler or
2743 object file to prevent the assembler and linker from molesting the symbol.</p>
2747 <!-- ======================================================================= -->
2748 <div class="doc_subsection">
2749 <a name="intg_compiler_used">The '<tt>llvm.compiler.used</tt>' Global Variable</a>
2752 <div class="doc_text">
2754 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2755 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2756 touching the symbol. On targets that support it, this allows an intelligent
2757 linker to optimize references to the symbol without being impeded as it would be
2758 by <tt>@llvm.used</tt>.</p>
2760 <p>This is a rare construct that should only be used in rare circumstances, and
2761 should not be exposed to source languages.</p>
2765 <!-- ======================================================================= -->
2766 <div class="doc_subsection">
2767 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2770 <div class="doc_text">
2772 <p>TODO: Describe this.</p>
2776 <!-- ======================================================================= -->
2777 <div class="doc_subsection">
2778 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2781 <div class="doc_text">
2783 <p>TODO: Describe this.</p>
2788 <!-- *********************************************************************** -->
2789 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2790 <!-- *********************************************************************** -->
2792 <div class="doc_text">
2794 <p>The LLVM instruction set consists of several different classifications of
2795 instructions: <a href="#terminators">terminator
2796 instructions</a>, <a href="#binaryops">binary instructions</a>,
2797 <a href="#bitwiseops">bitwise binary instructions</a>,
2798 <a href="#memoryops">memory instructions</a>, and
2799 <a href="#otherops">other instructions</a>.</p>
2803 <!-- ======================================================================= -->
2804 <div class="doc_subsection"> <a name="terminators">Terminator
2805 Instructions</a> </div>
2807 <div class="doc_text">
2809 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
2810 in a program ends with a "Terminator" instruction, which indicates which
2811 block should be executed after the current block is finished. These
2812 terminator instructions typically yield a '<tt>void</tt>' value: they produce
2813 control flow, not values (the one exception being the
2814 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2816 <p>There are seven different terminator instructions: the
2817 '<a href="#i_ret"><tt>ret</tt></a>' instruction, the
2818 '<a href="#i_br"><tt>br</tt></a>' instruction, the
2819 '<a href="#i_switch"><tt>switch</tt></a>' instruction, the
2820 '<a href="#i_indirectbr">'<tt>indirectbr</tt></a>' Instruction, the
2821 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the
2822 '<a href="#i_unwind"><tt>unwind</tt></a>' instruction, and the
2823 '<a href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2827 <!-- _______________________________________________________________________ -->
2828 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2829 Instruction</a> </div>
2831 <div class="doc_text">
2835 ret <type> <value> <i>; Return a value from a non-void function</i>
2836 ret void <i>; Return from void function</i>
2840 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
2841 a value) from a function back to the caller.</p>
2843 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
2844 value and then causes control flow, and one that just causes control flow to
2848 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
2849 return value. The type of the return value must be a
2850 '<a href="#t_firstclass">first class</a>' type.</p>
2852 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
2853 non-void return type and contains a '<tt>ret</tt>' instruction with no return
2854 value or a return value with a type that does not match its type, or if it
2855 has a void return type and contains a '<tt>ret</tt>' instruction with a
2859 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
2860 the calling function's context. If the caller is a
2861 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
2862 instruction after the call. If the caller was an
2863 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
2864 the beginning of the "normal" destination block. If the instruction returns
2865 a value, that value shall set the call or invoke instruction's return
2870 ret i32 5 <i>; Return an integer value of 5</i>
2871 ret void <i>; Return from a void function</i>
2872 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2876 <!-- _______________________________________________________________________ -->
2877 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2879 <div class="doc_text">
2883 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2887 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
2888 different basic block in the current function. There are two forms of this
2889 instruction, corresponding to a conditional branch and an unconditional
2893 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
2894 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
2895 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
2899 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2900 argument is evaluated. If the value is <tt>true</tt>, control flows to the
2901 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2902 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2907 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
2908 br i1 %cond, label %IfEqual, label %IfUnequal
2910 <a href="#i_ret">ret</a> i32 1
2912 <a href="#i_ret">ret</a> i32 0
2917 <!-- _______________________________________________________________________ -->
2918 <div class="doc_subsubsection">
2919 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2922 <div class="doc_text">
2926 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2930 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2931 several different places. It is a generalization of the '<tt>br</tt>'
2932 instruction, allowing a branch to occur to one of many possible
2936 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2937 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
2938 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
2939 The table is not allowed to contain duplicate constant entries.</p>
2942 <p>The <tt>switch</tt> instruction specifies a table of values and
2943 destinations. When the '<tt>switch</tt>' instruction is executed, this table
2944 is searched for the given value. If the value is found, control flow is
2945 transferred to the corresponding destination; otherwise, control flow is
2946 transferred to the default destination.</p>
2948 <h5>Implementation:</h5>
2949 <p>Depending on properties of the target machine and the particular
2950 <tt>switch</tt> instruction, this instruction may be code generated in
2951 different ways. For example, it could be generated as a series of chained
2952 conditional branches or with a lookup table.</p>
2956 <i>; Emulate a conditional br instruction</i>
2957 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2958 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2960 <i>; Emulate an unconditional br instruction</i>
2961 switch i32 0, label %dest [ ]
2963 <i>; Implement a jump table:</i>
2964 switch i32 %val, label %otherwise [ i32 0, label %onzero
2966 i32 2, label %ontwo ]
2972 <!-- _______________________________________________________________________ -->
2973 <div class="doc_subsubsection">
2974 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
2977 <div class="doc_text">
2981 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
2986 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
2987 within the current function, whose address is specified by
2988 "<tt>address</tt>". Address must be derived from a <a
2989 href="#blockaddress">blockaddress</a> constant.</p>
2993 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
2994 rest of the arguments indicate the full set of possible destinations that the
2995 address may point to. Blocks are allowed to occur multiple times in the
2996 destination list, though this isn't particularly useful.</p>
2998 <p>This destination list is required so that dataflow analysis has an accurate
2999 understanding of the CFG.</p>
3003 <p>Control transfers to the block specified in the address argument. All
3004 possible destination blocks must be listed in the label list, otherwise this
3005 instruction has undefined behavior. This implies that jumps to labels
3006 defined in other functions have undefined behavior as well.</p>
3008 <h5>Implementation:</h5>
3010 <p>This is typically implemented with a jump through a register.</p>
3014 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3020 <!-- _______________________________________________________________________ -->
3021 <div class="doc_subsubsection">
3022 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3025 <div class="doc_text">
3029 <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>]
3030 to label <normal label> unwind label <exception label>
3034 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3035 function, with the possibility of control flow transfer to either the
3036 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3037 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3038 control flow will return to the "normal" label. If the callee (or any
3039 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
3040 instruction, control is interrupted and continued at the dynamically nearest
3041 "exception" label.</p>
3044 <p>This instruction requires several arguments:</p>
3047 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3048 convention</a> the call should use. If none is specified, the call
3049 defaults to using C calling conventions.</li>
3051 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3052 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3053 '<tt>inreg</tt>' attributes are valid here.</li>
3055 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3056 function value being invoked. In most cases, this is a direct function
3057 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3058 off an arbitrary pointer to function value.</li>
3060 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3061 function to be invoked. </li>
3063 <li>'<tt>function args</tt>': argument list whose types match the function
3064 signature argument types and parameter attributes. All arguments must be
3065 of <a href="#t_firstclass">first class</a> type. If the function
3066 signature indicates the function accepts a variable number of arguments,
3067 the extra arguments can be specified.</li>
3069 <li>'<tt>normal label</tt>': the label reached when the called function
3070 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3072 <li>'<tt>exception label</tt>': the label reached when a callee returns with
3073 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
3075 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3076 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3077 '<tt>readnone</tt>' attributes are valid here.</li>
3081 <p>This instruction is designed to operate as a standard
3082 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3083 primary difference is that it establishes an association with a label, which
3084 is used by the runtime library to unwind the stack.</p>
3086 <p>This instruction is used in languages with destructors to ensure that proper
3087 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3088 exception. Additionally, this is important for implementation of
3089 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3091 <p>For the purposes of the SSA form, the definition of the value returned by the
3092 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3093 block to the "normal" label. If the callee unwinds then no return value is
3096 <p>Note that the code generator does not yet completely support unwind, and
3097 that the invoke/unwind semantics are likely to change in future versions.</p>
3101 %retval = invoke i32 @Test(i32 15) to label %Continue
3102 unwind label %TestCleanup <i>; {i32}:retval set</i>
3103 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3104 unwind label %TestCleanup <i>; {i32}:retval set</i>
3109 <!-- _______________________________________________________________________ -->
3111 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
3112 Instruction</a> </div>
3114 <div class="doc_text">
3122 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
3123 at the first callee in the dynamic call stack which used
3124 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
3125 This is primarily used to implement exception handling.</p>
3128 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
3129 immediately halt. The dynamic call stack is then searched for the
3130 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
3131 Once found, execution continues at the "exceptional" destination block
3132 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
3133 instruction in the dynamic call chain, undefined behavior results.</p>
3135 <p>Note that the code generator does not yet completely support unwind, and
3136 that the invoke/unwind semantics are likely to change in future versions.</p>
3140 <!-- _______________________________________________________________________ -->
3142 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
3143 Instruction</a> </div>
3145 <div class="doc_text">
3153 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3154 instruction is used to inform the optimizer that a particular portion of the
3155 code is not reachable. This can be used to indicate that the code after a
3156 no-return function cannot be reached, and other facts.</p>
3159 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3163 <!-- ======================================================================= -->
3164 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
3166 <div class="doc_text">
3168 <p>Binary operators are used to do most of the computation in a program. They
3169 require two operands of the same type, execute an operation on them, and
3170 produce a single value. The operands might represent multiple data, as is
3171 the case with the <a href="#t_vector">vector</a> data type. The result value
3172 has the same type as its operands.</p>
3174 <p>There are several different binary operators:</p>
3178 <!-- _______________________________________________________________________ -->
3179 <div class="doc_subsubsection">
3180 <a name="i_add">'<tt>add</tt>' Instruction</a>
3183 <div class="doc_text">
3187 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3188 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3189 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3190 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3194 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3197 <p>The two arguments to the '<tt>add</tt>' instruction must
3198 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3199 integer values. Both arguments must have identical types.</p>
3202 <p>The value produced is the integer sum of the two operands.</p>
3204 <p>If the sum has unsigned overflow, the result returned is the mathematical
3205 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3207 <p>Because LLVM integers use a two's complement representation, this instruction
3208 is appropriate for both signed and unsigned integers.</p>
3210 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3211 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3212 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3213 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3214 respectively, occurs.</p>
3218 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3223 <!-- _______________________________________________________________________ -->
3224 <div class="doc_subsubsection">
3225 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3228 <div class="doc_text">
3232 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3236 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3239 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3240 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3241 floating point values. Both arguments must have identical types.</p>
3244 <p>The value produced is the floating point sum of the two operands.</p>
3248 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3253 <!-- _______________________________________________________________________ -->
3254 <div class="doc_subsubsection">
3255 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3258 <div class="doc_text">
3262 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3263 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3264 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3265 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3269 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3272 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3273 '<tt>neg</tt>' instruction present in most other intermediate
3274 representations.</p>
3277 <p>The two arguments to the '<tt>sub</tt>' instruction must
3278 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3279 integer values. Both arguments must have identical types.</p>
3282 <p>The value produced is the integer difference of the two operands.</p>
3284 <p>If the difference has unsigned overflow, the result returned is the
3285 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3288 <p>Because LLVM integers use a two's complement representation, this instruction
3289 is appropriate for both signed and unsigned integers.</p>
3291 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3292 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3293 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3294 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3295 respectively, occurs.</p>
3299 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3300 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3305 <!-- _______________________________________________________________________ -->
3306 <div class="doc_subsubsection">
3307 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3310 <div class="doc_text">
3314 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3318 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3321 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3322 '<tt>fneg</tt>' instruction present in most other intermediate
3323 representations.</p>
3326 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3327 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3328 floating point values. Both arguments must have identical types.</p>
3331 <p>The value produced is the floating point difference of the two operands.</p>
3335 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3336 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3341 <!-- _______________________________________________________________________ -->
3342 <div class="doc_subsubsection">
3343 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3346 <div class="doc_text">
3350 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3351 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3352 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3353 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3357 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3360 <p>The two arguments to the '<tt>mul</tt>' instruction must
3361 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3362 integer values. Both arguments must have identical types.</p>
3365 <p>The value produced is the integer product of the two operands.</p>
3367 <p>If the result of the multiplication has unsigned overflow, the result
3368 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3369 width of the result.</p>
3371 <p>Because LLVM integers use a two's complement representation, and the result
3372 is the same width as the operands, this instruction returns the correct
3373 result for both signed and unsigned integers. If a full product
3374 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3375 be sign-extended or zero-extended as appropriate to the width of the full
3378 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3379 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3380 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3381 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3382 respectively, occurs.</p>
3386 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3391 <!-- _______________________________________________________________________ -->
3392 <div class="doc_subsubsection">
3393 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3396 <div class="doc_text">
3400 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3404 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3407 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3408 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3409 floating point values. Both arguments must have identical types.</p>
3412 <p>The value produced is the floating point product of the two operands.</p>
3416 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3421 <!-- _______________________________________________________________________ -->
3422 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
3425 <div class="doc_text">
3429 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3433 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3436 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3437 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3438 values. Both arguments must have identical types.</p>
3441 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3443 <p>Note that unsigned integer division and signed integer division are distinct
3444 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3446 <p>Division by zero leads to undefined behavior.</p>
3450 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3455 <!-- _______________________________________________________________________ -->
3456 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
3459 <div class="doc_text">
3463 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3464 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3468 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3471 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3472 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3473 values. Both arguments must have identical types.</p>
3476 <p>The value produced is the signed integer quotient of the two operands rounded
3479 <p>Note that signed integer division and unsigned integer division are distinct
3480 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3482 <p>Division by zero leads to undefined behavior. Overflow also leads to
3483 undefined behavior; this is a rare case, but can occur, for example, by doing
3484 a 32-bit division of -2147483648 by -1.</p>
3486 <p>If the <tt>exact</tt> keyword is present, the result value of the
3487 <tt>sdiv</tt> is a <a href="#trapvalues">trap value</a> if the result would
3488 be rounded or if overflow would occur.</p>
3492 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3497 <!-- _______________________________________________________________________ -->
3498 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
3499 Instruction</a> </div>
3501 <div class="doc_text">
3505 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3509 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3512 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3513 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3514 floating point values. Both arguments must have identical types.</p>
3517 <p>The value produced is the floating point quotient of the two operands.</p>
3521 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3526 <!-- _______________________________________________________________________ -->
3527 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3530 <div class="doc_text">
3534 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3538 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3539 division of its two arguments.</p>
3542 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3543 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3544 values. Both arguments must have identical types.</p>
3547 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3548 This instruction always performs an unsigned division to get the
3551 <p>Note that unsigned integer remainder and signed integer remainder are
3552 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3554 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3558 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3563 <!-- _______________________________________________________________________ -->
3564 <div class="doc_subsubsection">
3565 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3568 <div class="doc_text">
3572 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3576 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3577 division of its two operands. This instruction can also take
3578 <a href="#t_vector">vector</a> versions of the values in which case the
3579 elements must be integers.</p>
3582 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3583 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3584 values. Both arguments must have identical types.</p>
3587 <p>This instruction returns the <i>remainder</i> of a division (where the result
3588 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
3589 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
3590 a value. For more information about the difference,
3591 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3592 Math Forum</a>. For a table of how this is implemented in various languages,
3593 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3594 Wikipedia: modulo operation</a>.</p>
3596 <p>Note that signed integer remainder and unsigned integer remainder are
3597 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3599 <p>Taking the remainder of a division by zero leads to undefined behavior.
3600 Overflow also leads to undefined behavior; this is a rare case, but can
3601 occur, for example, by taking the remainder of a 32-bit division of
3602 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3603 lets srem be implemented using instructions that return both the result of
3604 the division and the remainder.)</p>
3608 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3613 <!-- _______________________________________________________________________ -->
3614 <div class="doc_subsubsection">
3615 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
3617 <div class="doc_text">
3621 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3625 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3626 its two operands.</p>
3629 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3630 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3631 floating point values. Both arguments must have identical types.</p>
3634 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3635 has the same sign as the dividend.</p>
3639 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3644 <!-- ======================================================================= -->
3645 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
3646 Operations</a> </div>
3648 <div class="doc_text">
3650 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3651 program. They are generally very efficient instructions and can commonly be
3652 strength reduced from other instructions. They require two operands of the
3653 same type, execute an operation on them, and produce a single value. The
3654 resulting value is the same type as its operands.</p>
3658 <!-- _______________________________________________________________________ -->
3659 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
3660 Instruction</a> </div>
3662 <div class="doc_text">
3666 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3670 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3671 a specified number of bits.</p>
3674 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3675 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3676 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3679 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3680 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3681 is (statically or dynamically) negative or equal to or larger than the number
3682 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3683 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3684 shift amount in <tt>op2</tt>.</p>
3688 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3689 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3690 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3691 <result> = shl i32 1, 32 <i>; undefined</i>
3692 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3697 <!-- _______________________________________________________________________ -->
3698 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
3699 Instruction</a> </div>
3701 <div class="doc_text">
3705 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3709 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
3710 operand shifted to the right a specified number of bits with zero fill.</p>
3713 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
3714 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3715 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3718 <p>This instruction always performs a logical shift right operation. The most
3719 significant bits of the result will be filled with zero bits after the shift.
3720 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
3721 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3722 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3723 shift amount in <tt>op2</tt>.</p>
3727 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
3728 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
3729 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
3730 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
3731 <result> = lshr i32 1, 32 <i>; undefined</i>
3732 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3737 <!-- _______________________________________________________________________ -->
3738 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
3739 Instruction</a> </div>
3740 <div class="doc_text">
3744 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3748 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
3749 operand shifted to the right a specified number of bits with sign
3753 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
3754 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3755 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3758 <p>This instruction always performs an arithmetic shift right operation, The
3759 most significant bits of the result will be filled with the sign bit
3760 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
3761 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
3762 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
3763 the corresponding shift amount in <tt>op2</tt>.</p>
3767 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
3768 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
3769 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
3770 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
3771 <result> = ashr i32 1, 32 <i>; undefined</i>
3772 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3777 <!-- _______________________________________________________________________ -->
3778 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
3779 Instruction</a> </div>
3781 <div class="doc_text">
3785 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3789 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
3793 <p>The two arguments to the '<tt>and</tt>' instruction must be
3794 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3795 values. Both arguments must have identical types.</p>
3798 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
3800 <table border="1" cellspacing="0" cellpadding="4">
3832 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
3833 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
3834 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
3837 <!-- _______________________________________________________________________ -->
3838 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
3840 <div class="doc_text">
3844 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3848 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
3852 <p>The two arguments to the '<tt>or</tt>' instruction must be
3853 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3854 values. Both arguments must have identical types.</p>
3857 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
3859 <table border="1" cellspacing="0" cellpadding="4">
3891 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3892 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3893 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3898 <!-- _______________________________________________________________________ -->
3899 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3900 Instruction</a> </div>
3902 <div class="doc_text">
3906 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3910 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
3911 its two operands. The <tt>xor</tt> is used to implement the "one's
3912 complement" operation, which is the "~" operator in C.</p>
3915 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3916 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3917 values. Both arguments must have identical types.</p>
3920 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3922 <table border="1" cellspacing="0" cellpadding="4">
3954 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3955 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3956 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3957 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3962 <!-- ======================================================================= -->
3963 <div class="doc_subsection">
3964 <a name="vectorops">Vector Operations</a>
3967 <div class="doc_text">
3969 <p>LLVM supports several instructions to represent vector operations in a
3970 target-independent manner. These instructions cover the element-access and
3971 vector-specific operations needed to process vectors effectively. While LLVM
3972 does directly support these vector operations, many sophisticated algorithms
3973 will want to use target-specific intrinsics to take full advantage of a
3974 specific target.</p>
3978 <!-- _______________________________________________________________________ -->
3979 <div class="doc_subsubsection">
3980 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3983 <div class="doc_text">
3987 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3991 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
3992 from a vector at a specified index.</p>
3996 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
3997 of <a href="#t_vector">vector</a> type. The second operand is an index
3998 indicating the position from which to extract the element. The index may be
4002 <p>The result is a scalar of the same type as the element type of
4003 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4004 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4005 results are undefined.</p>
4009 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4014 <!-- _______________________________________________________________________ -->
4015 <div class="doc_subsubsection">
4016 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4019 <div class="doc_text">
4023 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4027 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4028 vector at a specified index.</p>
4031 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4032 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4033 whose type must equal the element type of the first operand. The third
4034 operand is an index indicating the position at which to insert the value.
4035 The index may be a variable.</p>
4038 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4039 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4040 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4041 results are undefined.</p>
4045 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4050 <!-- _______________________________________________________________________ -->
4051 <div class="doc_subsubsection">
4052 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4055 <div class="doc_text">
4059 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4063 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4064 from two input vectors, returning a vector with the same element type as the
4065 input and length that is the same as the shuffle mask.</p>
4068 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4069 with types that match each other. The third argument is a shuffle mask whose
4070 element type is always 'i32'. The result of the instruction is a vector
4071 whose length is the same as the shuffle mask and whose element type is the
4072 same as the element type of the first two operands.</p>
4074 <p>The shuffle mask operand is required to be a constant vector with either
4075 constant integer or undef values.</p>
4078 <p>The elements of the two input vectors are numbered from left to right across
4079 both of the vectors. The shuffle mask operand specifies, for each element of
4080 the result vector, which element of the two input vectors the result element
4081 gets. The element selector may be undef (meaning "don't care") and the
4082 second operand may be undef if performing a shuffle from only one vector.</p>
4086 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4087 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4088 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4089 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4090 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4091 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4092 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4093 <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>
4098 <!-- ======================================================================= -->
4099 <div class="doc_subsection">
4100 <a name="aggregateops">Aggregate Operations</a>
4103 <div class="doc_text">
4105 <p>LLVM supports several instructions for working with
4106 <a href="#t_aggregate">aggregate</a> values.</p>
4110 <!-- _______________________________________________________________________ -->
4111 <div class="doc_subsubsection">
4112 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4115 <div class="doc_text">
4119 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4123 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4124 from an <a href="#t_aggregate">aggregate</a> value.</p>
4127 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4128 of <a href="#t_struct">struct</a>, <a href="#t_union">union</a> or
4129 <a href="#t_array">array</a> type. The operands are constant indices to
4130 specify which value to extract in a similar manner as indices in a
4131 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4134 <p>The result is the value at the position in the aggregate specified by the
4139 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4144 <!-- _______________________________________________________________________ -->
4145 <div class="doc_subsubsection">
4146 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4149 <div class="doc_text">
4153 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx> <i>; yields <aggregate type></i>
4157 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4158 in an <a href="#t_aggregate">aggregate</a> value.</p>
4161 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4162 of <a href="#t_struct">struct</a>, <a href="#t_union">union</a> or
4163 <a href="#t_array">array</a> type. The second operand is a first-class
4164 value to insert. The following operands are constant indices indicating
4165 the position at which to insert the value in a similar manner as indices in a
4166 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction. The
4167 value to insert must have the same type as the value identified by the
4171 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4172 that of <tt>val</tt> except that the value at the position specified by the
4173 indices is that of <tt>elt</tt>.</p>
4177 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4178 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4184 <!-- ======================================================================= -->
4185 <div class="doc_subsection">
4186 <a name="memoryops">Memory Access and Addressing Operations</a>
4189 <div class="doc_text">
4191 <p>A key design point of an SSA-based representation is how it represents
4192 memory. In LLVM, no memory locations are in SSA form, which makes things
4193 very simple. This section describes how to read, write, and allocate
4198 <!-- _______________________________________________________________________ -->
4199 <div class="doc_subsubsection">
4200 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4203 <div class="doc_text">
4207 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4211 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4212 currently executing function, to be automatically released when this function
4213 returns to its caller. The object is always allocated in the generic address
4214 space (address space zero).</p>
4217 <p>The '<tt>alloca</tt>' instruction
4218 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4219 runtime stack, returning a pointer of the appropriate type to the program.
4220 If "NumElements" is specified, it is the number of elements allocated,
4221 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4222 specified, the value result of the allocation is guaranteed to be aligned to
4223 at least that boundary. If not specified, or if zero, the target can choose
4224 to align the allocation on any convenient boundary compatible with the
4227 <p>'<tt>type</tt>' may be any sized type.</p>
4230 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4231 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4232 memory is automatically released when the function returns. The
4233 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4234 variables that must have an address available. When the function returns
4235 (either with the <tt><a href="#i_ret">ret</a></tt>
4236 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
4237 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4241 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4242 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4243 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4244 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4249 <!-- _______________________________________________________________________ -->
4250 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
4251 Instruction</a> </div>
4253 <div class="doc_text">
4257 <result> = load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4258 <result> = volatile load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4259 !<index> = !{ i32 1 }
4263 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4266 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4267 from which to load. The pointer must point to
4268 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4269 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4270 number or order of execution of this <tt>load</tt> with other <a
4271 href="#volatile">volatile operations</a>.</p>
4273 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4274 operation (that is, the alignment of the memory address). A value of 0 or an
4275 omitted <tt>align</tt> argument means that the operation has the preferential
4276 alignment for the target. It is the responsibility of the code emitter to
4277 ensure that the alignment information is correct. Overestimating the
4278 alignment results in undefined behavior. Underestimating the alignment may
4279 produce less efficient code. An alignment of 1 is always safe.</p>
4281 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4282 metatadata name <index> corresponding to a metadata node with
4283 one <tt>i32</tt> entry of value 1. The existence of
4284 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4285 and code generator that this load is not expected to be reused in the cache.
4286 The code generator may select special instructions to save cache bandwidth,
4287 such as the <tt>MOVNT</tt> instruction on x86.</p>
4290 <p>The location of memory pointed to is loaded. If the value being loaded is of
4291 scalar type then the number of bytes read does not exceed the minimum number
4292 of bytes needed to hold all bits of the type. For example, loading an
4293 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4294 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4295 is undefined if the value was not originally written using a store of the
4300 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4301 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4302 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4307 <!-- _______________________________________________________________________ -->
4308 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
4309 Instruction</a> </div>
4311 <div class="doc_text">
4315 store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4316 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4320 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4323 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4324 and an address at which to store it. The type of the
4325 '<tt><pointer></tt>' operand must be a pointer to
4326 the <a href="#t_firstclass">first class</a> type of the
4327 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
4328 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
4329 order of execution of this <tt>store</tt> with other <a
4330 href="#volatile">volatile operations</a>.</p>
4332 <p>The optional constant "align" argument specifies the alignment of the
4333 operation (that is, the alignment of the memory address). A value of 0 or an
4334 omitted "align" argument means that the operation has the preferential
4335 alignment for the target. It is the responsibility of the code emitter to
4336 ensure that the alignment information is correct. Overestimating the
4337 alignment results in an undefined behavior. Underestimating the alignment may
4338 produce less efficient code. An alignment of 1 is always safe.</p>
4340 <p>The optional !nontemporal metadata must reference a single metatadata
4341 name <index> corresponding to a metadata node with one i32 entry of
4342 value 1. The existence of the !nontemporal metatadata on the
4343 instruction tells the optimizer and code generator that this load is
4344 not expected to be reused in the cache. The code generator may
4345 select special instructions to save cache bandwidth, such as the
4346 MOVNT instruction on x86.</p>
4350 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4351 location specified by the '<tt><pointer></tt>' operand. If
4352 '<tt><value></tt>' is of scalar type then the number of bytes written
4353 does not exceed the minimum number of bytes needed to hold all bits of the
4354 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4355 writing a value of a type like <tt>i20</tt> with a size that is not an
4356 integral number of bytes, it is unspecified what happens to the extra bits
4357 that do not belong to the type, but they will typically be overwritten.</p>
4361 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4362 store i32 3, i32* %ptr <i>; yields {void}</i>
4363 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4368 <!-- _______________________________________________________________________ -->
4369 <div class="doc_subsubsection">
4370 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
4373 <div class="doc_text">
4377 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4378 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4382 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4383 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
4384 It performs address calculation only and does not access memory.</p>
4387 <p>The first argument is always a pointer, and forms the basis of the
4388 calculation. The remaining arguments are indices that indicate which of the
4389 elements of the aggregate object are indexed. The interpretation of each
4390 index is dependent on the type being indexed into. The first index always
4391 indexes the pointer value given as the first argument, the second index
4392 indexes a value of the type pointed to (not necessarily the value directly
4393 pointed to, since the first index can be non-zero), etc. The first type
4394 indexed into must be a pointer value, subsequent types can be arrays,
4395 vectors, structs and unions. Note that subsequent types being indexed into
4396 can never be pointers, since that would require loading the pointer before
4397 continuing calculation.</p>
4399 <p>The type of each index argument depends on the type it is indexing into.
4400 When indexing into a (optionally packed) structure or union, only <tt>i32</tt>
4401 integer <b>constants</b> are allowed. When indexing into an array, pointer
4402 or vector, integers of any width are allowed, and they are not required to be
4405 <p>For example, let's consider a C code fragment and how it gets compiled to
4408 <div class="doc_code">
4421 int *foo(struct ST *s) {
4422 return &s[1].Z.B[5][13];
4427 <p>The LLVM code generated by the GCC frontend is:</p>
4429 <div class="doc_code">
4431 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4432 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
4434 define i32* @foo(%ST* %s) {
4436 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
4443 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
4444 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
4445 }</tt>' type, a structure. The second index indexes into the third element
4446 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
4447 i8 }</tt>' type, another structure. The third index indexes into the second
4448 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
4449 array. The two dimensions of the array are subscripted into, yielding an
4450 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
4451 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
4453 <p>Note that it is perfectly legal to index partially through a structure,
4454 returning a pointer to an inner element. Because of this, the LLVM code for
4455 the given testcase is equivalent to:</p>
4458 define i32* @foo(%ST* %s) {
4459 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
4460 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
4461 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
4462 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
4463 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
4468 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
4469 <tt>getelementptr</tt> is a <a href="#trapvalues">trap value</a> if the
4470 base pointer is not an <i>in bounds</i> address of an allocated object,
4471 or if any of the addresses that would be formed by successive addition of
4472 the offsets implied by the indices to the base address with infinitely
4473 precise arithmetic are not an <i>in bounds</i> address of that allocated
4474 object. The <i>in bounds</i> addresses for an allocated object are all
4475 the addresses that point into the object, plus the address one byte past
4478 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
4479 the base address with silently-wrapping two's complement arithmetic, and
4480 the result value of the <tt>getelementptr</tt> may be outside the object
4481 pointed to by the base pointer. The result value may not necessarily be
4482 used to access memory though, even if it happens to point into allocated
4483 storage. See the <a href="#pointeraliasing">Pointer Aliasing Rules</a>
4484 section for more information.</p>
4486 <p>The getelementptr instruction is often confusing. For some more insight into
4487 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
4491 <i>; yields [12 x i8]*:aptr</i>
4492 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4493 <i>; yields i8*:vptr</i>
4494 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4495 <i>; yields i8*:eptr</i>
4496 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4497 <i>; yields i32*:iptr</i>
4498 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4503 <!-- ======================================================================= -->
4504 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
4507 <div class="doc_text">
4509 <p>The instructions in this category are the conversion instructions (casting)
4510 which all take a single operand and a type. They perform various bit
4511 conversions on the operand.</p>
4515 <!-- _______________________________________________________________________ -->
4516 <div class="doc_subsubsection">
4517 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
4519 <div class="doc_text">
4523 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
4527 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
4528 type <tt>ty2</tt>.</p>
4531 <p>The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
4532 be an <a href="#t_integer">integer</a> type, and a type that specifies the
4533 size and type of the result, which must be
4534 an <a href="#t_integer">integer</a> type. The bit size of <tt>value</tt> must
4535 be larger than the bit size of <tt>ty2</tt>. Equal sized types are not
4539 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
4540 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
4541 source size must be larger than the destination size, <tt>trunc</tt> cannot
4542 be a <i>no-op cast</i>. It will always truncate bits.</p>
4546 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
4547 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
4548 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
4553 <!-- _______________________________________________________________________ -->
4554 <div class="doc_subsubsection">
4555 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
4557 <div class="doc_text">
4561 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
4565 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
4570 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
4571 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4572 also be of <a href="#t_integer">integer</a> type. The bit size of the
4573 <tt>value</tt> must be smaller than the bit size of the destination type,
4577 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
4578 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
4580 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
4584 %X = zext i32 257 to i64 <i>; yields i64:257</i>
4585 %Y = zext i1 true to i32 <i>; yields i32:1</i>
4590 <!-- _______________________________________________________________________ -->
4591 <div class="doc_subsubsection">
4592 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
4594 <div class="doc_text">
4598 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
4602 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
4605 <p>The '<tt>sext</tt>' instruction takes a value to cast, which must be of
4606 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4607 also be of <a href="#t_integer">integer</a> type. The bit size of the
4608 <tt>value</tt> must be smaller than the bit size of the destination type,
4612 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
4613 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
4614 of the type <tt>ty2</tt>.</p>
4616 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
4620 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
4621 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
4626 <!-- _______________________________________________________________________ -->
4627 <div class="doc_subsubsection">
4628 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
4631 <div class="doc_text">
4635 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
4639 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
4643 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
4644 point</a> value to cast and a <a href="#t_floating">floating point</a> type
4645 to cast it to. The size of <tt>value</tt> must be larger than the size of
4646 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
4647 <i>no-op cast</i>.</p>
4650 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
4651 <a href="#t_floating">floating point</a> type to a smaller
4652 <a href="#t_floating">floating point</a> type. If the value cannot fit
4653 within the destination type, <tt>ty2</tt>, then the results are
4658 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
4659 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
4664 <!-- _______________________________________________________________________ -->
4665 <div class="doc_subsubsection">
4666 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
4668 <div class="doc_text">
4672 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
4676 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
4677 floating point value.</p>
4680 <p>The '<tt>fpext</tt>' instruction takes a
4681 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
4682 a <a href="#t_floating">floating point</a> type to cast it to. The source
4683 type must be smaller than the destination type.</p>
4686 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
4687 <a href="#t_floating">floating point</a> type to a larger
4688 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
4689 used to make a <i>no-op cast</i> because it always changes bits. Use
4690 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
4694 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
4695 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
4700 <!-- _______________________________________________________________________ -->
4701 <div class="doc_subsubsection">
4702 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
4704 <div class="doc_text">
4708 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
4712 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
4713 unsigned integer equivalent of type <tt>ty2</tt>.</p>
4716 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
4717 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4718 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4719 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4720 vector integer type with the same number of elements as <tt>ty</tt></p>
4723 <p>The '<tt>fptoui</tt>' instruction converts its
4724 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4725 towards zero) unsigned integer value. If the value cannot fit
4726 in <tt>ty2</tt>, the results are undefined.</p>
4730 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
4731 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
4732 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
4737 <!-- _______________________________________________________________________ -->
4738 <div class="doc_subsubsection">
4739 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
4741 <div class="doc_text">
4745 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
4749 <p>The '<tt>fptosi</tt>' instruction converts
4750 <a href="#t_floating">floating point</a> <tt>value</tt> to
4751 type <tt>ty2</tt>.</p>
4754 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
4755 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4756 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4757 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4758 vector integer type with the same number of elements as <tt>ty</tt></p>
4761 <p>The '<tt>fptosi</tt>' instruction converts its
4762 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4763 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
4764 the results are undefined.</p>
4768 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
4769 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
4770 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
4775 <!-- _______________________________________________________________________ -->
4776 <div class="doc_subsubsection">
4777 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
4779 <div class="doc_text">
4783 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4787 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4788 integer and converts that value to the <tt>ty2</tt> type.</p>
4791 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4792 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4793 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4794 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4795 floating point type with the same number of elements as <tt>ty</tt></p>
4798 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4799 integer quantity and converts it to the corresponding floating point
4800 value. If the value cannot fit in the floating point value, the results are
4805 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4806 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4811 <!-- _______________________________________________________________________ -->
4812 <div class="doc_subsubsection">
4813 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4815 <div class="doc_text">
4819 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4823 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
4824 and converts that value to the <tt>ty2</tt> type.</p>
4827 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4828 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4829 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4830 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4831 floating point type with the same number of elements as <tt>ty</tt></p>
4834 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
4835 quantity and converts it to the corresponding floating point value. If the
4836 value cannot fit in the floating point value, the results are undefined.</p>
4840 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4841 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4846 <!-- _______________________________________________________________________ -->
4847 <div class="doc_subsubsection">
4848 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4850 <div class="doc_text">
4854 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4858 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4859 the integer type <tt>ty2</tt>.</p>
4862 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4863 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4864 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4867 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4868 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4869 truncating or zero extending that value to the size of the integer type. If
4870 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4871 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4872 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4877 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4878 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4883 <!-- _______________________________________________________________________ -->
4884 <div class="doc_subsubsection">
4885 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4887 <div class="doc_text">
4891 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4895 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
4896 pointer type, <tt>ty2</tt>.</p>
4899 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4900 value to cast, and a type to cast it to, which must be a
4901 <a href="#t_pointer">pointer</a> type.</p>
4904 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4905 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4906 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4907 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
4908 than the size of a pointer then a zero extension is done. If they are the
4909 same size, nothing is done (<i>no-op cast</i>).</p>
4913 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4914 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4915 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4920 <!-- _______________________________________________________________________ -->
4921 <div class="doc_subsubsection">
4922 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4924 <div class="doc_text">
4928 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4932 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4933 <tt>ty2</tt> without changing any bits.</p>
4936 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
4937 non-aggregate first class value, and a type to cast it to, which must also be
4938 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
4939 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
4940 identical. If the source type is a pointer, the destination type must also be
4941 a pointer. This instruction supports bitwise conversion of vectors to
4942 integers and to vectors of other types (as long as they have the same
4946 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4947 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4948 this conversion. The conversion is done as if the <tt>value</tt> had been
4949 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
4950 be converted to other pointer types with this instruction. To convert
4951 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
4952 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4956 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4957 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4958 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4963 <!-- ======================================================================= -->
4964 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4966 <div class="doc_text">
4968 <p>The instructions in this category are the "miscellaneous" instructions, which
4969 defy better classification.</p>
4973 <!-- _______________________________________________________________________ -->
4974 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4977 <div class="doc_text">
4981 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4985 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
4986 boolean values based on comparison of its two integer, integer vector, or
4987 pointer operands.</p>
4990 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4991 the condition code indicating the kind of comparison to perform. It is not a
4992 value, just a keyword. The possible condition code are:</p>
4995 <li><tt>eq</tt>: equal</li>
4996 <li><tt>ne</tt>: not equal </li>
4997 <li><tt>ugt</tt>: unsigned greater than</li>
4998 <li><tt>uge</tt>: unsigned greater or equal</li>
4999 <li><tt>ult</tt>: unsigned less than</li>
5000 <li><tt>ule</tt>: unsigned less or equal</li>
5001 <li><tt>sgt</tt>: signed greater than</li>
5002 <li><tt>sge</tt>: signed greater or equal</li>
5003 <li><tt>slt</tt>: signed less than</li>
5004 <li><tt>sle</tt>: signed less or equal</li>
5007 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
5008 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5009 typed. They must also be identical types.</p>
5012 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5013 condition code given as <tt>cond</tt>. The comparison performed always yields
5014 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
5015 result, as follows:</p>
5018 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
5019 <tt>false</tt> otherwise. No sign interpretation is necessary or
5022 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
5023 <tt>false</tt> otherwise. No sign interpretation is necessary or
5026 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
5027 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5029 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
5030 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5031 to <tt>op2</tt>.</li>
5033 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
5034 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5036 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
5037 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5039 <li><tt>sgt</tt>: interprets the operands as signed values and yields
5040 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5042 <li><tt>sge</tt>: interprets the operands as signed values and yields
5043 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5044 to <tt>op2</tt>.</li>
5046 <li><tt>slt</tt>: interprets the operands as signed values and yields
5047 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5049 <li><tt>sle</tt>: interprets the operands as signed values and yields
5050 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5053 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
5054 values are compared as if they were integers.</p>
5056 <p>If the operands are integer vectors, then they are compared element by
5057 element. The result is an <tt>i1</tt> vector with the same number of elements
5058 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
5062 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
5063 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
5064 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
5065 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
5066 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
5067 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
5070 <p>Note that the code generator does not yet support vector types with
5071 the <tt>icmp</tt> instruction.</p>
5075 <!-- _______________________________________________________________________ -->
5076 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
5079 <div class="doc_text">
5083 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5087 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
5088 values based on comparison of its operands.</p>
5090 <p>If the operands are floating point scalars, then the result type is a boolean
5091 (<a href="#t_integer"><tt>i1</tt></a>).</p>
5093 <p>If the operands are floating point vectors, then the result type is a vector
5094 of boolean with the same number of elements as the operands being
5098 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
5099 the condition code indicating the kind of comparison to perform. It is not a
5100 value, just a keyword. The possible condition code are:</p>
5103 <li><tt>false</tt>: no comparison, always returns false</li>
5104 <li><tt>oeq</tt>: ordered and equal</li>
5105 <li><tt>ogt</tt>: ordered and greater than </li>
5106 <li><tt>oge</tt>: ordered and greater than or equal</li>
5107 <li><tt>olt</tt>: ordered and less than </li>
5108 <li><tt>ole</tt>: ordered and less than or equal</li>
5109 <li><tt>one</tt>: ordered and not equal</li>
5110 <li><tt>ord</tt>: ordered (no nans)</li>
5111 <li><tt>ueq</tt>: unordered or equal</li>
5112 <li><tt>ugt</tt>: unordered or greater than </li>
5113 <li><tt>uge</tt>: unordered or greater than or equal</li>
5114 <li><tt>ult</tt>: unordered or less than </li>
5115 <li><tt>ule</tt>: unordered or less than or equal</li>
5116 <li><tt>une</tt>: unordered or not equal</li>
5117 <li><tt>uno</tt>: unordered (either nans)</li>
5118 <li><tt>true</tt>: no comparison, always returns true</li>
5121 <p><i>Ordered</i> means that neither operand is a QNAN while
5122 <i>unordered</i> means that either operand may be a QNAN.</p>
5124 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
5125 a <a href="#t_floating">floating point</a> type or
5126 a <a href="#t_vector">vector</a> of floating point type. They must have
5127 identical types.</p>
5130 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
5131 according to the condition code given as <tt>cond</tt>. If the operands are
5132 vectors, then the vectors are compared element by element. Each comparison
5133 performed always yields an <a href="#t_integer">i1</a> result, as
5137 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
5139 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5140 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5142 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5143 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5145 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5146 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5148 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5149 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5151 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5152 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5154 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5155 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5157 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
5159 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
5160 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5162 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
5163 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5165 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
5166 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5168 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
5169 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5171 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
5172 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5174 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
5175 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5177 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
5179 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5184 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
5185 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5186 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5187 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
5190 <p>Note that the code generator does not yet support vector types with
5191 the <tt>fcmp</tt> instruction.</p>
5195 <!-- _______________________________________________________________________ -->
5196 <div class="doc_subsubsection">
5197 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
5200 <div class="doc_text">
5204 <result> = phi <ty> [ <val0>, <label0>], ...
5208 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
5209 SSA graph representing the function.</p>
5212 <p>The type of the incoming values is specified with the first type field. After
5213 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5214 one pair for each predecessor basic block of the current block. Only values
5215 of <a href="#t_firstclass">first class</a> type may be used as the value
5216 arguments to the PHI node. Only labels may be used as the label
5219 <p>There must be no non-phi instructions between the start of a basic block and
5220 the PHI instructions: i.e. PHI instructions must be first in a basic
5223 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5224 occur on the edge from the corresponding predecessor block to the current
5225 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5226 value on the same edge).</p>
5229 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
5230 specified by the pair corresponding to the predecessor basic block that
5231 executed just prior to the current block.</p>
5235 Loop: ; Infinite loop that counts from 0 on up...
5236 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5237 %nextindvar = add i32 %indvar, 1
5243 <!-- _______________________________________________________________________ -->
5244 <div class="doc_subsubsection">
5245 <a name="i_select">'<tt>select</tt>' Instruction</a>
5248 <div class="doc_text">
5252 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5254 <i>selty</i> is either i1 or {<N x i1>}
5258 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
5259 condition, without branching.</p>
5263 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5264 values indicating the condition, and two values of the
5265 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
5266 vectors and the condition is a scalar, then entire vectors are selected, not
5267 individual elements.</p>
5270 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
5271 first value argument; otherwise, it returns the second value argument.</p>
5273 <p>If the condition is a vector of i1, then the value arguments must be vectors
5274 of the same size, and the selection is done element by element.</p>
5278 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
5281 <p>Note that the code generator does not yet support conditions
5282 with vector type.</p>
5286 <!-- _______________________________________________________________________ -->
5287 <div class="doc_subsubsection">
5288 <a name="i_call">'<tt>call</tt>' Instruction</a>
5291 <div class="doc_text">
5295 <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>]
5299 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
5302 <p>This instruction requires several arguments:</p>
5305 <li>The optional "tail" marker indicates that the callee function does not
5306 access any allocas or varargs in the caller. Note that calls may be
5307 marked "tail" even if they do not occur before
5308 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
5309 present, the function call is eligible for tail call optimization,
5310 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
5311 optimized into a jump</a>. The code generator may optimize calls marked
5312 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
5313 sibling call optimization</a> when the caller and callee have
5314 matching signatures, or 2) forced tail call optimization when the
5315 following extra requirements are met:
5317 <li>Caller and callee both have the calling
5318 convention <tt>fastcc</tt>.</li>
5319 <li>The call is in tail position (ret immediately follows call and ret
5320 uses value of call or is void).</li>
5321 <li>Option <tt>-tailcallopt</tt> is enabled,
5322 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
5323 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
5324 constraints are met.</a></li>
5328 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5329 convention</a> the call should use. If none is specified, the call
5330 defaults to using C calling conventions. The calling convention of the
5331 call must match the calling convention of the target function, or else the
5332 behavior is undefined.</li>
5334 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5335 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5336 '<tt>inreg</tt>' attributes are valid here.</li>
5338 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5339 type of the return value. Functions that return no value are marked
5340 <tt><a href="#t_void">void</a></tt>.</li>
5342 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5343 being invoked. The argument types must match the types implied by this
5344 signature. This type can be omitted if the function is not varargs and if
5345 the function type does not return a pointer to a function.</li>
5347 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5348 be invoked. In most cases, this is a direct function invocation, but
5349 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5350 to function value.</li>
5352 <li>'<tt>function args</tt>': argument list whose types match the function
5353 signature argument types and parameter attributes. All arguments must be
5354 of <a href="#t_firstclass">first class</a> type. If the function
5355 signature indicates the function accepts a variable number of arguments,
5356 the extra arguments can be specified.</li>
5358 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5359 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5360 '<tt>readnone</tt>' attributes are valid here.</li>
5364 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5365 a specified function, with its incoming arguments bound to the specified
5366 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5367 function, control flow continues with the instruction after the function
5368 call, and the return value of the function is bound to the result
5373 %retval = call i32 @test(i32 %argc)
5374 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
5375 %X = tail call i32 @foo() <i>; yields i32</i>
5376 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5377 call void %foo(i8 97 signext)
5379 %struct.A = type { i32, i8 }
5380 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
5381 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5382 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
5383 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
5384 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
5387 <p>llvm treats calls to some functions with names and arguments that match the
5388 standard C99 library as being the C99 library functions, and may perform
5389 optimizations or generate code for them under that assumption. This is
5390 something we'd like to change in the future to provide better support for
5391 freestanding environments and non-C-based languages.</p>
5395 <!-- _______________________________________________________________________ -->
5396 <div class="doc_subsubsection">
5397 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5400 <div class="doc_text">
5404 <resultval> = va_arg <va_list*> <arglist>, <argty>
5408 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
5409 the "variable argument" area of a function call. It is used to implement the
5410 <tt>va_arg</tt> macro in C.</p>
5413 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
5414 argument. It returns a value of the specified argument type and increments
5415 the <tt>va_list</tt> to point to the next argument. The actual type
5416 of <tt>va_list</tt> is target specific.</p>
5419 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
5420 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
5421 to the next argument. For more information, see the variable argument
5422 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
5424 <p>It is legal for this instruction to be called in a function which does not
5425 take a variable number of arguments, for example, the <tt>vfprintf</tt>
5428 <p><tt>va_arg</tt> is an LLVM instruction instead of
5429 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
5433 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
5435 <p>Note that the code generator does not yet fully support va_arg on many
5436 targets. Also, it does not currently support va_arg with aggregate types on
5441 <!-- *********************************************************************** -->
5442 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
5443 <!-- *********************************************************************** -->
5445 <div class="doc_text">
5447 <p>LLVM supports the notion of an "intrinsic function". These functions have
5448 well known names and semantics and are required to follow certain
5449 restrictions. Overall, these intrinsics represent an extension mechanism for
5450 the LLVM language that does not require changing all of the transformations
5451 in LLVM when adding to the language (or the bitcode reader/writer, the
5452 parser, etc...).</p>
5454 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
5455 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5456 begin with this prefix. Intrinsic functions must always be external
5457 functions: you cannot define the body of intrinsic functions. Intrinsic
5458 functions may only be used in call or invoke instructions: it is illegal to
5459 take the address of an intrinsic function. Additionally, because intrinsic
5460 functions are part of the LLVM language, it is required if any are added that
5461 they be documented here.</p>
5463 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5464 family of functions that perform the same operation but on different data
5465 types. Because LLVM can represent over 8 million different integer types,
5466 overloading is used commonly to allow an intrinsic function to operate on any
5467 integer type. One or more of the argument types or the result type can be
5468 overloaded to accept any integer type. Argument types may also be defined as
5469 exactly matching a previous argument's type or the result type. This allows
5470 an intrinsic function which accepts multiple arguments, but needs all of them
5471 to be of the same type, to only be overloaded with respect to a single
5472 argument or the result.</p>
5474 <p>Overloaded intrinsics will have the names of its overloaded argument types
5475 encoded into its function name, each preceded by a period. Only those types
5476 which are overloaded result in a name suffix. Arguments whose type is matched
5477 against another type do not. For example, the <tt>llvm.ctpop</tt> function
5478 can take an integer of any width and returns an integer of exactly the same
5479 integer width. This leads to a family of functions such as
5480 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
5481 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5482 suffix is required. Because the argument's type is matched against the return
5483 type, it does not require its own name suffix.</p>
5485 <p>To learn how to add an intrinsic function, please see the
5486 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
5490 <!-- ======================================================================= -->
5491 <div class="doc_subsection">
5492 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
5495 <div class="doc_text">
5497 <p>Variable argument support is defined in LLVM with
5498 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
5499 intrinsic functions. These functions are related to the similarly named
5500 macros defined in the <tt><stdarg.h></tt> header file.</p>
5502 <p>All of these functions operate on arguments that use a target-specific value
5503 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5504 not define what this type is, so all transformations should be prepared to
5505 handle these functions regardless of the type used.</p>
5507 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
5508 instruction and the variable argument handling intrinsic functions are
5511 <div class="doc_code">
5513 define i32 @test(i32 %X, ...) {
5514 ; Initialize variable argument processing
5516 %ap2 = bitcast i8** %ap to i8*
5517 call void @llvm.va_start(i8* %ap2)
5519 ; Read a single integer argument
5520 %tmp = va_arg i8** %ap, i32
5522 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5524 %aq2 = bitcast i8** %aq to i8*
5525 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5526 call void @llvm.va_end(i8* %aq2)
5528 ; Stop processing of arguments.
5529 call void @llvm.va_end(i8* %ap2)
5533 declare void @llvm.va_start(i8*)
5534 declare void @llvm.va_copy(i8*, i8*)
5535 declare void @llvm.va_end(i8*)
5541 <!-- _______________________________________________________________________ -->
5542 <div class="doc_subsubsection">
5543 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
5547 <div class="doc_text">
5551 declare void %llvm.va_start(i8* <arglist>)
5555 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
5556 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
5559 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
5562 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
5563 macro available in C. In a target-dependent way, it initializes
5564 the <tt>va_list</tt> element to which the argument points, so that the next
5565 call to <tt>va_arg</tt> will produce the first variable argument passed to
5566 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
5567 need to know the last argument of the function as the compiler can figure
5572 <!-- _______________________________________________________________________ -->
5573 <div class="doc_subsubsection">
5574 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
5577 <div class="doc_text">
5581 declare void @llvm.va_end(i8* <arglist>)
5585 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
5586 which has been initialized previously
5587 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
5588 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
5591 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
5594 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
5595 macro available in C. In a target-dependent way, it destroys
5596 the <tt>va_list</tt> element to which the argument points. Calls
5597 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
5598 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
5599 with calls to <tt>llvm.va_end</tt>.</p>
5603 <!-- _______________________________________________________________________ -->
5604 <div class="doc_subsubsection">
5605 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
5608 <div class="doc_text">
5612 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
5616 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
5617 from the source argument list to the destination argument list.</p>
5620 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
5621 The second argument is a pointer to a <tt>va_list</tt> element to copy
5625 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
5626 macro available in C. In a target-dependent way, it copies the
5627 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
5628 element. This intrinsic is necessary because
5629 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
5630 arbitrarily complex and require, for example, memory allocation.</p>
5634 <!-- ======================================================================= -->
5635 <div class="doc_subsection">
5636 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
5639 <div class="doc_text">
5641 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
5642 Collection</a> (GC) requires the implementation and generation of these
5643 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
5644 roots on the stack</a>, as well as garbage collector implementations that
5645 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
5646 barriers. Front-ends for type-safe garbage collected languages should generate
5647 these intrinsics to make use of the LLVM garbage collectors. For more details,
5648 see <a href="GarbageCollection.html">Accurate Garbage Collection with
5651 <p>The garbage collection intrinsics only operate on objects in the generic
5652 address space (address space zero).</p>
5656 <!-- _______________________________________________________________________ -->
5657 <div class="doc_subsubsection">
5658 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
5661 <div class="doc_text">
5665 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5669 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
5670 the code generator, and allows some metadata to be associated with it.</p>
5673 <p>The first argument specifies the address of a stack object that contains the
5674 root pointer. The second pointer (which must be either a constant or a
5675 global value address) contains the meta-data to be associated with the
5679 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
5680 location. At compile-time, the code generator generates information to allow
5681 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
5682 intrinsic may only be used in a function which <a href="#gc">specifies a GC
5687 <!-- _______________________________________________________________________ -->
5688 <div class="doc_subsubsection">
5689 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
5692 <div class="doc_text">
5696 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5700 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
5701 locations, allowing garbage collector implementations that require read
5705 <p>The second argument is the address to read from, which should be an address
5706 allocated from the garbage collector. The first object is a pointer to the
5707 start of the referenced object, if needed by the language runtime (otherwise
5711 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
5712 instruction, but may be replaced with substantially more complex code by the
5713 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
5714 may only be used in a function which <a href="#gc">specifies a GC
5719 <!-- _______________________________________________________________________ -->
5720 <div class="doc_subsubsection">
5721 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5724 <div class="doc_text">
5728 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5732 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5733 locations, allowing garbage collector implementations that require write
5734 barriers (such as generational or reference counting collectors).</p>
5737 <p>The first argument is the reference to store, the second is the start of the
5738 object to store it to, and the third is the address of the field of Obj to
5739 store to. If the runtime does not require a pointer to the object, Obj may
5743 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5744 instruction, but may be replaced with substantially more complex code by the
5745 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5746 may only be used in a function which <a href="#gc">specifies a GC
5751 <!-- ======================================================================= -->
5752 <div class="doc_subsection">
5753 <a name="int_codegen">Code Generator Intrinsics</a>
5756 <div class="doc_text">
5758 <p>These intrinsics are provided by LLVM to expose special features that may
5759 only be implemented with code generator support.</p>
5763 <!-- _______________________________________________________________________ -->
5764 <div class="doc_subsubsection">
5765 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5768 <div class="doc_text">
5772 declare i8 *@llvm.returnaddress(i32 <level>)
5776 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5777 target-specific value indicating the return address of the current function
5778 or one of its callers.</p>
5781 <p>The argument to this intrinsic indicates which function to return the address
5782 for. Zero indicates the calling function, one indicates its caller, etc.
5783 The argument is <b>required</b> to be a constant integer value.</p>
5786 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
5787 indicating the return address of the specified call frame, or zero if it
5788 cannot be identified. The value returned by this intrinsic is likely to be
5789 incorrect or 0 for arguments other than zero, so it should only be used for
5790 debugging purposes.</p>
5792 <p>Note that calling this intrinsic does not prevent function inlining or other
5793 aggressive transformations, so the value returned may not be that of the
5794 obvious source-language caller.</p>
5798 <!-- _______________________________________________________________________ -->
5799 <div class="doc_subsubsection">
5800 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5803 <div class="doc_text">
5807 declare i8 *@llvm.frameaddress(i32 <level>)
5811 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5812 target-specific frame pointer value for the specified stack frame.</p>
5815 <p>The argument to this intrinsic indicates which function to return the frame
5816 pointer for. Zero indicates the calling function, one indicates its caller,
5817 etc. The argument is <b>required</b> to be a constant integer value.</p>
5820 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
5821 indicating the frame address of the specified call frame, or zero if it
5822 cannot be identified. The value returned by this intrinsic is likely to be
5823 incorrect or 0 for arguments other than zero, so it should only be used for
5824 debugging purposes.</p>
5826 <p>Note that calling this intrinsic does not prevent function inlining or other
5827 aggressive transformations, so the value returned may not be that of the
5828 obvious source-language caller.</p>
5832 <!-- _______________________________________________________________________ -->
5833 <div class="doc_subsubsection">
5834 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5837 <div class="doc_text">
5841 declare i8 *@llvm.stacksave()
5845 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
5846 of the function stack, for use
5847 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
5848 useful for implementing language features like scoped automatic variable
5849 sized arrays in C99.</p>
5852 <p>This intrinsic returns a opaque pointer value that can be passed
5853 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
5854 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
5855 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
5856 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
5857 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
5858 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
5862 <!-- _______________________________________________________________________ -->
5863 <div class="doc_subsubsection">
5864 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5867 <div class="doc_text">
5871 declare void @llvm.stackrestore(i8 * %ptr)
5875 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5876 the function stack to the state it was in when the
5877 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
5878 executed. This is useful for implementing language features like scoped
5879 automatic variable sized arrays in C99.</p>
5882 <p>See the description
5883 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
5887 <!-- _______________________________________________________________________ -->
5888 <div class="doc_subsubsection">
5889 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5892 <div class="doc_text">
5896 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5900 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
5901 insert a prefetch instruction if supported; otherwise, it is a noop.
5902 Prefetches have no effect on the behavior of the program but can change its
5903 performance characteristics.</p>
5906 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
5907 specifier determining if the fetch should be for a read (0) or write (1),
5908 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5909 locality, to (3) - extremely local keep in cache. The <tt>rw</tt>
5910 and <tt>locality</tt> arguments must be constant integers.</p>
5913 <p>This intrinsic does not modify the behavior of the program. In particular,
5914 prefetches cannot trap and do not produce a value. On targets that support
5915 this intrinsic, the prefetch can provide hints to the processor cache for
5916 better performance.</p>
5920 <!-- _______________________________________________________________________ -->
5921 <div class="doc_subsubsection">
5922 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5925 <div class="doc_text">
5929 declare void @llvm.pcmarker(i32 <id>)
5933 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
5934 Counter (PC) in a region of code to simulators and other tools. The method
5935 is target specific, but it is expected that the marker will use exported
5936 symbols to transmit the PC of the marker. The marker makes no guarantees
5937 that it will remain with any specific instruction after optimizations. It is
5938 possible that the presence of a marker will inhibit optimizations. The
5939 intended use is to be inserted after optimizations to allow correlations of
5940 simulation runs.</p>
5943 <p><tt>id</tt> is a numerical id identifying the marker.</p>
5946 <p>This intrinsic does not modify the behavior of the program. Backends that do
5947 not support this intrinsic may ignore it.</p>
5951 <!-- _______________________________________________________________________ -->
5952 <div class="doc_subsubsection">
5953 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5956 <div class="doc_text">
5960 declare i64 @llvm.readcyclecounter( )
5964 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5965 counter register (or similar low latency, high accuracy clocks) on those
5966 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5967 should map to RPCC. As the backing counters overflow quickly (on the order
5968 of 9 seconds on alpha), this should only be used for small timings.</p>
5971 <p>When directly supported, reading the cycle counter should not modify any
5972 memory. Implementations are allowed to either return a application specific
5973 value or a system wide value. On backends without support, this is lowered
5974 to a constant 0.</p>
5978 <!-- ======================================================================= -->
5979 <div class="doc_subsection">
5980 <a name="int_libc">Standard C Library Intrinsics</a>
5983 <div class="doc_text">
5985 <p>LLVM provides intrinsics for a few important standard C library functions.
5986 These intrinsics allow source-language front-ends to pass information about
5987 the alignment of the pointer arguments to the code generator, providing
5988 opportunity for more efficient code generation.</p>
5992 <!-- _______________________________________________________________________ -->
5993 <div class="doc_subsubsection">
5994 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5997 <div class="doc_text">
6000 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
6001 integer bit width and for different address spaces. Not all targets support
6002 all bit widths however.</p>
6005 declare void @llvm.memcpy.p0i8.p0i8.i32(i8 * <dest>, i8 * <src>,
6006 i32 <len>, i32 <align>, i1 <isvolatile>)
6007 declare void @llvm.memcpy.p0i8.p0i8.i64(i8 * <dest>, i8 * <src>,
6008 i64 <len>, i32 <align>, i1 <isvolatile>)
6012 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6013 source location to the destination location.</p>
6015 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
6016 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6017 and the pointers can be in specified address spaces.</p>
6021 <p>The first argument is a pointer to the destination, the second is a pointer
6022 to the source. The third argument is an integer argument specifying the
6023 number of bytes to copy, the fourth argument is the alignment of the
6024 source and destination locations, and the fifth is a boolean indicating a
6025 volatile access.</p>
6027 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6028 then the caller guarantees that both the source and destination pointers are
6029 aligned to that boundary.</p>
6031 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6032 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
6033 The detailed access behavior is not very cleanly specified and it is unwise
6034 to depend on it.</p>
6038 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6039 source location to the destination location, which are not allowed to
6040 overlap. It copies "len" bytes of memory over. If the argument is known to
6041 be aligned to some boundary, this can be specified as the fourth argument,
6042 otherwise it should be set to 0 or 1.</p>
6046 <!-- _______________________________________________________________________ -->
6047 <div class="doc_subsubsection">
6048 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
6051 <div class="doc_text">
6054 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6055 width and for different address space. Not all targets support all bit
6059 declare void @llvm.memmove.p0i8.p0i8.i32(i8 * <dest>, i8 * <src>,
6060 i32 <len>, i32 <align>, i1 <isvolatile>)
6061 declare void @llvm.memmove.p0i8.p0i8.i64(i8 * <dest>, i8 * <src>,
6062 i64 <len>, i32 <align>, i1 <isvolatile>)
6066 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
6067 source location to the destination location. It is similar to the
6068 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
6071 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
6072 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6073 and the pointers can be in specified address spaces.</p>
6077 <p>The first argument is a pointer to the destination, the second is a pointer
6078 to the source. The third argument is an integer argument specifying the
6079 number of bytes to copy, the fourth argument is the alignment of the
6080 source and destination locations, and the fifth is a boolean indicating a
6081 volatile access.</p>
6083 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6084 then the caller guarantees that the source and destination pointers are
6085 aligned to that boundary.</p>
6087 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6088 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
6089 The detailed access behavior is not very cleanly specified and it is unwise
6090 to depend on it.</p>
6094 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
6095 source location to the destination location, which may overlap. It copies
6096 "len" bytes of memory over. If the argument is known to be aligned to some
6097 boundary, this can be specified as the fourth argument, otherwise it should
6098 be set to 0 or 1.</p>
6102 <!-- _______________________________________________________________________ -->
6103 <div class="doc_subsubsection">
6104 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
6107 <div class="doc_text">
6110 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6111 width and for different address spaces. Not all targets support all bit
6115 declare void @llvm.memset.p0i8.i32(i8 * <dest>, i8 <val>,
6116 i32 <len>, i32 <align>, i1 <isvolatile>)
6117 declare void @llvm.memset.p0i8.i64(i8 * <dest>, i8 <val>,
6118 i64 <len>, i32 <align>, i1 <isvolatile>)
6122 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
6123 particular byte value.</p>
6125 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
6126 intrinsic does not return a value, takes extra alignment/volatile arguments,
6127 and the destination can be in an arbitrary address space.</p>
6130 <p>The first argument is a pointer to the destination to fill, the second is the
6131 byte value to fill it with, the third argument is an integer argument
6132 specifying the number of bytes to fill, and the fourth argument is the known
6133 alignment of destination location.</p>
6135 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6136 then the caller guarantees that the destination pointer is aligned to that
6139 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6140 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
6141 The detailed access behavior is not very cleanly specified and it is unwise
6142 to depend on it.</p>
6145 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
6146 at the destination location. If the argument is known to be aligned to some
6147 boundary, this can be specified as the fourth argument, otherwise it should
6148 be set to 0 or 1.</p>
6152 <!-- _______________________________________________________________________ -->
6153 <div class="doc_subsubsection">
6154 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
6157 <div class="doc_text">
6160 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
6161 floating point or vector of floating point type. Not all targets support all
6165 declare float @llvm.sqrt.f32(float %Val)
6166 declare double @llvm.sqrt.f64(double %Val)
6167 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6168 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6169 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6173 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
6174 returning the same value as the libm '<tt>sqrt</tt>' functions would.
6175 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
6176 behavior for negative numbers other than -0.0 (which allows for better
6177 optimization, because there is no need to worry about errno being
6178 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
6181 <p>The argument and return value are floating point numbers of the same
6185 <p>This function returns the sqrt of the specified operand if it is a
6186 nonnegative floating point number.</p>
6190 <!-- _______________________________________________________________________ -->
6191 <div class="doc_subsubsection">
6192 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
6195 <div class="doc_text">
6198 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
6199 floating point or vector of floating point type. Not all targets support all
6203 declare float @llvm.powi.f32(float %Val, i32 %power)
6204 declare double @llvm.powi.f64(double %Val, i32 %power)
6205 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6206 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6207 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6211 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
6212 specified (positive or negative) power. The order of evaluation of
6213 multiplications is not defined. When a vector of floating point type is
6214 used, the second argument remains a scalar integer value.</p>
6217 <p>The second argument is an integer power, and the first is a value to raise to
6221 <p>This function returns the first value raised to the second power with an
6222 unspecified sequence of rounding operations.</p>
6226 <!-- _______________________________________________________________________ -->
6227 <div class="doc_subsubsection">
6228 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
6231 <div class="doc_text">
6234 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
6235 floating point or vector of floating point type. Not all targets support all
6239 declare float @llvm.sin.f32(float %Val)
6240 declare double @llvm.sin.f64(double %Val)
6241 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6242 declare fp128 @llvm.sin.f128(fp128 %Val)
6243 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6247 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
6250 <p>The argument and return value are floating point numbers of the same
6254 <p>This function returns the sine of the specified operand, returning the same
6255 values as the libm <tt>sin</tt> functions would, and handles error conditions
6256 in the same way.</p>
6260 <!-- _______________________________________________________________________ -->
6261 <div class="doc_subsubsection">
6262 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
6265 <div class="doc_text">
6268 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
6269 floating point or vector of floating point type. Not all targets support all
6273 declare float @llvm.cos.f32(float %Val)
6274 declare double @llvm.cos.f64(double %Val)
6275 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6276 declare fp128 @llvm.cos.f128(fp128 %Val)
6277 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6281 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
6284 <p>The argument and return value are floating point numbers of the same
6288 <p>This function returns the cosine of the specified operand, returning the same
6289 values as the libm <tt>cos</tt> functions would, and handles error conditions
6290 in the same way.</p>
6294 <!-- _______________________________________________________________________ -->
6295 <div class="doc_subsubsection">
6296 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
6299 <div class="doc_text">
6302 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
6303 floating point or vector of floating point type. Not all targets support all
6307 declare float @llvm.pow.f32(float %Val, float %Power)
6308 declare double @llvm.pow.f64(double %Val, double %Power)
6309 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6310 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6311 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6315 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
6316 specified (positive or negative) power.</p>
6319 <p>The second argument is a floating point power, and the first is a value to
6320 raise to that power.</p>
6323 <p>This function returns the first value raised to the second power, returning
6324 the same values as the libm <tt>pow</tt> functions would, and handles error
6325 conditions in the same way.</p>
6329 <!-- ======================================================================= -->
6330 <div class="doc_subsection">
6331 <a name="int_manip">Bit Manipulation Intrinsics</a>
6334 <div class="doc_text">
6336 <p>LLVM provides intrinsics for a few important bit manipulation operations.
6337 These allow efficient code generation for some algorithms.</p>
6341 <!-- _______________________________________________________________________ -->
6342 <div class="doc_subsubsection">
6343 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
6346 <div class="doc_text">
6349 <p>This is an overloaded intrinsic function. You can use bswap on any integer
6350 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
6353 declare i16 @llvm.bswap.i16(i16 <id>)
6354 declare i32 @llvm.bswap.i32(i32 <id>)
6355 declare i64 @llvm.bswap.i64(i64 <id>)
6359 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
6360 values with an even number of bytes (positive multiple of 16 bits). These
6361 are useful for performing operations on data that is not in the target's
6362 native byte order.</p>
6365 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
6366 and low byte of the input i16 swapped. Similarly,
6367 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
6368 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
6369 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
6370 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
6371 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
6372 more, respectively).</p>
6376 <!-- _______________________________________________________________________ -->
6377 <div class="doc_subsubsection">
6378 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
6381 <div class="doc_text">
6384 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6385 width. Not all targets support all bit widths however.</p>
6388 declare i8 @llvm.ctpop.i8(i8 <src>)
6389 declare i16 @llvm.ctpop.i16(i16 <src>)
6390 declare i32 @llvm.ctpop.i32(i32 <src>)
6391 declare i64 @llvm.ctpop.i64(i64 <src>)
6392 declare i256 @llvm.ctpop.i256(i256 <src>)
6396 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
6400 <p>The only argument is the value to be counted. The argument may be of any
6401 integer type. The return type must match the argument type.</p>
6404 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.</p>
6408 <!-- _______________________________________________________________________ -->
6409 <div class="doc_subsubsection">
6410 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
6413 <div class="doc_text">
6416 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
6417 integer bit width. Not all targets support all bit widths however.</p>
6420 declare i8 @llvm.ctlz.i8 (i8 <src>)
6421 declare i16 @llvm.ctlz.i16(i16 <src>)
6422 declare i32 @llvm.ctlz.i32(i32 <src>)
6423 declare i64 @llvm.ctlz.i64(i64 <src>)
6424 declare i256 @llvm.ctlz.i256(i256 <src>)
6428 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
6429 leading zeros in a variable.</p>
6432 <p>The only argument is the value to be counted. The argument may be of any
6433 integer type. The return type must match the argument type.</p>
6436 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
6437 zeros in a variable. If the src == 0 then the result is the size in bits of
6438 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
6442 <!-- _______________________________________________________________________ -->
6443 <div class="doc_subsubsection">
6444 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
6447 <div class="doc_text">
6450 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
6451 integer bit width. Not all targets support all bit widths however.</p>
6454 declare i8 @llvm.cttz.i8 (i8 <src>)
6455 declare i16 @llvm.cttz.i16(i16 <src>)
6456 declare i32 @llvm.cttz.i32(i32 <src>)
6457 declare i64 @llvm.cttz.i64(i64 <src>)
6458 declare i256 @llvm.cttz.i256(i256 <src>)
6462 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
6466 <p>The only argument is the value to be counted. The argument may be of any
6467 integer type. The return type must match the argument type.</p>
6470 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
6471 zeros in a variable. If the src == 0 then the result is the size in bits of
6472 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
6476 <!-- ======================================================================= -->
6477 <div class="doc_subsection">
6478 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
6481 <div class="doc_text">
6483 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
6487 <!-- _______________________________________________________________________ -->
6488 <div class="doc_subsubsection">
6489 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
6492 <div class="doc_text">
6495 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
6496 on any integer bit width.</p>
6499 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6500 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6501 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6505 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6506 a signed addition of the two arguments, and indicate whether an overflow
6507 occurred during the signed summation.</p>
6510 <p>The arguments (%a and %b) and the first element of the result structure may
6511 be of integer types of any bit width, but they must have the same bit
6512 width. The second element of the result structure must be of
6513 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6514 undergo signed addition.</p>
6517 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6518 a signed addition of the two variables. They return a structure — the
6519 first element of which is the signed summation, and the second element of
6520 which is a bit specifying if the signed summation resulted in an
6525 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6526 %sum = extractvalue {i32, i1} %res, 0
6527 %obit = extractvalue {i32, i1} %res, 1
6528 br i1 %obit, label %overflow, label %normal
6533 <!-- _______________________________________________________________________ -->
6534 <div class="doc_subsubsection">
6535 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
6538 <div class="doc_text">
6541 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
6542 on any integer bit width.</p>
6545 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6546 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6547 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6551 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6552 an unsigned addition of the two arguments, and indicate whether a carry
6553 occurred during the unsigned summation.</p>
6556 <p>The arguments (%a and %b) and the first element of the result structure may
6557 be of integer types of any bit width, but they must have the same bit
6558 width. The second element of the result structure must be of
6559 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6560 undergo unsigned addition.</p>
6563 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6564 an unsigned addition of the two arguments. They return a structure —
6565 the first element of which is the sum, and the second element of which is a
6566 bit specifying if the unsigned summation resulted in a carry.</p>
6570 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6571 %sum = extractvalue {i32, i1} %res, 0
6572 %obit = extractvalue {i32, i1} %res, 1
6573 br i1 %obit, label %carry, label %normal
6578 <!-- _______________________________________________________________________ -->
6579 <div class="doc_subsubsection">
6580 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6583 <div class="doc_text">
6586 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6587 on any integer bit width.</p>
6590 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6591 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6592 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6596 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6597 a signed subtraction of the two arguments, and indicate whether an overflow
6598 occurred during the signed subtraction.</p>
6601 <p>The arguments (%a and %b) and the first element of the result structure may
6602 be of integer types of any bit width, but they must have the same bit
6603 width. The second element of the result structure must be of
6604 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6605 undergo signed subtraction.</p>
6608 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6609 a signed subtraction of the two arguments. They return a structure —
6610 the first element of which is the subtraction, and the second element of
6611 which is a bit specifying if the signed subtraction resulted in an
6616 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6617 %sum = extractvalue {i32, i1} %res, 0
6618 %obit = extractvalue {i32, i1} %res, 1
6619 br i1 %obit, label %overflow, label %normal
6624 <!-- _______________________________________________________________________ -->
6625 <div class="doc_subsubsection">
6626 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6629 <div class="doc_text">
6632 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6633 on any integer bit width.</p>
6636 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6637 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6638 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6642 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6643 an unsigned subtraction of the two arguments, and indicate whether an
6644 overflow occurred during the unsigned subtraction.</p>
6647 <p>The arguments (%a and %b) and the first element of the result structure may
6648 be of integer types of any bit width, but they must have the same bit
6649 width. The second element of the result structure must be of
6650 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6651 undergo unsigned subtraction.</p>
6654 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6655 an unsigned subtraction of the two arguments. They return a structure —
6656 the first element of which is the subtraction, and the second element of
6657 which is a bit specifying if the unsigned subtraction resulted in an
6662 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6663 %sum = extractvalue {i32, i1} %res, 0
6664 %obit = extractvalue {i32, i1} %res, 1
6665 br i1 %obit, label %overflow, label %normal
6670 <!-- _______________________________________________________________________ -->
6671 <div class="doc_subsubsection">
6672 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6675 <div class="doc_text">
6678 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6679 on any integer bit width.</p>
6682 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6683 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6684 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6689 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6690 a signed multiplication of the two arguments, and indicate whether an
6691 overflow occurred during the signed multiplication.</p>
6694 <p>The arguments (%a and %b) and the first element of the result structure may
6695 be of integer types of any bit width, but they must have the same bit
6696 width. The second element of the result structure must be of
6697 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6698 undergo signed multiplication.</p>
6701 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6702 a signed multiplication of the two arguments. They return a structure —
6703 the first element of which is the multiplication, and the second element of
6704 which is a bit specifying if the signed multiplication resulted in an
6709 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6710 %sum = extractvalue {i32, i1} %res, 0
6711 %obit = extractvalue {i32, i1} %res, 1
6712 br i1 %obit, label %overflow, label %normal
6717 <!-- _______________________________________________________________________ -->
6718 <div class="doc_subsubsection">
6719 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6722 <div class="doc_text">
6725 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6726 on any integer bit width.</p>
6729 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6730 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6731 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6735 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6736 a unsigned multiplication of the two arguments, and indicate whether an
6737 overflow occurred during the unsigned multiplication.</p>
6740 <p>The arguments (%a and %b) and the first element of the result structure may
6741 be of integer types of any bit width, but they must have the same bit
6742 width. The second element of the result structure must be of
6743 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6744 undergo unsigned multiplication.</p>
6747 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6748 an unsigned multiplication of the two arguments. They return a structure
6749 — the first element of which is the multiplication, and the second
6750 element of which is a bit specifying if the unsigned multiplication resulted
6755 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6756 %sum = extractvalue {i32, i1} %res, 0
6757 %obit = extractvalue {i32, i1} %res, 1
6758 br i1 %obit, label %overflow, label %normal
6763 <!-- ======================================================================= -->
6764 <div class="doc_subsection">
6765 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
6768 <div class="doc_text">
6770 <p>Half precision floating point is a storage-only format. This means that it is
6771 a dense encoding (in memory) but does not support computation in the
6774 <p>This means that code must first load the half-precision floating point
6775 value as an i16, then convert it to float with <a
6776 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
6777 Computation can then be performed on the float value (including extending to
6778 double etc). To store the value back to memory, it is first converted to
6779 float if needed, then converted to i16 with
6780 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
6781 storing as an i16 value.</p>
6784 <!-- _______________________________________________________________________ -->
6785 <div class="doc_subsubsection">
6786 <a name="int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a>
6789 <div class="doc_text">
6793 declare i16 @llvm.convert.to.fp16(f32 %a)
6797 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
6798 a conversion from single precision floating point format to half precision
6799 floating point format.</p>
6802 <p>The intrinsic function contains single argument - the value to be
6806 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
6807 a conversion from single precision floating point format to half precision
6808 floating point format. The return value is an <tt>i16</tt> which
6809 contains the converted number.</p>
6813 %res = call i16 @llvm.convert.to.fp16(f32 %a)
6814 store i16 %res, i16* @x, align 2
6819 <!-- _______________________________________________________________________ -->
6820 <div class="doc_subsubsection">
6821 <a name="int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a>
6824 <div class="doc_text">
6828 declare f32 @llvm.convert.from.fp16(i16 %a)
6832 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
6833 a conversion from half precision floating point format to single precision
6834 floating point format.</p>
6837 <p>The intrinsic function contains single argument - the value to be
6841 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
6842 conversion from half single precision floating point format to single
6843 precision floating point format. The input half-float value is represented by
6844 an <tt>i16</tt> value.</p>
6848 %a = load i16* @x, align 2
6849 %res = call f32 @llvm.convert.from.fp16(i16 %a)
6854 <!-- ======================================================================= -->
6855 <div class="doc_subsection">
6856 <a name="int_debugger">Debugger Intrinsics</a>
6859 <div class="doc_text">
6861 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
6862 prefix), are described in
6863 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
6864 Level Debugging</a> document.</p>
6868 <!-- ======================================================================= -->
6869 <div class="doc_subsection">
6870 <a name="int_eh">Exception Handling Intrinsics</a>
6873 <div class="doc_text">
6875 <p>The LLVM exception handling intrinsics (which all start with
6876 <tt>llvm.eh.</tt> prefix), are described in
6877 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6878 Handling</a> document.</p>
6882 <!-- ======================================================================= -->
6883 <div class="doc_subsection">
6884 <a name="int_trampoline">Trampoline Intrinsic</a>
6887 <div class="doc_text">
6889 <p>This intrinsic makes it possible to excise one parameter, marked with
6890 the <tt>nest</tt> attribute, from a function. The result is a callable
6891 function pointer lacking the nest parameter - the caller does not need to
6892 provide a value for it. Instead, the value to use is stored in advance in a
6893 "trampoline", a block of memory usually allocated on the stack, which also
6894 contains code to splice the nest value into the argument list. This is used
6895 to implement the GCC nested function address extension.</p>
6897 <p>For example, if the function is
6898 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6899 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
6902 <div class="doc_code">
6904 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6905 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6906 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6907 %fp = bitcast i8* %p to i32 (i32, i32)*
6911 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6912 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6916 <!-- _______________________________________________________________________ -->
6917 <div class="doc_subsubsection">
6918 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6921 <div class="doc_text">
6925 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6929 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
6930 function pointer suitable for executing it.</p>
6933 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6934 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
6935 sufficiently aligned block of memory; this memory is written to by the
6936 intrinsic. Note that the size and the alignment are target-specific - LLVM
6937 currently provides no portable way of determining them, so a front-end that
6938 generates this intrinsic needs to have some target-specific knowledge.
6939 The <tt>func</tt> argument must hold a function bitcast to
6940 an <tt>i8*</tt>.</p>
6943 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
6944 dependent code, turning it into a function. A pointer to this function is
6945 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
6946 function pointer type</a> before being called. The new function's signature
6947 is the same as that of <tt>func</tt> with any arguments marked with
6948 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
6949 is allowed, and it must be of pointer type. Calling the new function is
6950 equivalent to calling <tt>func</tt> with the same argument list, but
6951 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
6952 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
6953 by <tt>tramp</tt> is modified, then the effect of any later call to the
6954 returned function pointer is undefined.</p>
6958 <!-- ======================================================================= -->
6959 <div class="doc_subsection">
6960 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6963 <div class="doc_text">
6965 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
6966 hardware constructs for atomic operations and memory synchronization. This
6967 provides an interface to the hardware, not an interface to the programmer. It
6968 is aimed at a low enough level to allow any programming models or APIs
6969 (Application Programming Interfaces) which need atomic behaviors to map
6970 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6971 hardware provides a "universal IR" for source languages, it also provides a
6972 starting point for developing a "universal" atomic operation and
6973 synchronization IR.</p>
6975 <p>These do <em>not</em> form an API such as high-level threading libraries,
6976 software transaction memory systems, atomic primitives, and intrinsic
6977 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6978 application libraries. The hardware interface provided by LLVM should allow
6979 a clean implementation of all of these APIs and parallel programming models.
6980 No one model or paradigm should be selected above others unless the hardware
6981 itself ubiquitously does so.</p>
6985 <!-- _______________________________________________________________________ -->
6986 <div class="doc_subsubsection">
6987 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6989 <div class="doc_text">
6992 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device> )
6996 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6997 specific pairs of memory access types.</p>
7000 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
7001 The first four arguments enables a specific barrier as listed below. The
7002 fifth argument specifies that the barrier applies to io or device or uncached
7006 <li><tt>ll</tt>: load-load barrier</li>
7007 <li><tt>ls</tt>: load-store barrier</li>
7008 <li><tt>sl</tt>: store-load barrier</li>
7009 <li><tt>ss</tt>: store-store barrier</li>
7010 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
7014 <p>This intrinsic causes the system to enforce some ordering constraints upon
7015 the loads and stores of the program. This barrier does not
7016 indicate <em>when</em> any events will occur, it only enforces
7017 an <em>order</em> in which they occur. For any of the specified pairs of load
7018 and store operations (f.ex. load-load, or store-load), all of the first
7019 operations preceding the barrier will complete before any of the second
7020 operations succeeding the barrier begin. Specifically the semantics for each
7021 pairing is as follows:</p>
7024 <li><tt>ll</tt>: All loads before the barrier must complete before any load
7025 after the barrier begins.</li>
7026 <li><tt>ls</tt>: All loads before the barrier must complete before any
7027 store after the barrier begins.</li>
7028 <li><tt>ss</tt>: All stores before the barrier must complete before any
7029 store after the barrier begins.</li>
7030 <li><tt>sl</tt>: All stores before the barrier must complete before any
7031 load after the barrier begins.</li>
7034 <p>These semantics are applied with a logical "and" behavior when more than one
7035 is enabled in a single memory barrier intrinsic.</p>
7037 <p>Backends may implement stronger barriers than those requested when they do
7038 not support as fine grained a barrier as requested. Some architectures do
7039 not need all types of barriers and on such architectures, these become
7044 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7045 %ptr = bitcast i8* %mallocP to i32*
7048 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
7049 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
7050 <i>; guarantee the above finishes</i>
7051 store i32 8, %ptr <i>; before this begins</i>
7056 <!-- _______________________________________________________________________ -->
7057 <div class="doc_subsubsection">
7058 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
7061 <div class="doc_text">
7064 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
7065 any integer bit width and for different address spaces. Not all targets
7066 support all bit widths however.</p>
7069 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
7070 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
7071 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
7072 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
7076 <p>This loads a value in memory and compares it to a given value. If they are
7077 equal, it stores a new value into the memory.</p>
7080 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
7081 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
7082 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
7083 this integer type. While any bit width integer may be used, targets may only
7084 lower representations they support in hardware.</p>
7087 <p>This entire intrinsic must be executed atomically. It first loads the value
7088 in memory pointed to by <tt>ptr</tt> and compares it with the
7089 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
7090 memory. The loaded value is yielded in all cases. This provides the
7091 equivalent of an atomic compare-and-swap operation within the SSA
7096 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7097 %ptr = bitcast i8* %mallocP to i32*
7100 %val1 = add i32 4, 4
7101 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
7102 <i>; yields {i32}:result1 = 4</i>
7103 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7104 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7106 %val2 = add i32 1, 1
7107 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
7108 <i>; yields {i32}:result2 = 8</i>
7109 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
7111 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
7116 <!-- _______________________________________________________________________ -->
7117 <div class="doc_subsubsection">
7118 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
7120 <div class="doc_text">
7123 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
7124 integer bit width. Not all targets support all bit widths however.</p>
7127 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
7128 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
7129 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
7130 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
7134 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
7135 the value from memory. It then stores the value in <tt>val</tt> in the memory
7136 at <tt>ptr</tt>.</p>
7139 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
7140 the <tt>val</tt> argument and the result must be integers of the same bit
7141 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
7142 integer type. The targets may only lower integer representations they
7146 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
7147 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
7148 equivalent of an atomic swap operation within the SSA framework.</p>
7152 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7153 %ptr = bitcast i8* %mallocP to i32*
7156 %val1 = add i32 4, 4
7157 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
7158 <i>; yields {i32}:result1 = 4</i>
7159 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7160 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7162 %val2 = add i32 1, 1
7163 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
7164 <i>; yields {i32}:result2 = 8</i>
7166 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
7167 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
7172 <!-- _______________________________________________________________________ -->
7173 <div class="doc_subsubsection">
7174 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
7178 <div class="doc_text">
7181 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
7182 any integer bit width. Not all targets support all bit widths however.</p>
7185 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
7186 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
7187 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
7188 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
7192 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
7193 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7196 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7197 and the second an integer value. The result is also an integer value. These
7198 integer types can have any bit width, but they must all have the same bit
7199 width. The targets may only lower integer representations they support.</p>
7202 <p>This intrinsic does a series of operations atomically. It first loads the
7203 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
7204 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
7208 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7209 %ptr = bitcast i8* %mallocP to i32*
7211 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
7212 <i>; yields {i32}:result1 = 4</i>
7213 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
7214 <i>; yields {i32}:result2 = 8</i>
7215 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
7216 <i>; yields {i32}:result3 = 10</i>
7217 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
7222 <!-- _______________________________________________________________________ -->
7223 <div class="doc_subsubsection">
7224 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
7228 <div class="doc_text">
7231 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
7232 any integer bit width and for different address spaces. Not all targets
7233 support all bit widths however.</p>
7236 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
7237 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
7238 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
7239 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
7243 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
7244 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7247 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7248 and the second an integer value. The result is also an integer value. These
7249 integer types can have any bit width, but they must all have the same bit
7250 width. The targets may only lower integer representations they support.</p>
7253 <p>This intrinsic does a series of operations atomically. It first loads the
7254 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
7255 result to <tt>ptr</tt>. It yields the original value stored
7256 at <tt>ptr</tt>.</p>
7260 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7261 %ptr = bitcast i8* %mallocP to i32*
7263 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
7264 <i>; yields {i32}:result1 = 8</i>
7265 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
7266 <i>; yields {i32}:result2 = 4</i>
7267 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
7268 <i>; yields {i32}:result3 = 2</i>
7269 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
7274 <!-- _______________________________________________________________________ -->
7275 <div class="doc_subsubsection">
7276 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
7277 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
7278 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
7279 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
7282 <div class="doc_text">
7285 <p>These are overloaded intrinsics. You can
7286 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
7287 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
7288 bit width and for different address spaces. Not all targets support all bit
7292 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
7293 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
7294 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
7295 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
7299 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
7300 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
7301 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
7302 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
7306 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
7307 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
7308 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
7309 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
7313 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
7314 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
7315 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
7316 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
7320 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
7321 the value stored in memory at <tt>ptr</tt>. It yields the original value
7322 at <tt>ptr</tt>.</p>
7325 <p>These intrinsics take two arguments, the first a pointer to an integer value
7326 and the second an integer value. The result is also an integer value. These
7327 integer types can have any bit width, but they must all have the same bit
7328 width. The targets may only lower integer representations they support.</p>
7331 <p>These intrinsics does a series of operations atomically. They first load the
7332 value stored at <tt>ptr</tt>. They then do the bitwise
7333 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
7334 original value stored at <tt>ptr</tt>.</p>
7338 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7339 %ptr = bitcast i8* %mallocP to i32*
7340 store i32 0x0F0F, %ptr
7341 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
7342 <i>; yields {i32}:result0 = 0x0F0F</i>
7343 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
7344 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
7345 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
7346 <i>; yields {i32}:result2 = 0xF0</i>
7347 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
7348 <i>; yields {i32}:result3 = FF</i>
7349 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
7354 <!-- _______________________________________________________________________ -->
7355 <div class="doc_subsubsection">
7356 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
7357 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
7358 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
7359 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
7362 <div class="doc_text">
7365 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
7366 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
7367 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
7368 address spaces. Not all targets support all bit widths however.</p>
7371 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
7372 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
7373 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
7374 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
7378 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
7379 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
7380 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
7381 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
7385 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
7386 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
7387 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
7388 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
7392 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
7393 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
7394 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
7395 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
7399 <p>These intrinsics takes the signed or unsigned minimum or maximum of
7400 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
7401 original value at <tt>ptr</tt>.</p>
7404 <p>These intrinsics take two arguments, the first a pointer to an integer value
7405 and the second an integer value. The result is also an integer value. These
7406 integer types can have any bit width, but they must all have the same bit
7407 width. The targets may only lower integer representations they support.</p>
7410 <p>These intrinsics does a series of operations atomically. They first load the
7411 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
7412 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
7413 yield the original value stored at <tt>ptr</tt>.</p>
7417 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7418 %ptr = bitcast i8* %mallocP to i32*
7420 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
7421 <i>; yields {i32}:result0 = 7</i>
7422 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
7423 <i>; yields {i32}:result1 = -2</i>
7424 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
7425 <i>; yields {i32}:result2 = 8</i>
7426 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
7427 <i>; yields {i32}:result3 = 8</i>
7428 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
7434 <!-- ======================================================================= -->
7435 <div class="doc_subsection">
7436 <a name="int_memorymarkers">Memory Use Markers</a>
7439 <div class="doc_text">
7441 <p>This class of intrinsics exists to information about the lifetime of memory
7442 objects and ranges where variables are immutable.</p>
7446 <!-- _______________________________________________________________________ -->
7447 <div class="doc_subsubsection">
7448 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
7451 <div class="doc_text">
7455 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
7459 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
7460 object's lifetime.</p>
7463 <p>The first argument is a constant integer representing the size of the
7464 object, or -1 if it is variable sized. The second argument is a pointer to
7468 <p>This intrinsic indicates that before this point in the code, the value of the
7469 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7470 never be used and has an undefined value. A load from the pointer that
7471 precedes this intrinsic can be replaced with
7472 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
7476 <!-- _______________________________________________________________________ -->
7477 <div class="doc_subsubsection">
7478 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
7481 <div class="doc_text">
7485 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
7489 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
7490 object's lifetime.</p>
7493 <p>The first argument is a constant integer representing the size of the
7494 object, or -1 if it is variable sized. The second argument is a pointer to
7498 <p>This intrinsic indicates that after this point in the code, the value of the
7499 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7500 never be used and has an undefined value. Any stores into the memory object
7501 following this intrinsic may be removed as dead.
7505 <!-- _______________________________________________________________________ -->
7506 <div class="doc_subsubsection">
7507 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
7510 <div class="doc_text">
7514 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>) readonly
7518 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
7519 a memory object will not change.</p>
7522 <p>The first argument is a constant integer representing the size of the
7523 object, or -1 if it is variable sized. The second argument is a pointer to
7527 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
7528 the return value, the referenced memory location is constant and
7533 <!-- _______________________________________________________________________ -->
7534 <div class="doc_subsubsection">
7535 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
7538 <div class="doc_text">
7542 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7546 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
7547 a memory object are mutable.</p>
7550 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
7551 The second argument is a constant integer representing the size of the
7552 object, or -1 if it is variable sized and the third argument is a pointer
7556 <p>This intrinsic indicates that the memory is mutable again.</p>
7560 <!-- ======================================================================= -->
7561 <div class="doc_subsection">
7562 <a name="int_general">General Intrinsics</a>
7565 <div class="doc_text">
7567 <p>This class of intrinsics is designed to be generic and has no specific
7572 <!-- _______________________________________________________________________ -->
7573 <div class="doc_subsubsection">
7574 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
7577 <div class="doc_text">
7581 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
7585 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
7588 <p>The first argument is a pointer to a value, the second is a pointer to a
7589 global string, the third is a pointer to a global string which is the source
7590 file name, and the last argument is the line number.</p>
7593 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7594 This can be useful for special purpose optimizations that want to look for
7595 these annotations. These have no other defined use, they are ignored by code
7596 generation and optimization.</p>
7600 <!-- _______________________________________________________________________ -->
7601 <div class="doc_subsubsection">
7602 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
7605 <div class="doc_text">
7608 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
7609 any integer bit width.</p>
7612 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
7613 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
7614 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
7615 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
7616 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
7620 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
7623 <p>The first argument is an integer value (result of some expression), the
7624 second is a pointer to a global string, the third is a pointer to a global
7625 string which is the source file name, and the last argument is the line
7626 number. It returns the value of the first argument.</p>
7629 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7630 arbitrary strings. This can be useful for special purpose optimizations that
7631 want to look for these annotations. These have no other defined use, they
7632 are ignored by code generation and optimization.</p>
7636 <!-- _______________________________________________________________________ -->
7637 <div class="doc_subsubsection">
7638 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
7641 <div class="doc_text">
7645 declare void @llvm.trap()
7649 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
7655 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7656 target does not have a trap instruction, this intrinsic will be lowered to
7657 the call of the <tt>abort()</tt> function.</p>
7661 <!-- _______________________________________________________________________ -->
7662 <div class="doc_subsubsection">
7663 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
7666 <div class="doc_text">
7670 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
7674 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
7675 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
7676 ensure that it is placed on the stack before local variables.</p>
7679 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
7680 arguments. The first argument is the value loaded from the stack
7681 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
7682 that has enough space to hold the value of the guard.</p>
7685 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7686 the <tt>AllocaInst</tt> stack slot to be before local variables on the
7687 stack. This is to ensure that if a local variable on the stack is
7688 overwritten, it will destroy the value of the guard. When the function exits,
7689 the guard on the stack is checked against the original guard. If they're
7690 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
7695 <!-- _______________________________________________________________________ -->
7696 <div class="doc_subsubsection">
7697 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
7700 <div class="doc_text">
7704 declare i32 @llvm.objectsize.i32( i8* <object>, i1 <type> )
7705 declare i64 @llvm.objectsize.i64( i8* <object>, i1 <type> )
7709 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information
7710 to the optimizers to discover at compile time either a) when an
7711 operation like memcpy will either overflow a buffer that corresponds to
7712 an object, or b) to determine that a runtime check for overflow isn't
7713 necessary. An object in this context means an allocation of a
7714 specific class, structure, array, or other object.</p>
7717 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
7718 argument is a pointer to or into the <tt>object</tt>. The second argument
7719 is a boolean 0 or 1. This argument determines whether you want the
7720 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
7721 1, variables are not allowed.</p>
7724 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
7725 representing the size of the object concerned or <tt>i32/i64 -1 or 0</tt>
7726 (depending on the <tt>type</tt> argument if the size cannot be determined
7727 at compile time.</p>
7731 <!-- *********************************************************************** -->
7734 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
7735 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
7736 <a href="http://validator.w3.org/check/referer"><img
7737 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
7739 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
7740 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
7741 Last modified: $Date$