blob: 2b9ee24bf41a66f85632eea4636191e132942bba [file] [log] [blame]
Misha Brukmandaa4cb02004-03-01 17:47:27 +00001<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
Misha Brukman9d0919f2003-11-08 01:05:38 +00003<html>
4<head>
5 <title>LLVM Assembly Language Reference Manual</title>
Reid Spencer3921c742004-08-26 20:44:00 +00006 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
Eric Christopher6c7e8a02009-12-05 02:46:03 +00008 <meta name="description"
Reid Spencer3921c742004-08-26 20:44:00 +00009 content="LLVM Assembly Language Reference Manual.">
Misha Brukman9d0919f2003-11-08 01:05:38 +000010 <link rel="stylesheet" href="llvm.css" type="text/css">
11</head>
Chris Lattnerd7923912004-05-23 21:06:01 +000012
Misha Brukman9d0919f2003-11-08 01:05:38 +000013<body>
Chris Lattnerd7923912004-05-23 21:06:01 +000014
NAKAMURA Takumi05d02652011-04-18 23:59:50 +000015<h1>LLVM Language Reference Manual</h1>
Chris Lattner00950542001-06-06 20:29:01 +000016<ol>
Misha Brukman9d0919f2003-11-08 01:05:38 +000017 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
Chris Lattnerfa730212004-12-09 16:11:40 +000020 <li><a href="#highlevel">High Level Structure</a>
21 <ol>
22 <li><a href="#modulestructure">Module Structure</a></li>
Bill Wendling3d10a5a2009-07-20 01:03:30 +000023 <li><a href="#linkage">Linkage Types</a>
24 <ol>
Bill Wendling987e7eb2009-07-20 02:41:50 +000025 <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>
Bill Wendling5e721d72010-07-01 21:55:59 +000027 <li><a href="#linkage_linker_private_weak">'<tt>linker_private_weak</tt>' Linkage</a></li>
Bill Wendling55ae5152010-08-20 22:05:50 +000028 <li><a href="#linkage_linker_private_weak_def_auto">'<tt>linker_private_weak_def_auto</tt>' Linkage</a></li>
Bill Wendling987e7eb2009-07-20 02:41:50 +000029 <li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
30 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
31 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
32 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
33 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
34 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
35 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
Chris Lattner5a2d8752009-10-10 18:26:06 +000036 <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
Bill Wendling987e7eb2009-07-20 02:41:50 +000037 <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
38 <li><a href="#linkage_external">'<tt>externally visible</tt>' Linkage</a></li>
39 <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
40 <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
Bill Wendling3d10a5a2009-07-20 01:03:30 +000041 </ol>
42 </li>
Chris Lattnerbad10ee2005-05-06 22:57:40 +000043 <li><a href="#callingconv">Calling Conventions</a></li>
Chris Lattnere7886e42009-01-11 20:53:49 +000044 <li><a href="#namedtypes">Named Types</a></li>
Chris Lattnerfa730212004-12-09 16:11:40 +000045 <li><a href="#globalvars">Global Variables</a></li>
Chris Lattner4e9aba72006-01-23 23:23:47 +000046 <li><a href="#functionstructure">Functions</a></li>
Dan Gohman0e451ce2008-10-14 16:51:45 +000047 <li><a href="#aliasstructure">Aliases</a></li>
Devang Patelcd1fd252010-01-11 19:35:55 +000048 <li><a href="#namedmetadatastructure">Named Metadata</a></li>
Reid Spencerca86e162006-12-31 07:07:53 +000049 <li><a href="#paramattrs">Parameter Attributes</a></li>
Devang Patel2c9c3e72008-09-26 23:51:19 +000050 <li><a href="#fnattrs">Function Attributes</a></li>
Gordon Henriksen80a75bf2007-12-10 03:18:06 +000051 <li><a href="#gc">Garbage Collector Names</a></li>
Chris Lattner4e9aba72006-01-23 23:23:47 +000052 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
Reid Spencerde151942007-02-19 23:54:10 +000053 <li><a href="#datalayout">Data Layout</a></li>
Dan Gohman556ca272009-07-27 18:07:55 +000054 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
Jeffrey Yasskin93e066d2010-04-26 21:21:24 +000055 <li><a href="#volatile">Volatile Memory Accesses</a></li>
Eli Friedman5b60e1b2011-07-20 21:35:53 +000056 <li><a href="#memmodel">Memory Model for Concurrent Operations</a></li>
Eli Friedmanff030482011-07-28 21:48:00 +000057 <li><a href="#ordering">Atomic Memory Ordering Constraints</a></li>
Chris Lattnerfa730212004-12-09 16:11:40 +000058 </ol>
59 </li>
Chris Lattner00950542001-06-06 20:29:01 +000060 <li><a href="#typesystem">Type System</a>
61 <ol>
Chris Lattner4f69f462008-01-04 04:32:38 +000062 <li><a href="#t_classifications">Type Classifications</a></li>
Eric Christopher6c7e8a02009-12-05 02:46:03 +000063 <li><a href="#t_primitive">Primitive Types</a>
Chris Lattner261efe92003-11-25 01:02:51 +000064 <ol>
Nick Lewyckyec38da42009-09-27 00:45:11 +000065 <li><a href="#t_integer">Integer Type</a></li>
Chris Lattner4f69f462008-01-04 04:32:38 +000066 <li><a href="#t_floating">Floating Point Types</a></li>
Dale Johannesen21fe99b2010-10-01 00:48:59 +000067 <li><a href="#t_x86mmx">X86mmx Type</a></li>
Chris Lattner4f69f462008-01-04 04:32:38 +000068 <li><a href="#t_void">Void Type</a></li>
69 <li><a href="#t_label">Label Type</a></li>
Nick Lewycky7a0370f2009-05-30 05:06:04 +000070 <li><a href="#t_metadata">Metadata Type</a></li>
Chris Lattner261efe92003-11-25 01:02:51 +000071 </ol>
72 </li>
Chris Lattner00950542001-06-06 20:29:01 +000073 <li><a href="#t_derived">Derived Types</a>
74 <ol>
Chris Lattnerfdfeb692010-02-12 20:49:41 +000075 <li><a href="#t_aggregate">Aggregate Types</a>
76 <ol>
77 <li><a href="#t_array">Array Type</a></li>
78 <li><a href="#t_struct">Structure Type</a></li>
Chris Lattner628ed392011-07-23 19:59:08 +000079 <li><a href="#t_opaque">Opaque Structure Types</a></li>
Chris Lattnerfdfeb692010-02-12 20:49:41 +000080 <li><a href="#t_vector">Vector Type</a></li>
81 </ol>
82 </li>
Misha Brukman9d0919f2003-11-08 01:05:38 +000083 <li><a href="#t_function">Function Type</a></li>
84 <li><a href="#t_pointer">Pointer Type</a></li>
Chris Lattner261efe92003-11-25 01:02:51 +000085 </ol>
86 </li>
87 </ol>
88 </li>
Chris Lattnerfa730212004-12-09 16:11:40 +000089 <li><a href="#constants">Constants</a>
Chris Lattnerc3f59762004-12-09 17:30:23 +000090 <ol>
Dan Gohman0e451ce2008-10-14 16:51:45 +000091 <li><a href="#simpleconstants">Simple Constants</a></li>
Chris Lattner70882792009-02-28 18:32:25 +000092 <li><a href="#complexconstants">Complex Constants</a></li>
Dan Gohman0e451ce2008-10-14 16:51:45 +000093 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
94 <li><a href="#undefvalues">Undefined Values</a></li>
Dan Gohmanfff6c532010-04-22 23:14:21 +000095 <li><a href="#trapvalues">Trap Values</a></li>
Chris Lattnerf9d078e2009-10-27 21:19:13 +000096 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
Dan Gohman0e451ce2008-10-14 16:51:45 +000097 <li><a href="#constantexprs">Constant Expressions</a></li>
Chris Lattnerc3f59762004-12-09 17:30:23 +000098 </ol>
Chris Lattner261efe92003-11-25 01:02:51 +000099 </li>
Chris Lattnere87d6532006-01-25 23:47:57 +0000100 <li><a href="#othervalues">Other Values</a>
101 <ol>
Dan Gohman0e451ce2008-10-14 16:51:45 +0000102 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
Devang Patelcd1fd252010-01-11 19:35:55 +0000103 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a></li>
Chris Lattnere87d6532006-01-25 23:47:57 +0000104 </ol>
105 </li>
Chris Lattner857755c2009-07-20 05:55:19 +0000106 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
107 <ol>
108 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
Chris Lattner401e10c2009-07-20 06:14:25 +0000109 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
110 Global Variable</a></li>
Chris Lattner857755c2009-07-20 05:55:19 +0000111 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
112 Global Variable</a></li>
113 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
114 Global Variable</a></li>
115 </ol>
116 </li>
Chris Lattner00950542001-06-06 20:29:01 +0000117 <li><a href="#instref">Instruction Reference</a>
118 <ol>
119 <li><a href="#terminators">Terminator Instructions</a>
120 <ol>
Chris Lattner261efe92003-11-25 01:02:51 +0000121 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
122 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
Misha Brukman9d0919f2003-11-08 01:05:38 +0000123 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
Chris Lattnerab21db72009-10-28 00:19:10 +0000124 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
Misha Brukman9d0919f2003-11-08 01:05:38 +0000125 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
Chris Lattner261efe92003-11-25 01:02:51 +0000126 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
Bill Wendlingdccc03b2011-07-31 06:30:59 +0000127 <li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li>
Chris Lattner35eca582004-10-16 18:04:13 +0000128 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
Chris Lattner261efe92003-11-25 01:02:51 +0000129 </ol>
130 </li>
Chris Lattner00950542001-06-06 20:29:01 +0000131 <li><a href="#binaryops">Binary Operations</a>
132 <ol>
Chris Lattner261efe92003-11-25 01:02:51 +0000133 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
Dan Gohmanae3a0be2009-06-04 22:49:04 +0000134 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
Chris Lattner261efe92003-11-25 01:02:51 +0000135 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
Dan Gohmanae3a0be2009-06-04 22:49:04 +0000136 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
Chris Lattner261efe92003-11-25 01:02:51 +0000137 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
Dan Gohmanae3a0be2009-06-04 22:49:04 +0000138 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
Reid Spencer1628cec2006-10-26 06:15:43 +0000139 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
140 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
141 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
Reid Spencer0a783f72006-11-02 01:53:59 +0000142 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
143 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
144 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
Chris Lattner261efe92003-11-25 01:02:51 +0000145 </ol>
146 </li>
Chris Lattner00950542001-06-06 20:29:01 +0000147 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
148 <ol>
Reid Spencer8e11bf82007-02-02 13:57:07 +0000149 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
150 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
151 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
Misha Brukman9d0919f2003-11-08 01:05:38 +0000152 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
Chris Lattner261efe92003-11-25 01:02:51 +0000153 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
Misha Brukman9d0919f2003-11-08 01:05:38 +0000154 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
Chris Lattner261efe92003-11-25 01:02:51 +0000155 </ol>
156 </li>
Chris Lattner3df241e2006-04-08 23:07:04 +0000157 <li><a href="#vectorops">Vector Operations</a>
158 <ol>
159 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
160 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
161 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
Chris Lattner3df241e2006-04-08 23:07:04 +0000162 </ol>
163 </li>
Dan Gohmana334d5f2008-05-12 23:51:09 +0000164 <li><a href="#aggregateops">Aggregate Operations</a>
165 <ol>
166 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
167 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
168 </ol>
169 </li>
Chris Lattner884a9702006-08-15 00:45:58 +0000170 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
Chris Lattner00950542001-06-06 20:29:01 +0000171 <ol>
Eli Friedmanff030482011-07-28 21:48:00 +0000172 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
173 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
174 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
175 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
176 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
177 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
Robert Bocchino7b81c752006-02-17 21:18:08 +0000178 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
Chris Lattner261efe92003-11-25 01:02:51 +0000179 </ol>
180 </li>
Reid Spencer2fd21e62006-11-08 01:18:52 +0000181 <li><a href="#convertops">Conversion Operations</a>
Reid Spencer9dee3ac2006-11-08 01:11:31 +0000182 <ol>
183 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
184 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
185 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
186 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
187 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
Reid Spencerd4448792006-11-09 23:03:26 +0000188 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
189 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
190 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
191 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
Reid Spencer72679252006-11-11 21:00:47 +0000192 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
193 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
Reid Spencer5c0ef472006-11-11 23:08:07 +0000194 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
Reid Spencer9dee3ac2006-11-08 01:11:31 +0000195 </ol>
Dan Gohman0e451ce2008-10-14 16:51:45 +0000196 </li>
Chris Lattner00950542001-06-06 20:29:01 +0000197 <li><a href="#otherops">Other Operations</a>
198 <ol>
Reid Spencerf3a70a62006-11-18 21:50:54 +0000199 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
200 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
Chris Lattner261efe92003-11-25 01:02:51 +0000201 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
Chris Lattnercc37aae2004-03-12 05:50:16 +0000202 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
Chris Lattner261efe92003-11-25 01:02:51 +0000203 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
Chris Lattnerfb6977d2006-01-13 23:26:01 +0000204 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
Bill Wendlingf78faf82011-08-02 21:52:38 +0000205 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
Chris Lattner00950542001-06-06 20:29:01 +0000206 </ol>
Chris Lattner261efe92003-11-25 01:02:51 +0000207 </li>
Chris Lattner00950542001-06-06 20:29:01 +0000208 </ol>
Chris Lattner261efe92003-11-25 01:02:51 +0000209 </li>
Chris Lattnerd9ad5b32003-05-08 04:57:36 +0000210 <li><a href="#intrinsics">Intrinsic Functions</a>
Chris Lattnerd9ad5b32003-05-08 04:57:36 +0000211 <ol>
Chris Lattner261efe92003-11-25 01:02:51 +0000212 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
213 <ol>
Reid Spencera3e435f2007-04-04 02:42:35 +0000214 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
215 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
216 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
Chris Lattner261efe92003-11-25 01:02:51 +0000217 </ol>
218 </li>
Chris Lattnerd7923912004-05-23 21:06:01 +0000219 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
220 <ol>
Reid Spencera3e435f2007-04-04 02:42:35 +0000221 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
222 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
223 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
Chris Lattnerd7923912004-05-23 21:06:01 +0000224 </ol>
225 </li>
Chris Lattner10610642004-02-14 04:08:35 +0000226 <li><a href="#int_codegen">Code Generator Intrinsics</a>
227 <ol>
Reid Spencera3e435f2007-04-04 02:42:35 +0000228 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
229 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
230 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
231 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
232 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
233 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
Dan Gohman31f1af12010-05-26 21:56:15 +0000234 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
John Criswell7123e272004-04-09 16:43:20 +0000235 </ol>
236 </li>
Chris Lattner33aec9e2004-02-12 17:01:32 +0000237 <li><a href="#int_libc">Standard C Library Intrinsics</a>
238 <ol>
Reid Spencera3e435f2007-04-04 02:42:35 +0000239 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
240 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
241 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
242 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
243 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
Dan Gohman91c284c2007-10-15 20:30:11 +0000244 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
245 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
246 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
Dan Gohman08b280b2011-05-27 00:36:31 +0000247 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
248 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
Cameron Zwarich33390842011-07-08 21:39:21 +0000249 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
Chris Lattner33aec9e2004-02-12 17:01:32 +0000250 </ol>
251 </li>
Nate Begeman7e36c472006-01-13 23:26:38 +0000252 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
Andrew Lenharthec370fd2005-05-03 18:01:48 +0000253 <ol>
Reid Spencera3e435f2007-04-04 02:42:35 +0000254 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
Chris Lattner8a886be2006-01-16 22:34:14 +0000255 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
256 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
257 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
Andrew Lenharthec370fd2005-05-03 18:01:48 +0000258 </ol>
259 </li>
Bill Wendlingac1df8e2009-02-08 01:40:31 +0000260 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
261 <ol>
Bill Wendlingda01af72009-02-08 04:04:40 +0000262 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
263 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
264 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
265 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
266 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
Bill Wendling41b485c2009-02-08 23:00:09 +0000267 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
Bill Wendlingac1df8e2009-02-08 01:40:31 +0000268 </ol>
269 </li>
Anton Korobeynikovf02e7302010-03-14 18:42:47 +0000270 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
271 <ol>
Chris Lattner82c3dc62010-03-14 23:03:31 +0000272 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
273 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
Anton Korobeynikovf02e7302010-03-14 18:42:47 +0000274 </ol>
275 </li>
Chris Lattnerd7923912004-05-23 21:06:01 +0000276 <li><a href="#int_debugger">Debugger intrinsics</a></li>
Jim Laskeydd4ef1b2007-03-14 19:31:19 +0000277 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
Duncan Sandsf7331b32007-09-11 14:10:23 +0000278 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
Duncan Sands36397f52007-07-27 12:58:54 +0000279 <ol>
280 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
Duncan Sands36397f52007-07-27 12:58:54 +0000281 </ol>
282 </li>
Bill Wendling3c44f5b2008-11-18 22:10:53 +0000283 <li><a href="#int_atomics">Atomic intrinsics</a>
284 <ol>
285 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
286 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
287 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
288 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
289 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
290 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
291 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
292 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
293 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
294 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
295 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
296 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
297 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
298 </ol>
299 </li>
Nick Lewyckycc271862009-10-13 07:03:23 +0000300 <li><a href="#int_memorymarkers">Memory Use Markers</a>
301 <ol>
302 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
303 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
304 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
305 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
306 </ol>
307 </li>
Reid Spencer20677642007-07-20 19:59:11 +0000308 <li><a href="#int_general">General intrinsics</a>
Tanya Lattner6d806e92007-06-15 20:50:54 +0000309 <ol>
Reid Spencer20677642007-07-20 19:59:11 +0000310 <li><a href="#int_var_annotation">
Bill Wendling69e4adb2008-11-19 05:56:17 +0000311 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
Tanya Lattnerb6367882007-09-21 22:59:12 +0000312 <li><a href="#int_annotation">
Bill Wendling69e4adb2008-11-19 05:56:17 +0000313 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
Anton Korobeynikov4cb86182008-01-15 22:31:34 +0000314 <li><a href="#int_trap">
Bill Wendling69e4adb2008-11-19 05:56:17 +0000315 '<tt>llvm.trap</tt>' Intrinsic</a></li>
316 <li><a href="#int_stackprotector">
317 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
Eric Christopher0e671492009-11-30 08:03:53 +0000318 <li><a href="#int_objectsize">
319 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
Tanya Lattnerb6367882007-09-21 22:59:12 +0000320 </ol>
Tanya Lattner6d806e92007-06-15 20:50:54 +0000321 </li>
Chris Lattner261efe92003-11-25 01:02:51 +0000322 </ol>
323 </li>
Chris Lattner00950542001-06-06 20:29:01 +0000324</ol>
Chris Lattnerd7923912004-05-23 21:06:01 +0000325
326<div class="doc_author">
327 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
328 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
Misha Brukman9d0919f2003-11-08 01:05:38 +0000329</div>
Chris Lattnerd7923912004-05-23 21:06:01 +0000330
Chris Lattner00950542001-06-06 20:29:01 +0000331<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000332<h2><a name="abstract">Abstract</a></h2>
Chris Lattner261efe92003-11-25 01:02:51 +0000333<!-- *********************************************************************** -->
Chris Lattnerd7923912004-05-23 21:06:01 +0000334
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000335<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000336
337<p>This document is a reference manual for the LLVM assembly language. LLVM is
338 a Static Single Assignment (SSA) based representation that provides type
339 safety, low-level operations, flexibility, and the capability of representing
340 'all' high-level languages cleanly. It is the common code representation
341 used throughout all phases of the LLVM compilation strategy.</p>
342
Misha Brukman9d0919f2003-11-08 01:05:38 +0000343</div>
Chris Lattnerd7923912004-05-23 21:06:01 +0000344
Chris Lattner00950542001-06-06 20:29:01 +0000345<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000346<h2><a name="introduction">Introduction</a></h2>
Chris Lattner261efe92003-11-25 01:02:51 +0000347<!-- *********************************************************************** -->
Chris Lattnerd7923912004-05-23 21:06:01 +0000348
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000349<div>
Chris Lattnerd7923912004-05-23 21:06:01 +0000350
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000351<p>The LLVM code representation is designed to be used in three different forms:
352 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
353 for fast loading by a Just-In-Time compiler), and as a human readable
354 assembly language representation. This allows LLVM to provide a powerful
355 intermediate representation for efficient compiler transformations and
356 analysis, while providing a natural means to debug and visualize the
357 transformations. The three different forms of LLVM are all equivalent. This
358 document describes the human readable representation and notation.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +0000359
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000360<p>The LLVM representation aims to be light-weight and low-level while being
361 expressive, typed, and extensible at the same time. It aims to be a
362 "universal IR" of sorts, by being at a low enough level that high-level ideas
363 may be cleanly mapped to it (similar to how microprocessors are "universal
364 IR's", allowing many source languages to be mapped to them). By providing
365 type information, LLVM can be used as the target of optimizations: for
366 example, through pointer analysis, it can be proven that a C automatic
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000367 variable is never accessed outside of the current function, allowing it to
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000368 be promoted to a simple SSA value instead of a memory location.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +0000369
Chris Lattner00950542001-06-06 20:29:01 +0000370<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000371<h4>
372 <a name="wellformed">Well-Formedness</a>
373</h4>
Chris Lattnerd7923912004-05-23 21:06:01 +0000374
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000375<div>
Chris Lattnerd7923912004-05-23 21:06:01 +0000376
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000377<p>It is important to note that this document describes 'well formed' LLVM
378 assembly language. There is a difference between what the parser accepts and
379 what is considered 'well formed'. For example, the following instruction is
380 syntactically okay, but not well formed:</p>
Chris Lattnerd7923912004-05-23 21:06:01 +0000381
Benjamin Kramer26fe25f2010-07-13 12:26:09 +0000382<pre class="doc_code">
Bill Wendling2f7a8b02007-05-29 09:04:49 +0000383%x = <a href="#i_add">add</a> i32 1, %x
Chris Lattnerd7923912004-05-23 21:06:01 +0000384</pre>
385
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000386<p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
387 LLVM infrastructure provides a verification pass that may be used to verify
388 that an LLVM module is well formed. This pass is automatically run by the
389 parser after parsing input assembly and by the optimizer before it outputs
390 bitcode. The violations pointed out by the verifier pass indicate bugs in
391 transformation passes or input to the parser.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000392
Bill Wendling2f7a8b02007-05-29 09:04:49 +0000393</div>
Chris Lattnerd7923912004-05-23 21:06:01 +0000394
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000395</div>
396
Chris Lattnercc689392007-10-03 17:34:29 +0000397<!-- Describe the typesetting conventions here. -->
Chris Lattnerd7923912004-05-23 21:06:01 +0000398
Chris Lattner00950542001-06-06 20:29:01 +0000399<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000400<h2><a name="identifiers">Identifiers</a></h2>
Chris Lattner00950542001-06-06 20:29:01 +0000401<!-- *********************************************************************** -->
Chris Lattnerd7923912004-05-23 21:06:01 +0000402
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000403<div>
Chris Lattnerd7923912004-05-23 21:06:01 +0000404
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000405<p>LLVM identifiers come in two basic types: global and local. Global
406 identifiers (functions, global variables) begin with the <tt>'@'</tt>
407 character. Local identifiers (register names, types) begin with
408 the <tt>'%'</tt> character. Additionally, there are three different formats
409 for identifiers, for different purposes:</p>
Chris Lattnerd7923912004-05-23 21:06:01 +0000410
Chris Lattner00950542001-06-06 20:29:01 +0000411<ol>
Reid Spencer2c452282007-08-07 14:34:28 +0000412 <li>Named values are represented as a string of characters with their prefix.
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000413 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
414 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
415 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
416 other characters in their names can be surrounded with quotes. Special
417 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
418 ASCII code for the character in hexadecimal. In this way, any character
419 can be used in a name value, even quotes themselves.</li>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000420
Reid Spencer2c452282007-08-07 14:34:28 +0000421 <li>Unnamed values are represented as an unsigned numeric value with their
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000422 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000423
Reid Spencercc16dc32004-12-09 18:02:53 +0000424 <li>Constants, which are described in a <a href="#constants">section about
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000425 constants</a>, below.</li>
Misha Brukman9d0919f2003-11-08 01:05:38 +0000426</ol>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000427
Reid Spencer2c452282007-08-07 14:34:28 +0000428<p>LLVM requires that values start with a prefix for two reasons: Compilers
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000429 don't need to worry about name clashes with reserved words, and the set of
430 reserved words may be expanded in the future without penalty. Additionally,
431 unnamed identifiers allow a compiler to quickly come up with a temporary
432 variable without having to avoid symbol table conflicts.</p>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000433
Chris Lattner261efe92003-11-25 01:02:51 +0000434<p>Reserved words in LLVM are very similar to reserved words in other
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000435 languages. There are keywords for different opcodes
436 ('<tt><a href="#i_add">add</a></tt>',
437 '<tt><a href="#i_bitcast">bitcast</a></tt>',
438 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
439 ('<tt><a href="#t_void">void</a></tt>',
440 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
441 reserved words cannot conflict with variable names, because none of them
442 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000443
444<p>Here is an example of LLVM code to multiply the integer variable
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000445 '<tt>%X</tt>' by 8:</p>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000446
Misha Brukman9d0919f2003-11-08 01:05:38 +0000447<p>The easy way:</p>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000448
Benjamin Kramer26fe25f2010-07-13 12:26:09 +0000449<pre class="doc_code">
Bill Wendling2f7a8b02007-05-29 09:04:49 +0000450%result = <a href="#i_mul">mul</a> i32 %X, 8
Chris Lattnere5d947b2004-12-09 16:36:40 +0000451</pre>
452
Misha Brukman9d0919f2003-11-08 01:05:38 +0000453<p>After strength reduction:</p>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000454
Benjamin Kramer26fe25f2010-07-13 12:26:09 +0000455<pre class="doc_code">
Bill Wendling2f7a8b02007-05-29 09:04:49 +0000456%result = <a href="#i_shl">shl</a> i32 %X, i8 3
Chris Lattnere5d947b2004-12-09 16:36:40 +0000457</pre>
458
Misha Brukman9d0919f2003-11-08 01:05:38 +0000459<p>And the hard way:</p>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000460
Benjamin Kramer26fe25f2010-07-13 12:26:09 +0000461<pre class="doc_code">
Gabor Greifec58f752009-10-28 13:05:07 +0000462%0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
463%1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
Bill Wendling2f7a8b02007-05-29 09:04:49 +0000464%result = <a href="#i_add">add</a> i32 %1, %1
Chris Lattnere5d947b2004-12-09 16:36:40 +0000465</pre>
466
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000467<p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
468 lexical features of LLVM:</p>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000469
Chris Lattner00950542001-06-06 20:29:01 +0000470<ol>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000471 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000472 line.</li>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000473
474 <li>Unnamed temporaries are created when the result of a computation is not
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000475 assigned to a named value.</li>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000476
Misha Brukman9d0919f2003-11-08 01:05:38 +0000477 <li>Unnamed temporaries are numbered sequentially</li>
478</ol>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000479
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000480<p>It also shows a convention that we follow in this document. When
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000481 demonstrating instructions, we will follow an instruction with a comment that
482 defines the type and name of value produced. Comments are shown in italic
483 text.</p>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000484
Misha Brukman9d0919f2003-11-08 01:05:38 +0000485</div>
Chris Lattnerfa730212004-12-09 16:11:40 +0000486
487<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000488<h2><a name="highlevel">High Level Structure</a></h2>
Chris Lattnerfa730212004-12-09 16:11:40 +0000489<!-- *********************************************************************** -->
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000490<div>
Chris Lattnerfa730212004-12-09 16:11:40 +0000491<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000492<h3>
493 <a name="modulestructure">Module Structure</a>
494</h3>
Chris Lattnerfa730212004-12-09 16:11:40 +0000495
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000496<div>
Chris Lattnerfa730212004-12-09 16:11:40 +0000497
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000498<p>LLVM programs are composed of "Module"s, each of which is a translation unit
499 of the input programs. Each module consists of functions, global variables,
500 and symbol table entries. Modules may be combined together with the LLVM
501 linker, which merges function (and global variable) definitions, resolves
502 forward declarations, and merges symbol table entries. Here is an example of
503 the "hello world" module:</p>
Chris Lattnerfa730212004-12-09 16:11:40 +0000504
Benjamin Kramer26fe25f2010-07-13 12:26:09 +0000505<pre class="doc_code">
Chris Lattner63e4ccb2010-08-17 17:13:42 +0000506<i>; Declare the string constant as a global constant.</i>&nbsp;
Nick Lewyckydb9cd762011-01-29 01:09:53 +0000507<a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a>&nbsp;<a href="#globalvars">constant</a>&nbsp;<a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>&nbsp;
Chris Lattnerfa730212004-12-09 16:11:40 +0000508
Chris Lattner63e4ccb2010-08-17 17:13:42 +0000509<i>; External declaration of the puts function</i>&nbsp;
510<a href="#functionstructure">declare</a> i32 @puts(i8*) <i>; i32 (i8*)* </i>&nbsp;
Chris Lattnerfa730212004-12-09 16:11:40 +0000511
512<i>; Definition of main function</i>
Chris Lattner63e4ccb2010-08-17 17:13:42 +0000513define i32 @main() { <i>; i32()* </i>&nbsp;
514 <i>; Convert [13 x i8]* to i8 *...</i>&nbsp;
515 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8*</i>&nbsp;
Chris Lattnerfa730212004-12-09 16:11:40 +0000516
Chris Lattner63e4ccb2010-08-17 17:13:42 +0000517 <i>; Call puts function to write out the string to stdout.</i>&nbsp;
518 <a href="#i_call">call</a> i32 @puts(i8* %cast210) <i>; i32</i>&nbsp;
519 <a href="#i_ret">ret</a> i32 0&nbsp;
520}
Devang Patelcd1fd252010-01-11 19:35:55 +0000521
522<i>; Named metadata</i>
523!1 = metadata !{i32 41}
524!foo = !{!1, null}
Bill Wendling2f7a8b02007-05-29 09:04:49 +0000525</pre>
Chris Lattnerfa730212004-12-09 16:11:40 +0000526
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000527<p>This example is made up of a <a href="#globalvars">global variable</a> named
Devang Patelcd1fd252010-01-11 19:35:55 +0000528 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000529 a <a href="#functionstructure">function definition</a> for
Devang Patelcd1fd252010-01-11 19:35:55 +0000530 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
531 "<tt>foo"</tt>.</p>
Chris Lattnerfa730212004-12-09 16:11:40 +0000532
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000533<p>In general, a module is made up of a list of global values, where both
534 functions and global variables are global values. Global values are
535 represented by a pointer to a memory location (in this case, a pointer to an
536 array of char, and a pointer to a function), and have one of the
537 following <a href="#linkage">linkage types</a>.</p>
Chris Lattnerfa730212004-12-09 16:11:40 +0000538
Chris Lattnere5d947b2004-12-09 16:36:40 +0000539</div>
540
541<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000542<h3>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000543 <a name="linkage">Linkage Types</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000544</h3>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000545
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000546<div>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000547
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000548<p>All Global Variables and Functions have one of the following types of
549 linkage:</p>
Chris Lattnerfa730212004-12-09 16:11:40 +0000550
551<dl>
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000552 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
Bill Wendling5e721d72010-07-01 21:55:59 +0000553 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
554 by objects in the current module. In particular, linking code into a
555 module with an private global value may cause the private to be renamed as
556 necessary to avoid collisions. Because the symbol is private to the
557 module, all references can be updated. This doesn't show up in any symbol
558 table in the object file.</dd>
Rafael Espindolabb46f522009-01-15 20:18:42 +0000559
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000560 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
Bill Wendling5e721d72010-07-01 21:55:59 +0000561 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
562 assembler and evaluated by the linker. Unlike normal strong symbols, they
563 are removed by the linker from the final linked image (executable or
564 dynamic library).</dd>
565
566 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
567 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
568 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
569 linker. The symbols are removed by the linker from the final linked image
570 (executable or dynamic library).</dd>
Bill Wendling3d10a5a2009-07-20 01:03:30 +0000571
Bill Wendling55ae5152010-08-20 22:05:50 +0000572 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
573 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
574 of the object is not taken. For instance, functions that had an inline
575 definition, but the compiler decided not to inline it. Note,
576 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
577 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
578 visibility. The symbols are removed by the linker from the final linked
579 image (executable or dynamic library).</dd>
580
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000581 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
Bill Wendling07d31772010-06-29 22:34:52 +0000582 <dd>Similar to private, but the value shows as a local symbol
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000583 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
584 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000585
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000586 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
Chris Lattner266c7bb2009-04-13 05:44:34 +0000587 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000588 into the object file corresponding to the LLVM module. They exist to
589 allow inlining and other optimizations to take place given knowledge of
590 the definition of the global, which is known to be somewhere outside the
591 module. Globals with <tt>available_externally</tt> linkage are allowed to
592 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
593 This linkage type is only allowed on definitions, not declarations.</dd>
Chris Lattner266c7bb2009-04-13 05:44:34 +0000594
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000595 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
Chris Lattner4887bd82007-01-14 06:51:48 +0000596 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
Chris Lattner873187c2010-01-09 19:15:14 +0000597 the same name when linkage occurs. This can be used to implement
598 some forms of inline functions, templates, or other code which must be
599 generated in each translation unit that uses it, but where the body may
600 be overridden with a more definitive definition later. Unreferenced
601 <tt>linkonce</tt> globals are allowed to be discarded. Note that
602 <tt>linkonce</tt> linkage does not actually allow the optimizer to
603 inline the body of this function into callers because it doesn't know if
604 this definition of the function is the definitive definition within the
605 program or whether it will be overridden by a stronger definition.
606 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
607 linkage.</dd>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000608
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000609 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
Chris Lattner26d054d2009-08-05 05:21:07 +0000610 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
611 <tt>linkonce</tt> linkage, except that unreferenced globals with
612 <tt>weak</tt> linkage may not be discarded. This is used for globals that
613 are declared "weak" in C source code.</dd>
614
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000615 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
Chris Lattner26d054d2009-08-05 05:21:07 +0000616 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
617 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
618 global scope.
619 Symbols with "<tt>common</tt>" linkage are merged in the same way as
620 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
Chris Lattnercd81f5d2009-08-05 05:41:44 +0000621 <tt>common</tt> symbols may not have an explicit section,
Eric Christopher6c7e8a02009-12-05 02:46:03 +0000622 must have a zero initializer, and may not be marked '<a
Chris Lattnercd81f5d2009-08-05 05:41:44 +0000623 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
624 have common linkage.</dd>
Chris Lattner26d054d2009-08-05 05:21:07 +0000625
Chris Lattnere5d947b2004-12-09 16:36:40 +0000626
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000627 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000628 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000629 pointer to array type. When two global variables with appending linkage
630 are linked together, the two global arrays are appended together. This is
631 the LLVM, typesafe, equivalent of having the system linker append together
632 "sections" with identical names when .o files are linked.</dd>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000633
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000634 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000635 <dd>The semantics of this linkage follow the ELF object file model: the symbol
636 is weak until linked, if not linked, the symbol becomes null instead of
637 being an undefined reference.</dd>
Anton Korobeynikov7f705592007-01-12 19:20:47 +0000638
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000639 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
640 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000641 <dd>Some languages allow differing globals to be merged, such as two functions
642 with different semantics. Other languages, such as <tt>C++</tt>, ensure
Bill Wendling5e721d72010-07-01 21:55:59 +0000643 that only equivalent globals are ever merged (the "one definition rule"
644 &mdash; "ODR"). Such languages can use the <tt>linkonce_odr</tt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000645 and <tt>weak_odr</tt> linkage types to indicate that the global will only
646 be merged with equivalent globals. These linkage types are otherwise the
647 same as their non-<tt>odr</tt> versions.</dd>
Duncan Sands667d4b82009-03-07 15:45:40 +0000648
Chris Lattnerfa730212004-12-09 16:11:40 +0000649 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
Chris Lattnere5d947b2004-12-09 16:36:40 +0000650 <dd>If none of the above identifiers are used, the global is externally
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000651 visible, meaning that it participates in linkage and can be used to
652 resolve external symbol references.</dd>
Reid Spencerc8910842007-04-11 23:49:50 +0000653</dl>
Anton Korobeynikovb74ed072006-09-14 18:23:27 +0000654
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000655<p>The next two types of linkage are targeted for Microsoft Windows platform
656 only. They are designed to support importing (exporting) symbols from (to)
657 DLLs (Dynamic Link Libraries).</p>
Anton Korobeynikovb74ed072006-09-14 18:23:27 +0000658
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000659<dl>
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000660 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
Anton Korobeynikovb74ed072006-09-14 18:23:27 +0000661 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000662 or variable via a global pointer to a pointer that is set up by the DLL
663 exporting the symbol. On Microsoft Windows targets, the pointer name is
664 formed by combining <code>__imp_</code> and the function or variable
665 name.</dd>
Anton Korobeynikovb74ed072006-09-14 18:23:27 +0000666
Bill Wendlingf82d40a2009-11-02 00:24:16 +0000667 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
Anton Korobeynikovb74ed072006-09-14 18:23:27 +0000668 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000669 pointer to a pointer in a DLL, so that it can be referenced with the
670 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
671 name is formed by combining <code>__imp_</code> and the function or
672 variable name.</dd>
Chris Lattnerfa730212004-12-09 16:11:40 +0000673</dl>
674
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000675<p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
676 another module defined a "<tt>.LC0</tt>" variable and was linked with this
677 one, one of the two would be renamed, preventing a collision. Since
678 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
679 declarations), they are accessible outside of the current module.</p>
680
681<p>It is illegal for a function <i>declaration</i> to have any linkage type
682 other than "externally visible", <tt>dllimport</tt>
683 or <tt>extern_weak</tt>.</p>
684
Duncan Sands667d4b82009-03-07 15:45:40 +0000685<p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000686 or <tt>weak_odr</tt> linkages.</p>
687
Chris Lattnerfa730212004-12-09 16:11:40 +0000688</div>
689
690<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000691<h3>
Chris Lattnerbad10ee2005-05-06 22:57:40 +0000692 <a name="callingconv">Calling Conventions</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000693</h3>
Chris Lattnerbad10ee2005-05-06 22:57:40 +0000694
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000695<div>
Chris Lattnerbad10ee2005-05-06 22:57:40 +0000696
697<p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000698 and <a href="#i_invoke">invokes</a> can all have an optional calling
699 convention specified for the call. The calling convention of any pair of
700 dynamic caller/callee must match, or the behavior of the program is
701 undefined. The following calling conventions are supported by LLVM, and more
702 may be added in the future:</p>
Chris Lattnerbad10ee2005-05-06 22:57:40 +0000703
704<dl>
705 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
Chris Lattnerbad10ee2005-05-06 22:57:40 +0000706 <dd>This calling convention (the default if no other calling convention is
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000707 specified) matches the target C calling conventions. This calling
708 convention supports varargs function calls and tolerates some mismatch in
709 the declared prototype and implemented declaration of the function (as
710 does normal C).</dd>
Chris Lattnerbad10ee2005-05-06 22:57:40 +0000711
712 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
Chris Lattnerbad10ee2005-05-06 22:57:40 +0000713 <dd>This calling convention attempts to make calls as fast as possible
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000714 (e.g. by passing things in registers). This calling convention allows the
715 target to use whatever tricks it wants to produce fast code for the
716 target, without having to conform to an externally specified ABI
Jeffrey Yasskin95fa80a2010-01-09 19:44:16 +0000717 (Application Binary Interface).
718 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
Chris Lattner29689432010-03-11 00:22:57 +0000719 when this or the GHC convention is used.</a> This calling convention
720 does not support varargs and requires the prototype of all callees to
721 exactly match the prototype of the function definition.</dd>
Chris Lattnerbad10ee2005-05-06 22:57:40 +0000722
723 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
Chris Lattnerbad10ee2005-05-06 22:57:40 +0000724 <dd>This calling convention attempts to make code in the caller as efficient
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000725 as possible under the assumption that the call is not commonly executed.
726 As such, these calls often preserve all registers so that the call does
727 not break any live ranges in the caller side. This calling convention
728 does not support varargs and requires the prototype of all callees to
729 exactly match the prototype of the function definition.</dd>
Chris Lattnerbad10ee2005-05-06 22:57:40 +0000730
Chris Lattner29689432010-03-11 00:22:57 +0000731 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
732 <dd>This calling convention has been implemented specifically for use by the
733 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
734 It passes everything in registers, going to extremes to achieve this by
735 disabling callee save registers. This calling convention should not be
736 used lightly but only for specific situations such as an alternative to
737 the <em>register pinning</em> performance technique often used when
738 implementing functional programming languages.At the moment only X86
739 supports this convention and it has the following limitations:
740 <ul>
741 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
742 floating point types are supported.</li>
743 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
744 6 floating point parameters.</li>
745 </ul>
746 This calling convention supports
747 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
748 requires both the caller and callee are using it.
749 </dd>
750
Chris Lattnercfe6b372005-05-07 01:46:40 +0000751 <dt><b>"<tt>cc &lt;<em>n</em>&gt;</tt>" - Numbered convention</b>:</dt>
Chris Lattnerbad10ee2005-05-06 22:57:40 +0000752 <dd>Any calling convention may be specified by number, allowing
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000753 target-specific calling conventions to be used. Target specific calling
754 conventions start at 64.</dd>
Chris Lattnercfe6b372005-05-07 01:46:40 +0000755</dl>
Chris Lattnerbad10ee2005-05-06 22:57:40 +0000756
757<p>More calling conventions can be added/defined on an as-needed basis, to
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000758 support Pascal conventions or any other well-known target-independent
759 convention.</p>
Chris Lattnerbad10ee2005-05-06 22:57:40 +0000760
761</div>
762
763<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000764<h3>
Anton Korobeynikov8cea37b2007-01-23 12:35:46 +0000765 <a name="visibility">Visibility Styles</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000766</h3>
Anton Korobeynikov8cea37b2007-01-23 12:35:46 +0000767
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000768<div>
Anton Korobeynikov8cea37b2007-01-23 12:35:46 +0000769
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000770<p>All Global Variables and Functions have one of the following visibility
771 styles:</p>
Anton Korobeynikov8cea37b2007-01-23 12:35:46 +0000772
773<dl>
774 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
Chris Lattnerd3eda892008-08-05 18:29:16 +0000775 <dd>On targets that use the ELF object file format, default visibility means
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000776 that the declaration is visible to other modules and, in shared libraries,
777 means that the declared entity may be overridden. On Darwin, default
778 visibility means that the declaration is visible to other modules. Default
779 visibility corresponds to "external linkage" in the language.</dd>
Anton Korobeynikov8cea37b2007-01-23 12:35:46 +0000780
781 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
Anton Korobeynikov8cea37b2007-01-23 12:35:46 +0000782 <dd>Two declarations of an object with hidden visibility refer to the same
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000783 object if they are in the same shared object. Usually, hidden visibility
784 indicates that the symbol will not be placed into the dynamic symbol
785 table, so no other module (executable or shared library) can reference it
786 directly.</dd>
Anton Korobeynikov8cea37b2007-01-23 12:35:46 +0000787
Anton Korobeynikov6f9896f2007-04-29 18:35:00 +0000788 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
Anton Korobeynikov6f9896f2007-04-29 18:35:00 +0000789 <dd>On ELF, protected visibility indicates that the symbol will be placed in
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000790 the dynamic symbol table, but that references within the defining module
791 will bind to the local symbol. That is, the symbol cannot be overridden by
792 another module.</dd>
Anton Korobeynikov8cea37b2007-01-23 12:35:46 +0000793</dl>
794
795</div>
796
797<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000798<h3>
Chris Lattnere7886e42009-01-11 20:53:49 +0000799 <a name="namedtypes">Named Types</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000800</h3>
Chris Lattnere7886e42009-01-11 20:53:49 +0000801
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000802<div>
Chris Lattnere7886e42009-01-11 20:53:49 +0000803
804<p>LLVM IR allows you to specify name aliases for certain types. This can make
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000805 it easier to read the IR and make the IR more condensed (particularly when
806 recursive types are involved). An example of a name specification is:</p>
Chris Lattnere7886e42009-01-11 20:53:49 +0000807
Benjamin Kramer26fe25f2010-07-13 12:26:09 +0000808<pre class="doc_code">
Chris Lattnere7886e42009-01-11 20:53:49 +0000809%mytype = type { %mytype*, i32 }
810</pre>
Chris Lattnere7886e42009-01-11 20:53:49 +0000811
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000812<p>You may give a name to any <a href="#typesystem">type</a> except
Chris Lattnerdc65f222010-08-17 23:26:04 +0000813 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000814 is expected with the syntax "%mytype".</p>
Chris Lattnere7886e42009-01-11 20:53:49 +0000815
816<p>Note that type names are aliases for the structural type that they indicate,
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000817 and that you can therefore specify multiple names for the same type. This
818 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
819 uses structural typing, the name is not part of the type. When printing out
820 LLVM IR, the printer will pick <em>one name</em> to render all types of a
821 particular shape. This means that if you have code where two different
822 source types end up having the same LLVM type, that the dumper will sometimes
823 print the "wrong" or unexpected type. This is an important design point and
824 isn't going to change.</p>
Chris Lattnere7886e42009-01-11 20:53:49 +0000825
826</div>
827
Chris Lattnere7886e42009-01-11 20:53:49 +0000828<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000829<h3>
Chris Lattnerfa730212004-12-09 16:11:40 +0000830 <a name="globalvars">Global Variables</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000831</h3>
Chris Lattnerfa730212004-12-09 16:11:40 +0000832
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000833<div>
Chris Lattnerfa730212004-12-09 16:11:40 +0000834
Chris Lattner3689a342005-02-12 19:30:21 +0000835<p>Global variables define regions of memory allocated at compilation time
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000836 instead of run-time. Global variables may optionally be initialized, may
837 have an explicit section to be placed in, and may have an optional explicit
838 alignment specified. A variable may be defined as "thread_local", which
839 means that it will not be shared by threads (each thread will have a
840 separated copy of the variable). A variable may be defined as a global
841 "constant," which indicates that the contents of the variable
842 will <b>never</b> be modified (enabling better optimization, allowing the
843 global data to be placed in the read-only section of an executable, etc).
844 Note that variables that need runtime initialization cannot be marked
845 "constant" as there is a store to the variable.</p>
Chris Lattner3689a342005-02-12 19:30:21 +0000846
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000847<p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
848 constant, even if the final definition of the global is not. This capability
849 can be used to enable slightly better optimization of the program, but
850 requires the language definition to guarantee that optimizations based on the
851 'constantness' are valid for the translation units that do not include the
852 definition.</p>
Chris Lattnerfa730212004-12-09 16:11:40 +0000853
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000854<p>As SSA values, global variables define pointer values that are in scope
855 (i.e. they dominate) all basic blocks in the program. Global variables
856 always define a pointer to their "content" type because they describe a
857 region of memory, and all memory objects in LLVM are accessed through
858 pointers.</p>
Chris Lattnerfa730212004-12-09 16:11:40 +0000859
Rafael Espindolabea46262011-01-08 16:42:36 +0000860<p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
861 that the address is not significant, only the content. Constants marked
Rafael Espindolaa5eaa862011-01-15 08:20:57 +0000862 like this can be merged with other constants if they have the same
863 initializer. Note that a constant with significant address <em>can</em>
864 be merged with a <tt>unnamed_addr</tt> constant, the result being a
865 constant whose address is significant.</p>
Rafael Espindolabea46262011-01-08 16:42:36 +0000866
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000867<p>A global variable may be declared to reside in a target-specific numbered
868 address space. For targets that support them, address spaces may affect how
869 optimizations are performed and/or what target instructions are used to
870 access the variable. The default address space is zero. The address space
871 qualifier must precede any other attributes.</p>
Christopher Lamb284d9922007-12-11 09:31:00 +0000872
Chris Lattner88f6c462005-11-12 00:45:07 +0000873<p>LLVM allows an explicit section to be specified for globals. If the target
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000874 supports it, it will emit globals to the section specified.</p>
Chris Lattner88f6c462005-11-12 00:45:07 +0000875
Chris Lattnerce99fa92010-04-28 00:13:42 +0000876<p>An explicit alignment may be specified for a global, which must be a power
877 of 2. If not present, or if the alignment is set to zero, the alignment of
878 the global is set by the target to whatever it feels convenient. If an
879 explicit alignment is specified, the global is forced to have exactly that
Chris Lattner2d4b8ee2010-04-28 00:31:12 +0000880 alignment. Targets and optimizers are not allowed to over-align the global
881 if the global has an assigned section. In this case, the extra alignment
882 could be observable: for example, code could assume that the globals are
883 densely packed in their section and try to iterate over them as an array,
884 alignment padding would break this iteration.</p>
Chris Lattner2cbdc452005-11-06 08:02:57 +0000885
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000886<p>For example, the following defines a global in a numbered address space with
887 an initializer, section, and alignment:</p>
Chris Lattner68027ea2007-01-14 00:27:09 +0000888
Benjamin Kramer26fe25f2010-07-13 12:26:09 +0000889<pre class="doc_code">
Dan Gohman398873c2009-01-11 00:40:00 +0000890@G = addrspace(5) constant float 1.0, section "foo", align 4
Chris Lattner68027ea2007-01-14 00:27:09 +0000891</pre>
892
Chris Lattnerfa730212004-12-09 16:11:40 +0000893</div>
894
895
896<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000897<h3>
Chris Lattnerfa730212004-12-09 16:11:40 +0000898 <a name="functionstructure">Functions</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000899</h3>
Chris Lattnerfa730212004-12-09 16:11:40 +0000900
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000901<div>
Chris Lattnerfa730212004-12-09 16:11:40 +0000902
Dan Gohmanb55a1ee2010-03-01 17:41:39 +0000903<p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000904 optional <a href="#linkage">linkage type</a>, an optional
905 <a href="#visibility">visibility style</a>, an optional
Rafael Espindolabea46262011-01-08 16:42:36 +0000906 <a href="#callingconv">calling convention</a>,
907 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000908 <a href="#paramattrs">parameter attribute</a> for the return type, a function
909 name, a (possibly empty) argument list (each with optional
910 <a href="#paramattrs">parameter attributes</a>), optional
911 <a href="#fnattrs">function attributes</a>, an optional section, an optional
912 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
913 curly brace, a list of basic blocks, and a closing curly brace.</p>
Anton Korobeynikov8cea37b2007-01-23 12:35:46 +0000914
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000915<p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
916 optional <a href="#linkage">linkage type</a>, an optional
Eric Christopher6c7e8a02009-12-05 02:46:03 +0000917 <a href="#visibility">visibility style</a>, an optional
Rafael Espindolabea46262011-01-08 16:42:36 +0000918 <a href="#callingconv">calling convention</a>,
919 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000920 <a href="#paramattrs">parameter attribute</a> for the return type, a function
921 name, a possibly empty list of arguments, an optional alignment, and an
922 optional <a href="#gc">garbage collector name</a>.</p>
Chris Lattnerfa730212004-12-09 16:11:40 +0000923
Chris Lattnerd3eda892008-08-05 18:29:16 +0000924<p>A function definition contains a list of basic blocks, forming the CFG
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000925 (Control Flow Graph) for the function. Each basic block may optionally start
926 with a label (giving the basic block a symbol table entry), contains a list
927 of instructions, and ends with a <a href="#terminators">terminator</a>
928 instruction (such as a branch or function return).</p>
Chris Lattnerfa730212004-12-09 16:11:40 +0000929
Chris Lattner4a3c9012007-06-08 16:52:14 +0000930<p>The first basic block in a function is special in two ways: it is immediately
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000931 executed on entrance to the function, and it is not allowed to have
932 predecessor basic blocks (i.e. there can not be any branches to the entry
933 block of a function). Because the block can have no predecessors, it also
934 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
Chris Lattnerfa730212004-12-09 16:11:40 +0000935
Chris Lattner88f6c462005-11-12 00:45:07 +0000936<p>LLVM allows an explicit section to be specified for functions. If the target
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000937 supports it, it will emit functions to the section specified.</p>
Chris Lattner88f6c462005-11-12 00:45:07 +0000938
Chris Lattner2cbdc452005-11-06 08:02:57 +0000939<p>An explicit alignment may be specified for a function. If not present, or if
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000940 the alignment is set to zero, the alignment of the function is set by the
941 target to whatever it feels convenient. If an explicit alignment is
942 specified, the function is forced to have at least that much alignment. All
943 alignments must be a power of 2.</p>
Chris Lattner2cbdc452005-11-06 08:02:57 +0000944
Rafael Espindolabea46262011-01-08 16:42:36 +0000945<p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
946 be significant and two identical functions can be merged</p>.
947
Bill Wendlingc39e3e02009-07-20 02:39:26 +0000948<h5>Syntax:</h5>
Benjamin Kramer26fe25f2010-07-13 12:26:09 +0000949<pre class="doc_code">
Chris Lattner50ad45c2008-10-13 16:55:18 +0000950define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000951 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
952 &lt;ResultType&gt; @&lt;FunctionName&gt; ([argument list])
953 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
954 [<a href="#gc">gc</a>] { ... }
955</pre>
Devang Patel307e8ab2008-10-07 17:48:33 +0000956
Chris Lattnerfa730212004-12-09 16:11:40 +0000957</div>
958
Anton Korobeynikov8b0a8c82007-04-25 14:27:10 +0000959<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000960<h3>
Anton Korobeynikov8b0a8c82007-04-25 14:27:10 +0000961 <a name="aliasstructure">Aliases</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000962</h3>
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000963
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000964<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +0000965
966<p>Aliases act as "second name" for the aliasee value (which can be either
967 function, global variable, another alias or bitcast of global value). Aliases
968 may have an optional <a href="#linkage">linkage type</a>, and an
969 optional <a href="#visibility">visibility style</a>.</p>
Anton Korobeynikov8b0a8c82007-04-25 14:27:10 +0000970
Bill Wendlingc39e3e02009-07-20 02:39:26 +0000971<h5>Syntax:</h5>
Benjamin Kramer26fe25f2010-07-13 12:26:09 +0000972<pre class="doc_code">
Duncan Sands0b23ac12008-09-12 20:48:21 +0000973@&lt;Name&gt; = alias [Linkage] [Visibility] &lt;AliaseeTy&gt; @&lt;Aliasee&gt;
Bill Wendlingaac388b2007-05-29 09:42:13 +0000974</pre>
Anton Korobeynikov8b0a8c82007-04-25 14:27:10 +0000975
976</div>
977
Chris Lattner4e9aba72006-01-23 23:23:47 +0000978<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000979<h3>
Devang Patelcd1fd252010-01-11 19:35:55 +0000980 <a name="namedmetadatastructure">Named Metadata</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +0000981</h3>
Devang Patelcd1fd252010-01-11 19:35:55 +0000982
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +0000983<div>
Devang Patelcd1fd252010-01-11 19:35:55 +0000984
Chris Lattnere6a5ddd2010-01-15 21:50:19 +0000985<p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
Dan Gohman872814a2010-07-21 18:54:18 +0000986 nodes</a> (but not metadata strings) are the only valid operands for
Chris Lattnere6a5ddd2010-01-15 21:50:19 +0000987 a named metadata.</p>
Devang Patelcd1fd252010-01-11 19:35:55 +0000988
989<h5>Syntax:</h5>
Benjamin Kramer26fe25f2010-07-13 12:26:09 +0000990<pre class="doc_code">
Dan Gohman872814a2010-07-21 18:54:18 +0000991; Some unnamed metadata nodes, which are referenced by the named metadata.
992!0 = metadata !{metadata !"zero"}
Devang Patelcd1fd252010-01-11 19:35:55 +0000993!1 = metadata !{metadata !"one"}
Dan Gohman872814a2010-07-21 18:54:18 +0000994!2 = metadata !{metadata !"two"}
Dan Gohman1005bc52010-07-13 19:48:13 +0000995; A named metadata.
Dan Gohman872814a2010-07-21 18:54:18 +0000996!name = !{!0, !1, !2}
Devang Patelcd1fd252010-01-11 19:35:55 +0000997</pre>
Devang Patelcd1fd252010-01-11 19:35:55 +0000998
999</div>
1000
1001<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001002<h3>
1003 <a name="paramattrs">Parameter Attributes</a>
1004</h3>
Reid Spencerca86e162006-12-31 07:07:53 +00001005
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001006<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001007
1008<p>The return type and each parameter of a function type may have a set of
1009 <i>parameter attributes</i> associated with them. Parameter attributes are
1010 used to communicate additional information about the result or parameters of
1011 a function. Parameter attributes are considered to be part of the function,
1012 not of the function type, so functions with different parameter attributes
1013 can have the same function type.</p>
1014
1015<p>Parameter attributes are simple keywords that follow the type specified. If
1016 multiple parameter attributes are needed, they are space separated. For
1017 example:</p>
Bill Wendling2f7a8b02007-05-29 09:04:49 +00001018
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00001019<pre class="doc_code">
Nick Lewyckyb6a7d252009-02-15 23:06:14 +00001020declare i32 @printf(i8* noalias nocapture, ...)
Chris Lattner66d922c2008-10-04 18:33:34 +00001021declare i32 @atoi(i8 zeroext)
1022declare signext i8 @returns_signed_char()
Bill Wendling2f7a8b02007-05-29 09:04:49 +00001023</pre>
Bill Wendling2f7a8b02007-05-29 09:04:49 +00001024
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001025<p>Note that any attributes for the function result (<tt>nounwind</tt>,
1026 <tt>readonly</tt>) come immediately after the argument list.</p>
Reid Spencerca86e162006-12-31 07:07:53 +00001027
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001028<p>Currently, only the following parameter attributes are defined:</p>
Chris Lattner47507de2008-01-11 06:20:47 +00001029
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001030<dl>
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001031 <dt><tt><b>zeroext</b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001032 <dd>This indicates to the code generator that the parameter or return value
Cameron Zwarichebe81732011-03-16 22:20:18 +00001033 should be zero-extended to the extent required by the target's ABI (which
1034 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1035 parameter) or the callee (for a return value).</dd>
Chris Lattner47507de2008-01-11 06:20:47 +00001036
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001037 <dt><tt><b>signext</b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001038 <dd>This indicates to the code generator that the parameter or return value
Cameron Zwarich9e69ff92011-03-17 14:21:58 +00001039 should be sign-extended to the extent required by the target's ABI (which
1040 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1041 return value).</dd>
Chris Lattner47507de2008-01-11 06:20:47 +00001042
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001043 <dt><tt><b>inreg</b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001044 <dd>This indicates that this parameter or return value should be treated in a
1045 special target-dependent fashion during while emitting code for a function
1046 call or return (usually, by putting it in a register as opposed to memory,
1047 though some targets use it to distinguish between two different kinds of
1048 registers). Use of this attribute is target-specific.</dd>
Chris Lattner47507de2008-01-11 06:20:47 +00001049
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001050 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
Chris Lattnera6fd81d2010-11-20 23:49:06 +00001051 <dd><p>This indicates that the pointer parameter should really be passed by
1052 value to the function. The attribute implies that a hidden copy of the
1053 pointee
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001054 is made between the caller and the callee, so the callee is unable to
1055 modify the value in the callee. This attribute is only valid on LLVM
1056 pointer arguments. It is generally used to pass structs and arrays by
1057 value, but is also valid on pointers to scalars. The copy is considered
1058 to belong to the caller not the callee (for example,
1059 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1060 <tt>byval</tt> parameters). This is not a valid attribute for return
Chris Lattnera6fd81d2010-11-20 23:49:06 +00001061 values.</p>
1062
1063 <p>The byval attribute also supports specifying an alignment with
1064 the align attribute. It indicates the alignment of the stack slot to
1065 form and the known alignment of the pointer specified to the call site. If
1066 the alignment is not specified, then the code generator makes a
1067 target-specific assumption.</p></dd>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001068
Dan Gohmanff235352010-07-02 23:18:08 +00001069 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001070 <dd>This indicates that the pointer parameter specifies the address of a
1071 structure that is the return value of the function in the source program.
1072 This pointer must be guaranteed by the caller to be valid: loads and
1073 stores to the structure may be assumed by the callee to not to trap. This
1074 may only be applied to the first parameter. This is not a valid attribute
1075 for return values. </dd>
1076
Dan Gohmanff235352010-07-02 23:18:08 +00001077 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
Dan Gohman1e109622010-07-02 18:41:32 +00001078 <dd>This indicates that pointer values
1079 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
Dan Gohmanefca7f92010-07-02 23:46:54 +00001080 value do not alias pointer values which are not <i>based</i> on it,
1081 ignoring certain "irrelevant" dependencies.
1082 For a call to the parent function, dependencies between memory
1083 references from before or after the call and from those during the call
1084 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1085 return value used in that call.
Dan Gohman1e109622010-07-02 18:41:32 +00001086 The caller shares the responsibility with the callee for ensuring that
1087 these requirements are met.
1088 For further details, please see the discussion of the NoAlias response in
Dan Gohmanff70fe42010-07-06 15:26:33 +00001089 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1090<br>
John McCall191d4ee2010-07-06 21:07:14 +00001091 Note that this definition of <tt>noalias</tt> is intentionally
1092 similar to the definition of <tt>restrict</tt> in C99 for function
Chris Lattner211244a2010-07-06 20:51:35 +00001093 arguments, though it is slightly weaker.
Dan Gohmanff70fe42010-07-06 15:26:33 +00001094<br>
1095 For function return values, C99's <tt>restrict</tt> is not meaningful,
1096 while LLVM's <tt>noalias</tt> is.
1097 </dd>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001098
Dan Gohmanff235352010-07-02 23:18:08 +00001099 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001100 <dd>This indicates that the callee does not make any copies of the pointer
1101 that outlive the callee itself. This is not a valid attribute for return
1102 values.</dd>
1103
Dan Gohmanff235352010-07-02 23:18:08 +00001104 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001105 <dd>This indicates that the pointer parameter can be excised using the
1106 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1107 attribute for return values.</dd>
1108</dl>
Reid Spencerca86e162006-12-31 07:07:53 +00001109
Reid Spencerca86e162006-12-31 07:07:53 +00001110</div>
1111
1112<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001113<h3>
Gordon Henriksen80a75bf2007-12-10 03:18:06 +00001114 <a name="gc">Garbage Collector Names</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001115</h3>
Gordon Henriksen80a75bf2007-12-10 03:18:06 +00001116
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001117<div>
Gordon Henriksen80a75bf2007-12-10 03:18:06 +00001118
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001119<p>Each function may specify a garbage collector name, which is simply a
1120 string:</p>
1121
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00001122<pre class="doc_code">
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001123define void @f() gc "name" { ... }
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001124</pre>
Gordon Henriksen80a75bf2007-12-10 03:18:06 +00001125
1126<p>The compiler declares the supported values of <i>name</i>. Specifying a
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001127 collector which will cause the compiler to alter its output in order to
1128 support the named garbage collection algorithm.</p>
1129
Gordon Henriksen80a75bf2007-12-10 03:18:06 +00001130</div>
1131
1132<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001133<h3>
Devang Patel2c9c3e72008-09-26 23:51:19 +00001134 <a name="fnattrs">Function Attributes</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001135</h3>
Devang Patelf8b94812008-09-04 23:05:13 +00001136
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001137<div>
Devang Patel2c9c3e72008-09-26 23:51:19 +00001138
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001139<p>Function attributes are set to communicate additional information about a
1140 function. Function attributes are considered to be part of the function, not
1141 of the function type, so functions with different parameter attributes can
1142 have the same function type.</p>
Devang Patel2c9c3e72008-09-26 23:51:19 +00001143
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001144<p>Function attributes are simple keywords that follow the type specified. If
1145 multiple attributes are needed, they are space separated. For example:</p>
Devang Patelf8b94812008-09-04 23:05:13 +00001146
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00001147<pre class="doc_code">
Devang Patel2c9c3e72008-09-26 23:51:19 +00001148define void @f() noinline { ... }
1149define void @f() alwaysinline { ... }
1150define void @f() alwaysinline optsize { ... }
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001151define void @f() optsize { ... }
Bill Wendlinge36dccc2008-09-07 10:26:33 +00001152</pre>
Devang Patelf8b94812008-09-04 23:05:13 +00001153
Bill Wendlinge36dccc2008-09-07 10:26:33 +00001154<dl>
Charles Davis1e063d12010-02-12 00:31:15 +00001155 <dt><tt><b>alignstack(&lt;<em>n</em>&gt;)</b></tt></dt>
1156 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1157 the backend should forcibly align the stack pointer. Specify the
1158 desired alignment, which must be a power of two, in parentheses.
1159
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001160 <dt><tt><b>alwaysinline</b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001161 <dd>This attribute indicates that the inliner should attempt to inline this
1162 function into callers whenever possible, ignoring any active inlining size
1163 threshold for this caller.</dd>
Bill Wendlinge36dccc2008-09-07 10:26:33 +00001164
Charles Davis970bfcc2010-10-25 15:37:09 +00001165 <dt><tt><b>hotpatch</b></tt></dt>
Charles Davis6f12e292010-10-25 16:29:03 +00001166 <dd>This attribute indicates that the function should be 'hotpatchable',
Charles Davis0076d202010-10-25 19:07:39 +00001167 meaning the function can be patched and/or hooked even while it is
1168 loaded into memory. On x86, the function prologue will be preceded
1169 by six bytes of padding and will begin with a two-byte instruction.
1170 Most of the functions in the Windows system DLLs in Windows XP SP2 or
1171 higher were compiled in this fashion.</dd>
Charles Davis970bfcc2010-10-25 15:37:09 +00001172
Dan Gohman129bd562011-06-16 16:03:13 +00001173 <dt><tt><b>nonlazybind</b></tt></dt>
1174 <dd>This attribute suppresses lazy symbol binding for the function. This
1175 may make calls to the function faster, at the cost of extra program
1176 startup time if the function is not called during program startup.</dd>
1177
Jakob Stoklund Olesen570a4a52010-02-06 01:16:28 +00001178 <dt><tt><b>inlinehint</b></tt></dt>
1179 <dd>This attribute indicates that the source code contained a hint that inlining
1180 this function is desirable (such as the "inline" keyword in C/C++). It
1181 is just a hint; it imposes no requirements on the inliner.</dd>
1182
Nick Lewycky76ec37a2010-07-06 18:24:09 +00001183 <dt><tt><b>naked</b></tt></dt>
1184 <dd>This attribute disables prologue / epilogue emission for the function.
1185 This can have very system-specific consequences.</dd>
1186
1187 <dt><tt><b>noimplicitfloat</b></tt></dt>
1188 <dd>This attributes disables implicit floating point instructions.</dd>
1189
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001190 <dt><tt><b>noinline</b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001191 <dd>This attribute indicates that the inliner should never inline this
1192 function in any situation. This attribute may not be used together with
1193 the <tt>alwaysinline</tt> attribute.</dd>
Devang Patel2c9c3e72008-09-26 23:51:19 +00001194
Nick Lewycky76ec37a2010-07-06 18:24:09 +00001195 <dt><tt><b>noredzone</b></tt></dt>
1196 <dd>This attribute indicates that the code generator should not use a red
1197 zone, even if the target-specific ABI normally permits it.</dd>
Devang Patel2c9c3e72008-09-26 23:51:19 +00001198
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001199 <dt><tt><b>noreturn</b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001200 <dd>This function attribute indicates that the function never returns
1201 normally. This produces undefined behavior at runtime if the function
1202 ever does dynamically return.</dd>
Bill Wendling31359ba2008-11-13 01:02:51 +00001203
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001204 <dt><tt><b>nounwind</b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001205 <dd>This function attribute indicates that the function never returns with an
1206 unwind or exceptional control flow. If the function does unwind, its
1207 runtime behavior is undefined.</dd>
Bill Wendlingfbaa7ed2008-11-26 19:07:40 +00001208
Nick Lewycky76ec37a2010-07-06 18:24:09 +00001209 <dt><tt><b>optsize</b></tt></dt>
1210 <dd>This attribute suggests that optimization passes and code generator passes
1211 make choices that keep the code size of this function low, and otherwise
1212 do optimizations specifically to reduce code size.</dd>
1213
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001214 <dt><tt><b>readnone</b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001215 <dd>This attribute indicates that the function computes its result (or decides
1216 to unwind an exception) based strictly on its arguments, without
1217 dereferencing any pointer arguments or otherwise accessing any mutable
1218 state (e.g. memory, control registers, etc) visible to caller functions.
1219 It does not write through any pointer arguments
1220 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1221 changes any state visible to callers. This means that it cannot unwind
1222 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1223 could use the <tt>unwind</tt> instruction.</dd>
Devang Patel5d96fda2009-06-12 19:45:19 +00001224
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001225 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001226 <dd>This attribute indicates that the function does not write through any
1227 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1228 arguments) or otherwise modify any state (e.g. memory, control registers,
1229 etc) visible to caller functions. It may dereference pointer arguments
1230 and read state that may be set in the caller. A readonly function always
1231 returns the same value (or unwinds an exception identically) when called
1232 with the same set of arguments and global state. It cannot unwind an
1233 exception by calling the <tt>C++</tt> exception throwing methods, but may
1234 use the <tt>unwind</tt> instruction.</dd>
Anton Korobeynikovc5ec8a72009-07-17 18:07:26 +00001235
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001236 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001237 <dd>This attribute indicates that the function should emit a stack smashing
1238 protector. It is in the form of a "canary"&mdash;a random value placed on
1239 the stack before the local variables that's checked upon return from the
1240 function to see if it has been overwritten. A heuristic is used to
1241 determine if a function needs stack protectors or not.<br>
1242<br>
1243 If a function that has an <tt>ssp</tt> attribute is inlined into a
1244 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1245 function will have an <tt>ssp</tt> attribute.</dd>
1246
Bill Wendlingf82d40a2009-11-02 00:24:16 +00001247 <dt><tt><b>sspreq</b></tt></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001248 <dd>This attribute indicates that the function should <em>always</em> emit a
1249 stack smashing protector. This overrides
Bill Wendlingc39e3e02009-07-20 02:39:26 +00001250 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1251<br>
1252 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1253 function that doesn't have an <tt>sspreq</tt> attribute or which has
1254 an <tt>ssp</tt> attribute, then the resulting function will have
1255 an <tt>sspreq</tt> attribute.</dd>
Rafael Espindolafbff0ec2011-07-25 15:27:59 +00001256
1257 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1258 <dd>This attribute indicates that the ABI being targeted requires that
1259 an unwind table entry be produce for this function even if we can
1260 show that no exceptions passes by it. This is normally the case for
1261 the ELF x86-64 abi, but it can be disabled for some compilation
1262 units.</dd>
1263
Bill Wendlinge36dccc2008-09-07 10:26:33 +00001264</dl>
1265
Devang Patelf8b94812008-09-04 23:05:13 +00001266</div>
1267
1268<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001269<h3>
Chris Lattner1eeeb0c2006-04-08 04:40:53 +00001270 <a name="moduleasm">Module-Level Inline Assembly</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001271</h3>
Chris Lattner4e9aba72006-01-23 23:23:47 +00001272
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001273<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001274
1275<p>Modules may contain "module-level inline asm" blocks, which corresponds to
1276 the GCC "file scope inline asm" blocks. These blocks are internally
1277 concatenated by LLVM and treated as a single unit, but may be separated in
1278 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
Chris Lattner4e9aba72006-01-23 23:23:47 +00001279
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00001280<pre class="doc_code">
Bill Wendling2f7a8b02007-05-29 09:04:49 +00001281module asm "inline asm code goes here"
1282module asm "more can go here"
1283</pre>
Chris Lattner4e9aba72006-01-23 23:23:47 +00001284
1285<p>The strings can contain any character by escaping non-printable characters.
1286 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001287 for the number.</p>
Chris Lattner4e9aba72006-01-23 23:23:47 +00001288
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001289<p>The inline asm code is simply printed to the machine code .s file when
1290 assembly code is generated.</p>
1291
Chris Lattner4e9aba72006-01-23 23:23:47 +00001292</div>
Chris Lattnerfa730212004-12-09 16:11:40 +00001293
Reid Spencerde151942007-02-19 23:54:10 +00001294<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001295<h3>
Reid Spencerde151942007-02-19 23:54:10 +00001296 <a name="datalayout">Data Layout</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001297</h3>
Reid Spencerde151942007-02-19 23:54:10 +00001298
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001299<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001300
Reid Spencerde151942007-02-19 23:54:10 +00001301<p>A module may specify a target specific data layout string that specifies how
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001302 data is to be laid out in memory. The syntax for the data layout is
1303 simply:</p>
1304
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00001305<pre class="doc_code">
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001306target datalayout = "<i>layout specification</i>"
1307</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001308
1309<p>The <i>layout specification</i> consists of a list of specifications
1310 separated by the minus sign character ('-'). Each specification starts with
1311 a letter and may include other information after the letter to define some
1312 aspect of the data layout. The specifications accepted are as follows:</p>
1313
Reid Spencerde151942007-02-19 23:54:10 +00001314<dl>
1315 <dt><tt>E</tt></dt>
1316 <dd>Specifies that the target lays out data in big-endian form. That is, the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001317 bits with the most significance have the lowest address location.</dd>
1318
Reid Spencerde151942007-02-19 23:54:10 +00001319 <dt><tt>e</tt></dt>
Chris Lattnerd3eda892008-08-05 18:29:16 +00001320 <dd>Specifies that the target lays out data in little-endian form. That is,
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001321 the bits with the least significance have the lowest address
1322 location.</dd>
1323
Reid Spencerde151942007-02-19 23:54:10 +00001324 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00001325 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001326 <i>preferred</i> alignments. All sizes are in bits. Specifying
1327 the <i>pref</i> alignment is optional. If omitted, the
1328 preceding <tt>:</tt> should be omitted too.</dd>
1329
Reid Spencerde151942007-02-19 23:54:10 +00001330 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1331 <dd>This specifies the alignment for an integer type of a given bit
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001332 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1333
Reid Spencerde151942007-02-19 23:54:10 +00001334 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00001335 <dd>This specifies the alignment for a vector type of a given bit
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001336 <i>size</i>.</dd>
1337
Reid Spencerde151942007-02-19 23:54:10 +00001338 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00001339 <dd>This specifies the alignment for a floating point type of a given bit
Dale Johannesen9d8d2212010-05-28 18:54:47 +00001340 <i>size</i>. Only values of <i>size</i> that are supported by the target
1341 will work. 32 (float) and 64 (double) are supported on all targets;
1342 80 or 128 (different flavors of long double) are also supported on some
1343 targets.
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001344
Reid Spencerde151942007-02-19 23:54:10 +00001345 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1346 <dd>This specifies the alignment for an aggregate type of a given bit
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001347 <i>size</i>.</dd>
1348
Daniel Dunbar87bde0b2009-06-08 22:17:53 +00001349 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1350 <dd>This specifies the alignment for a stack object of a given bit
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001351 <i>size</i>.</dd>
Chris Lattnere82bdc42009-11-07 09:35:34 +00001352
1353 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1354 <dd>This specifies a set of native integer widths for the target CPU
1355 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1356 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
Eric Christopher6c7e8a02009-12-05 02:46:03 +00001357 this set are considered to support most general arithmetic
Chris Lattnere82bdc42009-11-07 09:35:34 +00001358 operations efficiently.</dd>
Reid Spencerde151942007-02-19 23:54:10 +00001359</dl>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001360
Reid Spencerde151942007-02-19 23:54:10 +00001361<p>When constructing the data layout for a given target, LLVM starts with a
Dan Gohman1c70c002010-04-28 00:36:01 +00001362 default set of specifications which are then (possibly) overridden by the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001363 specifications in the <tt>datalayout</tt> keyword. The default specifications
1364 are given in this list:</p>
1365
Reid Spencerde151942007-02-19 23:54:10 +00001366<ul>
1367 <li><tt>E</tt> - big endian</li>
Dan Gohmanfdf2e8c2010-02-23 02:44:03 +00001368 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
Reid Spencerde151942007-02-19 23:54:10 +00001369 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1370 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1371 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1372 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
Chris Lattnerd3eda892008-08-05 18:29:16 +00001373 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
Reid Spencerde151942007-02-19 23:54:10 +00001374 alignment of 64-bits</li>
1375 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1376 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1377 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1378 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1379 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
Daniel Dunbar87bde0b2009-06-08 22:17:53 +00001380 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
Reid Spencerde151942007-02-19 23:54:10 +00001381</ul>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001382
1383<p>When LLVM is determining the alignment for a given type, it uses the
1384 following rules:</p>
1385
Reid Spencerde151942007-02-19 23:54:10 +00001386<ol>
1387 <li>If the type sought is an exact match for one of the specifications, that
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001388 specification is used.</li>
1389
Reid Spencerde151942007-02-19 23:54:10 +00001390 <li>If no match is found, and the type sought is an integer type, then the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001391 smallest integer type that is larger than the bitwidth of the sought type
1392 is used. If none of the specifications are larger than the bitwidth then
1393 the the largest integer type is used. For example, given the default
1394 specifications above, the i7 type will use the alignment of i8 (next
1395 largest) while both i65 and i256 will use the alignment of i64 (largest
1396 specified).</li>
1397
Reid Spencerde151942007-02-19 23:54:10 +00001398 <li>If no match is found, and the type sought is a vector type, then the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001399 largest vector type that is smaller than the sought vector type will be
1400 used as a fall back. This happens because &lt;128 x double&gt; can be
1401 implemented in terms of 64 &lt;2 x double&gt;, for example.</li>
Reid Spencerde151942007-02-19 23:54:10 +00001402</ol>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001403
Reid Spencerde151942007-02-19 23:54:10 +00001404</div>
Chris Lattnerfa730212004-12-09 16:11:40 +00001405
Dan Gohman556ca272009-07-27 18:07:55 +00001406<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001407<h3>
Dan Gohman556ca272009-07-27 18:07:55 +00001408 <a name="pointeraliasing">Pointer Aliasing Rules</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001409</h3>
Dan Gohman556ca272009-07-27 18:07:55 +00001410
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001411<div>
Dan Gohman556ca272009-07-27 18:07:55 +00001412
Andreas Bolka55e459a2009-07-29 00:02:05 +00001413<p>Any memory access must be done through a pointer value associated
Andreas Bolka99a82052009-07-27 20:37:10 +00001414with an address range of the memory access, otherwise the behavior
Dan Gohman556ca272009-07-27 18:07:55 +00001415is undefined. Pointer values are associated with address ranges
1416according to the following rules:</p>
1417
1418<ul>
Dan Gohman1e109622010-07-02 18:41:32 +00001419 <li>A pointer value is associated with the addresses associated with
1420 any value it is <i>based</i> on.
Andreas Bolka55e459a2009-07-29 00:02:05 +00001421 <li>An address of a global variable is associated with the address
Dan Gohman556ca272009-07-27 18:07:55 +00001422 range of the variable's storage.</li>
1423 <li>The result value of an allocation instruction is associated with
1424 the address range of the allocated storage.</li>
1425 <li>A null pointer in the default address-space is associated with
Andreas Bolka55e459a2009-07-29 00:02:05 +00001426 no address.</li>
Dan Gohman556ca272009-07-27 18:07:55 +00001427 <li>An integer constant other than zero or a pointer value returned
1428 from a function not defined within LLVM may be associated with address
1429 ranges allocated through mechanisms other than those provided by
Andreas Bolka55e459a2009-07-29 00:02:05 +00001430 LLVM. Such ranges shall not overlap with any ranges of addresses
Dan Gohman556ca272009-07-27 18:07:55 +00001431 allocated by mechanisms provided by LLVM.</li>
Dan Gohman1e109622010-07-02 18:41:32 +00001432</ul>
1433
1434<p>A pointer value is <i>based</i> on another pointer value according
1435 to the following rules:</p>
1436
1437<ul>
1438 <li>A pointer value formed from a
1439 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1440 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1441 <li>The result value of a
1442 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1443 of the <tt>bitcast</tt>.</li>
1444 <li>A pointer value formed by an
1445 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1446 pointer values that contribute (directly or indirectly) to the
1447 computation of the pointer's value.</li>
1448 <li>The "<i>based</i> on" relationship is transitive.</li>
1449</ul>
1450
1451<p>Note that this definition of <i>"based"</i> is intentionally
1452 similar to the definition of <i>"based"</i> in C99, though it is
1453 slightly weaker.</p>
Dan Gohman556ca272009-07-27 18:07:55 +00001454
1455<p>LLVM IR does not associate types with memory. The result type of a
Andreas Bolka55e459a2009-07-29 00:02:05 +00001456<tt><a href="#i_load">load</a></tt> merely indicates the size and
1457alignment of the memory from which to load, as well as the
Dan Gohmanc22c0f32010-06-17 19:23:50 +00001458interpretation of the value. The first operand type of a
Andreas Bolka55e459a2009-07-29 00:02:05 +00001459<tt><a href="#i_store">store</a></tt> similarly only indicates the size
1460and alignment of the store.</p>
Dan Gohman556ca272009-07-27 18:07:55 +00001461
1462<p>Consequently, type-based alias analysis, aka TBAA, aka
1463<tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1464LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1465additional information which specialized optimization passes may use
1466to implement type-based alias analysis.</p>
1467
1468</div>
1469
Jeffrey Yasskin93e066d2010-04-26 21:21:24 +00001470<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001471<h3>
Jeffrey Yasskin93e066d2010-04-26 21:21:24 +00001472 <a name="volatile">Volatile Memory Accesses</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001473</h3>
Jeffrey Yasskin93e066d2010-04-26 21:21:24 +00001474
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001475<div>
Jeffrey Yasskin93e066d2010-04-26 21:21:24 +00001476
1477<p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1478href="#i_store"><tt>store</tt></a>s, and <a
1479href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1480The optimizers must not change the number of volatile operations or change their
1481order of execution relative to other volatile operations. The optimizers
1482<i>may</i> change the order of volatile operations relative to non-volatile
1483operations. This is not Java's "volatile" and has no cross-thread
1484synchronization behavior.</p>
1485
1486</div>
1487
Eli Friedman5b60e1b2011-07-20 21:35:53 +00001488<!-- ======================================================================= -->
1489<h3>
1490 <a name="memmodel">Memory Model for Concurrent Operations</a>
1491</h3>
1492
1493<div>
1494
1495<p>The LLVM IR does not define any way to start parallel threads of execution
1496or to register signal handlers. Nonetheless, there are platform-specific
1497ways to create them, and we define LLVM IR's behavior in their presence. This
1498model is inspired by the C++0x memory model.</p>
1499
Eli Friedman234bccd2011-08-22 21:35:27 +00001500<p>For a more informal introduction to this model, see the
1501<a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1502
Eli Friedman5b60e1b2011-07-20 21:35:53 +00001503<p>We define a <i>happens-before</i> partial order as the least partial order
1504that</p>
1505<ul>
1506 <li>Is a superset of single-thread program order, and</li>
1507 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1508 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1509 by platform-specific techniques, like pthread locks, thread
Eli Friedmanff030482011-07-28 21:48:00 +00001510 creation, thread joining, etc., and by atomic instructions.
1511 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1512 </li>
Eli Friedman5b60e1b2011-07-20 21:35:53 +00001513</ul>
1514
1515<p>Note that program order does not introduce <i>happens-before</i> edges
1516between a thread and signals executing inside that thread.</p>
1517
1518<p>Every (defined) read operation (load instructions, memcpy, atomic
1519loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1520(defined) write operations (store instructions, atomic
Eli Friedman118973a2011-07-22 03:04:45 +00001521stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1522initialized globals are considered to have a write of the initializer which is
1523atomic and happens before any other read or write of the memory in question.
1524For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1525any write to the same byte, except:</p>
Eli Friedman5b60e1b2011-07-20 21:35:53 +00001526
1527<ul>
1528 <li>If <var>write<sub>1</sub></var> happens before
1529 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1530 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
Eli Friedman118973a2011-07-22 03:04:45 +00001531 does not see <var>write<sub>1</sub></var>.
Bill Wendling0246bb72011-07-31 06:45:03 +00001532 <li>If <var>R<sub>byte</sub></var> happens before
1533 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1534 see <var>write<sub>3</sub></var>.
Eli Friedman5b60e1b2011-07-20 21:35:53 +00001535</ul>
1536
1537<p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1538<ul>
Eli Friedman234bccd2011-08-22 21:35:27 +00001539 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1540 is supposed to give guarantees which can support
1541 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1542 addresses which do not behave like normal memory. It does not generally
1543 provide cross-thread synchronization.)
1544 <li>Otherwise, if there is no write to the same byte that happens before
Eli Friedman5b60e1b2011-07-20 21:35:53 +00001545 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1546 <tt>undef</tt> for that byte.
Eli Friedman118973a2011-07-22 03:04:45 +00001547 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
Eli Friedman5b60e1b2011-07-20 21:35:53 +00001548 <var>R<sub>byte</sub></var> returns the value written by that
1549 write.</li>
Eli Friedman118973a2011-07-22 03:04:45 +00001550 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1551 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
Eli Friedmanff030482011-07-28 21:48:00 +00001552 values written. See the <a href="#ordering">Atomic Memory Ordering
1553 Constraints</a> section for additional constraints on how the choice
1554 is made.
Eli Friedman5b60e1b2011-07-20 21:35:53 +00001555 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1556</ul>
1557
1558<p><var>R</var> returns the value composed of the series of bytes it read.
1559This implies that some bytes within the value may be <tt>undef</tt>
1560<b>without</b> the entire value being <tt>undef</tt>. Note that this only
1561defines the semantics of the operation; it doesn't mean that targets will
1562emit more than one instruction to read the series of bytes.</p>
1563
1564<p>Note that in cases where none of the atomic intrinsics are used, this model
1565places only one restriction on IR transformations on top of what is required
1566for single-threaded execution: introducing a store to a byte which might not
Eli Friedman101c81d2011-08-02 01:15:34 +00001567otherwise be stored is not allowed in general. (Specifically, in the case
1568where another thread might write to and read from an address, introducing a
1569store can change a load that may see exactly one write into a load that may
1570see multiple writes.)</p>
Eli Friedman5b60e1b2011-07-20 21:35:53 +00001571
1572<!-- FIXME: This model assumes all targets where concurrency is relevant have
1573a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1574none of the backends currently in the tree fall into this category; however,
1575there might be targets which care. If there are, we want a paragraph
1576like the following:
1577
1578Targets may specify that stores narrower than a certain width are not
1579available; on such a target, for the purposes of this model, treat any
1580non-atomic write with an alignment or width less than the minimum width
1581as if it writes to the relevant surrounding bytes.
1582-->
1583
1584</div>
1585
Eli Friedmanff030482011-07-28 21:48:00 +00001586<!-- ======================================================================= -->
NAKAMURA Takumib5bb29c2011-08-12 06:17:17 +00001587<h3>
Eli Friedmanff030482011-07-28 21:48:00 +00001588 <a name="ordering">Atomic Memory Ordering Constraints</a>
NAKAMURA Takumib5bb29c2011-08-12 06:17:17 +00001589</h3>
Eli Friedmanff030482011-07-28 21:48:00 +00001590
NAKAMURA Takumib5bb29c2011-08-12 06:17:17 +00001591<div>
Eli Friedmanff030482011-07-28 21:48:00 +00001592
1593<p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
Eli Friedman21006d42011-08-09 23:02:53 +00001594<a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1595<a href="#i_fence"><code>fence</code></a>,
1596<a href="#i_load"><code>atomic load</code></a>, and
Eli Friedman8fa281a2011-08-09 23:26:12 +00001597<a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
Eli Friedmanff030482011-07-28 21:48:00 +00001598that determines which other atomic instructions on the same address they
1599<i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1600but are somewhat more colloquial. If these descriptions aren't precise enough,
Eli Friedman234bccd2011-08-22 21:35:27 +00001601check those specs (see spec references in the
1602<a href="Atomic.html#introduction">atomics guide</a>).
1603<a href="#i_fence"><code>fence</code></a> instructions
Eli Friedmanff030482011-07-28 21:48:00 +00001604treat these orderings somewhat differently since they don't take an address.
1605See that instruction's documentation for details.</p>
1606
Eli Friedman234bccd2011-08-22 21:35:27 +00001607<p>For a simpler introduction to the ordering constraints, see the
1608<a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1609
Eli Friedmanff030482011-07-28 21:48:00 +00001610<dl>
Eli Friedmanff030482011-07-28 21:48:00 +00001611<dt><code>unordered</code></dt>
1612<dd>The set of values that can be read is governed by the happens-before
1613partial order. A value cannot be read unless some operation wrote it.
1614This is intended to provide a guarantee strong enough to model Java's
1615non-volatile shared variables. This ordering cannot be specified for
1616read-modify-write operations; it is not strong enough to make them atomic
1617in any interesting way.</dd>
1618<dt><code>monotonic</code></dt>
1619<dd>In addition to the guarantees of <code>unordered</code>, there is a single
1620total order for modifications by <code>monotonic</code> operations on each
1621address. All modification orders must be compatible with the happens-before
1622order. There is no guarantee that the modification orders can be combined to
1623a global total order for the whole program (and this often will not be
1624possible). The read in an atomic read-modify-write operation
1625(<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1626<a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1627reads the value in the modification order immediately before the value it
1628writes. If one atomic read happens before another atomic read of the same
1629address, the later read must see the same value or a later value in the
1630address's modification order. This disallows reordering of
1631<code>monotonic</code> (or stronger) operations on the same address. If an
1632address is written <code>monotonic</code>ally by one thread, and other threads
1633<code>monotonic</code>ally read that address repeatedly, the other threads must
Eli Friedman234bccd2011-08-22 21:35:27 +00001634eventually see the write. This corresponds to the C++0x/C1x
1635<code>memory_order_relaxed</code>.</dd>
Eli Friedmanff030482011-07-28 21:48:00 +00001636<dt><code>acquire</code></dt>
Eli Friedmanff030482011-07-28 21:48:00 +00001637<dd>In addition to the guarantees of <code>monotonic</code>,
Eli Friedmanc264b2f2011-08-24 20:28:39 +00001638a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1639operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1640<dt><code>release</code></dt>
1641<dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1642writes a value which is subsequently read by an <code>acquire</code> operation,
1643it <i>synchronizes-with</i> that operation. (This isn't a complete
1644description; see the C++0x definition of a release sequence.) This corresponds
1645to the C++0x/C1x <code>memory_order_release</code>.</dd>
Eli Friedmanff030482011-07-28 21:48:00 +00001646<dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
Eli Friedman234bccd2011-08-22 21:35:27 +00001647<code>acquire</code> and <code>release</code> operation on its address.
1648This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
Eli Friedmanff030482011-07-28 21:48:00 +00001649<dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1650<dd>In addition to the guarantees of <code>acq_rel</code>
1651(<code>acquire</code> for an operation which only reads, <code>release</code>
1652for an operation which only writes), there is a global total order on all
1653sequentially-consistent operations on all addresses, which is consistent with
1654the <i>happens-before</i> partial order and with the modification orders of
1655all the affected addresses. Each sequentially-consistent read sees the last
Eli Friedman234bccd2011-08-22 21:35:27 +00001656preceding write to the same address in this global order. This corresponds
1657to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
Eli Friedmanff030482011-07-28 21:48:00 +00001658</dl>
1659
1660<p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1661it only <i>synchronizes with</i> or participates in modification and seq_cst
1662total orderings with other operations running in the same thread (for example,
1663in signal handlers).</p>
1664
1665</div>
1666
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001667</div>
1668
Chris Lattner00950542001-06-06 20:29:01 +00001669<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001670<h2><a name="typesystem">Type System</a></h2>
Chris Lattner261efe92003-11-25 01:02:51 +00001671<!-- *********************************************************************** -->
Chris Lattnerfa730212004-12-09 16:11:40 +00001672
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001673<div>
Chris Lattnerfa730212004-12-09 16:11:40 +00001674
Misha Brukman9d0919f2003-11-08 01:05:38 +00001675<p>The LLVM type system is one of the most important features of the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001676 intermediate representation. Being typed enables a number of optimizations
1677 to be performed on the intermediate representation directly, without having
1678 to do extra analyses on the side before the transformation. A strong type
1679 system makes it easier to read the generated code and enables novel analyses
1680 and transformations that are not feasible to perform on normal three address
1681 code representations.</p>
Chris Lattnerfa730212004-12-09 16:11:40 +00001682
Chris Lattner00950542001-06-06 20:29:01 +00001683<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001684<h3>
1685 <a name="t_classifications">Type Classifications</a>
1686</h3>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001687
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001688<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001689
1690<p>The types fall into a few useful classifications:</p>
Misha Brukmandaa4cb02004-03-01 17:47:27 +00001691
1692<table border="1" cellspacing="0" cellpadding="4">
Chris Lattner261efe92003-11-25 01:02:51 +00001693 <tbody>
Reid Spencerd3f876c2004-11-01 08:19:36 +00001694 <tr><th>Classification</th><th>Types</th></tr>
Chris Lattner261efe92003-11-25 01:02:51 +00001695 <tr>
Chris Lattner4f69f462008-01-04 04:32:38 +00001696 <td><a href="#t_integer">integer</a></td>
Reid Spencer2b916312007-05-16 18:44:01 +00001697 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
Chris Lattner261efe92003-11-25 01:02:51 +00001698 </tr>
1699 <tr>
Chris Lattner4f69f462008-01-04 04:32:38 +00001700 <td><a href="#t_floating">floating point</a></td>
1701 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
Chris Lattner261efe92003-11-25 01:02:51 +00001702 </tr>
1703 <tr>
1704 <td><a name="t_firstclass">first class</a></td>
Chris Lattner4f69f462008-01-04 04:32:38 +00001705 <td><a href="#t_integer">integer</a>,
1706 <a href="#t_floating">floating point</a>,
1707 <a href="#t_pointer">pointer</a>,
Dan Gohman0066db62008-06-18 18:42:13 +00001708 <a href="#t_vector">vector</a>,
Dan Gohmana334d5f2008-05-12 23:51:09 +00001709 <a href="#t_struct">structure</a>,
1710 <a href="#t_array">array</a>,
Nick Lewycky7a0370f2009-05-30 05:06:04 +00001711 <a href="#t_label">label</a>,
1712 <a href="#t_metadata">metadata</a>.
Reid Spencerca86e162006-12-31 07:07:53 +00001713 </td>
Chris Lattner261efe92003-11-25 01:02:51 +00001714 </tr>
Chris Lattner4f69f462008-01-04 04:32:38 +00001715 <tr>
1716 <td><a href="#t_primitive">primitive</a></td>
1717 <td><a href="#t_label">label</a>,
1718 <a href="#t_void">void</a>,
Tobias Grosser05387292010-12-28 20:29:31 +00001719 <a href="#t_integer">integer</a>,
Nick Lewycky7a0370f2009-05-30 05:06:04 +00001720 <a href="#t_floating">floating point</a>,
Dale Johannesen21fe99b2010-10-01 00:48:59 +00001721 <a href="#t_x86mmx">x86mmx</a>,
Nick Lewycky7a0370f2009-05-30 05:06:04 +00001722 <a href="#t_metadata">metadata</a>.</td>
Chris Lattner4f69f462008-01-04 04:32:38 +00001723 </tr>
1724 <tr>
1725 <td><a href="#t_derived">derived</a></td>
Chris Lattnerfdfeb692010-02-12 20:49:41 +00001726 <td><a href="#t_array">array</a>,
Chris Lattner4f69f462008-01-04 04:32:38 +00001727 <a href="#t_function">function</a>,
1728 <a href="#t_pointer">pointer</a>,
1729 <a href="#t_struct">structure</a>,
Chris Lattner4f69f462008-01-04 04:32:38 +00001730 <a href="#t_vector">vector</a>,
1731 <a href="#t_opaque">opaque</a>.
Dan Gohman01ac1012008-10-14 16:32:04 +00001732 </td>
Chris Lattner4f69f462008-01-04 04:32:38 +00001733 </tr>
Chris Lattner261efe92003-11-25 01:02:51 +00001734 </tbody>
Misha Brukman9d0919f2003-11-08 01:05:38 +00001735</table>
Misha Brukmandaa4cb02004-03-01 17:47:27 +00001736
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001737<p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1738 important. Values of these types are the only ones which can be produced by
Nick Lewyckyec38da42009-09-27 00:45:11 +00001739 instructions.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001740
Misha Brukman9d0919f2003-11-08 01:05:38 +00001741</div>
Chris Lattnerc3f59762004-12-09 17:30:23 +00001742
Chris Lattner00950542001-06-06 20:29:01 +00001743<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001744<h3>
1745 <a name="t_primitive">Primitive Types</a>
1746</h3>
Chris Lattner8f8c7b72008-01-04 04:34:14 +00001747
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001748<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001749
Chris Lattner4f69f462008-01-04 04:32:38 +00001750<p>The primitive types are the fundamental building blocks of the LLVM
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001751 system.</p>
Chris Lattner4f69f462008-01-04 04:32:38 +00001752
1753<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001754<h4>
1755 <a name="t_integer">Integer Type</a>
1756</h4>
Nick Lewyckyec38da42009-09-27 00:45:11 +00001757
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001758<div>
Nick Lewyckyec38da42009-09-27 00:45:11 +00001759
1760<h5>Overview:</h5>
1761<p>The integer type is a very simple type that simply specifies an arbitrary
1762 bit width for the integer type desired. Any bit width from 1 bit to
1763 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1764
1765<h5>Syntax:</h5>
1766<pre>
1767 iN
1768</pre>
1769
1770<p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1771 value.</p>
1772
1773<h5>Examples:</h5>
1774<table class="layout">
1775 <tr class="layout">
1776 <td class="left"><tt>i1</tt></td>
1777 <td class="left">a single-bit integer.</td>
1778 </tr>
1779 <tr class="layout">
1780 <td class="left"><tt>i32</tt></td>
1781 <td class="left">a 32-bit integer.</td>
1782 </tr>
1783 <tr class="layout">
1784 <td class="left"><tt>i1942652</tt></td>
1785 <td class="left">a really big integer of over 1 million bits.</td>
1786 </tr>
1787</table>
1788
Nick Lewyckyec38da42009-09-27 00:45:11 +00001789</div>
1790
1791<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001792<h4>
1793 <a name="t_floating">Floating Point Types</a>
1794</h4>
Chris Lattner4f69f462008-01-04 04:32:38 +00001795
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001796<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001797
1798<table>
1799 <tbody>
1800 <tr><th>Type</th><th>Description</th></tr>
1801 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1802 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1803 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1804 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1805 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1806 </tbody>
1807</table>
1808
Chris Lattner4f69f462008-01-04 04:32:38 +00001809</div>
1810
1811<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001812<h4>
1813 <a name="t_x86mmx">X86mmx Type</a>
1814</h4>
Dale Johannesen21fe99b2010-10-01 00:48:59 +00001815
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001816<div>
Dale Johannesen21fe99b2010-10-01 00:48:59 +00001817
1818<h5>Overview:</h5>
1819<p>The x86mmx type represents a value held in an MMX register on an x86 machine. The operations allowed on it are quite limited: parameters and return values, load and store, and bitcast. User-specified MMX instructions are represented as intrinsic or asm calls with arguments and/or results of this type. There are no arrays, vectors or constants of this type.</p>
1820
1821<h5>Syntax:</h5>
1822<pre>
Dale Johannesen473a8c82010-10-01 01:07:02 +00001823 x86mmx
Dale Johannesen21fe99b2010-10-01 00:48:59 +00001824</pre>
1825
1826</div>
1827
1828<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001829<h4>
1830 <a name="t_void">Void Type</a>
1831</h4>
Chris Lattner4f69f462008-01-04 04:32:38 +00001832
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001833<div>
Bill Wendlingc39e3e02009-07-20 02:39:26 +00001834
Chris Lattner4f69f462008-01-04 04:32:38 +00001835<h5>Overview:</h5>
1836<p>The void type does not represent any value and has no size.</p>
1837
1838<h5>Syntax:</h5>
Chris Lattner4f69f462008-01-04 04:32:38 +00001839<pre>
1840 void
1841</pre>
Bill Wendlingc39e3e02009-07-20 02:39:26 +00001842
Chris Lattner4f69f462008-01-04 04:32:38 +00001843</div>
1844
1845<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001846<h4>
1847 <a name="t_label">Label Type</a>
1848</h4>
Chris Lattner4f69f462008-01-04 04:32:38 +00001849
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001850<div>
Bill Wendlingc39e3e02009-07-20 02:39:26 +00001851
Chris Lattner4f69f462008-01-04 04:32:38 +00001852<h5>Overview:</h5>
1853<p>The label type represents code labels.</p>
1854
1855<h5>Syntax:</h5>
Chris Lattner4f69f462008-01-04 04:32:38 +00001856<pre>
1857 label
1858</pre>
Bill Wendlingc39e3e02009-07-20 02:39:26 +00001859
Chris Lattner4f69f462008-01-04 04:32:38 +00001860</div>
1861
Nick Lewycky7a0370f2009-05-30 05:06:04 +00001862<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001863<h4>
1864 <a name="t_metadata">Metadata Type</a>
1865</h4>
Nick Lewycky7a0370f2009-05-30 05:06:04 +00001866
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001867<div>
Bill Wendlingc39e3e02009-07-20 02:39:26 +00001868
Nick Lewycky7a0370f2009-05-30 05:06:04 +00001869<h5>Overview:</h5>
Nick Lewyckyc261df92009-09-27 23:27:42 +00001870<p>The metadata type represents embedded metadata. No derived types may be
1871 created from metadata except for <a href="#t_function">function</a>
1872 arguments.
Nick Lewycky7a0370f2009-05-30 05:06:04 +00001873
1874<h5>Syntax:</h5>
Nick Lewycky7a0370f2009-05-30 05:06:04 +00001875<pre>
1876 metadata
1877</pre>
Bill Wendlingc39e3e02009-07-20 02:39:26 +00001878
Nick Lewycky7a0370f2009-05-30 05:06:04 +00001879</div>
1880
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001881</div>
Chris Lattner4f69f462008-01-04 04:32:38 +00001882
1883<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001884<h3>
1885 <a name="t_derived">Derived Types</a>
1886</h3>
Chris Lattnerc3f59762004-12-09 17:30:23 +00001887
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001888<div>
Chris Lattnerc3f59762004-12-09 17:30:23 +00001889
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001890<p>The real power in LLVM comes from the derived types in the system. This is
1891 what allows a programmer to represent arrays, functions, pointers, and other
Nick Lewyckyec38da42009-09-27 00:45:11 +00001892 useful types. Each of these types contain one or more element types which
1893 may be a primitive type, or another derived type. For example, it is
1894 possible to have a two dimensional array, using an array as the element type
1895 of another array.</p>
Dan Gohmand8791e52009-01-24 15:58:40 +00001896
Chris Lattner1afcace2011-07-09 17:41:24 +00001897</div>
1898
1899
Chris Lattnerfdfeb692010-02-12 20:49:41 +00001900<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001901<h4>
1902 <a name="t_aggregate">Aggregate Types</a>
1903</h4>
Chris Lattnerfdfeb692010-02-12 20:49:41 +00001904
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001905<div>
Chris Lattnerfdfeb692010-02-12 20:49:41 +00001906
1907<p>Aggregate Types are a subset of derived types that can contain multiple
1908 member types. <a href="#t_array">Arrays</a>,
Chris Lattner61c70e92010-08-28 04:09:24 +00001909 <a href="#t_struct">structs</a>, and <a href="#t_vector">vectors</a> are
1910 aggregate types.</p>
Chris Lattnerfdfeb692010-02-12 20:49:41 +00001911
1912</div>
1913
Reid Spencer2b916312007-05-16 18:44:01 +00001914<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001915<h4>
1916 <a name="t_array">Array Type</a>
1917</h4>
Chris Lattnerc3f59762004-12-09 17:30:23 +00001918
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001919<div>
Chris Lattnerc3f59762004-12-09 17:30:23 +00001920
Chris Lattner00950542001-06-06 20:29:01 +00001921<h5>Overview:</h5>
Misha Brukman9d0919f2003-11-08 01:05:38 +00001922<p>The array type is a very simple derived type that arranges elements
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001923 sequentially in memory. The array type requires a size (number of elements)
1924 and an underlying data type.</p>
Chris Lattnerc3f59762004-12-09 17:30:23 +00001925
Chris Lattner7faa8832002-04-14 06:13:44 +00001926<h5>Syntax:</h5>
Chris Lattnerc3f59762004-12-09 17:30:23 +00001927<pre>
1928 [&lt;# elements&gt; x &lt;elementtype&gt;]
1929</pre>
1930
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001931<p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1932 be any type with a size.</p>
Chris Lattnerc3f59762004-12-09 17:30:23 +00001933
Chris Lattner7faa8832002-04-14 06:13:44 +00001934<h5>Examples:</h5>
Reid Spencerd3f876c2004-11-01 08:19:36 +00001935<table class="layout">
1936 <tr class="layout">
Chris Lattner23ff1f92007-12-19 05:04:11 +00001937 <td class="left"><tt>[40 x i32]</tt></td>
1938 <td class="left">Array of 40 32-bit integer values.</td>
1939 </tr>
1940 <tr class="layout">
1941 <td class="left"><tt>[41 x i32]</tt></td>
1942 <td class="left">Array of 41 32-bit integer values.</td>
1943 </tr>
1944 <tr class="layout">
1945 <td class="left"><tt>[4 x i8]</tt></td>
1946 <td class="left">Array of 4 8-bit integer values.</td>
Reid Spencerd3f876c2004-11-01 08:19:36 +00001947 </tr>
Chris Lattner00950542001-06-06 20:29:01 +00001948</table>
Reid Spencerd3f876c2004-11-01 08:19:36 +00001949<p>Here are some examples of multidimensional arrays:</p>
1950<table class="layout">
1951 <tr class="layout">
Chris Lattner23ff1f92007-12-19 05:04:11 +00001952 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1953 <td class="left">3x4 array of 32-bit integer values.</td>
1954 </tr>
1955 <tr class="layout">
1956 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1957 <td class="left">12x10 array of single precision floating point values.</td>
1958 </tr>
1959 <tr class="layout">
1960 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1961 <td class="left">2x3x4 array of 16-bit integer values.</td>
Reid Spencerd3f876c2004-11-01 08:19:36 +00001962 </tr>
1963</table>
Chris Lattnere67a9512005-06-24 17:22:57 +00001964
Dan Gohman7657f6b2009-11-09 19:01:53 +00001965<p>There is no restriction on indexing beyond the end of the array implied by
1966 a static type (though there are restrictions on indexing beyond the bounds
1967 of an allocated object in some cases). This means that single-dimension
1968 'variable sized array' addressing can be implemented in LLVM with a zero
1969 length array type. An implementation of 'pascal style arrays' in LLVM could
1970 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
Chris Lattnere67a9512005-06-24 17:22:57 +00001971
Misha Brukman9d0919f2003-11-08 01:05:38 +00001972</div>
Reid Spencerd3f876c2004-11-01 08:19:36 +00001973
Chris Lattner00950542001-06-06 20:29:01 +00001974<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00001975<h4>
1976 <a name="t_function">Function Type</a>
1977</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001978
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00001979<div>
Chris Lattnerf4cde4e2008-04-23 04:59:35 +00001980
Chris Lattner00950542001-06-06 20:29:01 +00001981<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001982<p>The function type can be thought of as a function signature. It consists of
1983 a return type and a list of formal parameter types. The return type of a
Chris Lattner61c70e92010-08-28 04:09:24 +00001984 function type is a first class type or a void type.</p>
Devang Patelc3fc6df2008-03-10 20:49:15 +00001985
Chris Lattner00950542001-06-06 20:29:01 +00001986<h5>Syntax:</h5>
Chris Lattnerf4cde4e2008-04-23 04:59:35 +00001987<pre>
Nick Lewycky51386942009-09-27 07:55:32 +00001988 &lt;returntype&gt; (&lt;parameter list&gt;)
Chris Lattnerf4cde4e2008-04-23 04:59:35 +00001989</pre>
1990
John Criswell0ec250c2005-10-24 16:17:18 +00001991<p>...where '<tt>&lt;parameter list&gt;</tt>' is a comma-separated list of type
Bill Wendlinge910b4c2009-07-20 02:29:24 +00001992 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1993 which indicates that the function takes a variable number of arguments.
1994 Variable argument functions can access their arguments with
1995 the <a href="#int_varargs">variable argument handling intrinsic</a>
Chris Lattner0724fbd2010-03-02 06:36:51 +00001996 functions. '<tt>&lt;returntype&gt;</tt>' is any type except
Nick Lewyckyc261df92009-09-27 23:27:42 +00001997 <a href="#t_label">label</a>.</p>
Chris Lattnerf4cde4e2008-04-23 04:59:35 +00001998
Chris Lattner00950542001-06-06 20:29:01 +00001999<h5>Examples:</h5>
Reid Spencerd3f876c2004-11-01 08:19:36 +00002000<table class="layout">
2001 <tr class="layout">
Reid Spencer92f82302006-12-31 07:18:34 +00002002 <td class="left"><tt>i32 (i32)</tt></td>
2003 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
Reid Spencerd3f876c2004-11-01 08:19:36 +00002004 </td>
Reid Spencer92f82302006-12-31 07:18:34 +00002005 </tr><tr class="layout">
Chris Lattner0724fbd2010-03-02 06:36:51 +00002006 <td class="left"><tt>float&nbsp;(i16,&nbsp;i32&nbsp;*)&nbsp;*
Reid Spencerf17a0b72006-12-31 07:20:23 +00002007 </tt></td>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00002008 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
Chris Lattner0724fbd2010-03-02 06:36:51 +00002009 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2010 returning <tt>float</tt>.
Reid Spencer92f82302006-12-31 07:18:34 +00002011 </td>
2012 </tr><tr class="layout">
2013 <td class="left"><tt>i32 (i8*, ...)</tt></td>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00002014 <td class="left">A vararg function that takes at least one
2015 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2016 which returns an integer. This is the signature for <tt>printf</tt> in
Reid Spencer92f82302006-12-31 07:18:34 +00002017 LLVM.
Reid Spencerd3f876c2004-11-01 08:19:36 +00002018 </td>
Devang Patela582f402008-03-24 05:35:41 +00002019 </tr><tr class="layout">
2020 <td class="left"><tt>{i32, i32} (i32)</tt></td>
Nick Lewycky51386942009-09-27 07:55:32 +00002021 <td class="left">A function taking an <tt>i32</tt>, returning a
2022 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
Devang Patela582f402008-03-24 05:35:41 +00002023 </td>
Reid Spencerd3f876c2004-11-01 08:19:36 +00002024 </tr>
Chris Lattner00950542001-06-06 20:29:01 +00002025</table>
Misha Brukmandaa4cb02004-03-01 17:47:27 +00002026
Misha Brukman9d0919f2003-11-08 01:05:38 +00002027</div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002028
Chris Lattner00950542001-06-06 20:29:01 +00002029<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002030<h4>
2031 <a name="t_struct">Structure Type</a>
2032</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002033
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002034<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002035
Chris Lattner00950542001-06-06 20:29:01 +00002036<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002037<p>The structure type is used to represent a collection of data members together
Chris Lattner1afcace2011-07-09 17:41:24 +00002038 in memory. The elements of a structure may be any type that has a size.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002039
Jeffrey Yasskin7a088cf2010-01-11 19:19:26 +00002040<p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2041 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2042 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2043 Structures in registers are accessed using the
2044 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2045 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
Chris Lattner1afcace2011-07-09 17:41:24 +00002046
2047<p>Structures may optionally be "packed" structures, which indicate that the
2048 alignment of the struct is one byte, and that there is no padding between
Chris Lattner2c38d652011-08-12 17:31:02 +00002049 the elements. In non-packed structs, padding between field types is inserted
2050 as defined by the TargetData string in the module, which is required to match
2051 what the underlying processor expects.</p>
Chris Lattner1afcace2011-07-09 17:41:24 +00002052
Chris Lattner2c38d652011-08-12 17:31:02 +00002053<p>Structures can either be "literal" or "identified". A literal structure is
2054 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2055 types are always defined at the top level with a name. Literal types are
2056 uniqued by their contents and can never be recursive or opaque since there is
Chris Lattneraa175c32011-08-12 18:12:40 +00002057 no way to write one. Identified types can be recursive, can be opaqued, and are
Chris Lattner2c38d652011-08-12 17:31:02 +00002058 never uniqued.
Chris Lattner1afcace2011-07-09 17:41:24 +00002059</p>
2060
Chris Lattner00950542001-06-06 20:29:01 +00002061<h5>Syntax:</h5>
Bill Wendlingc39e3e02009-07-20 02:39:26 +00002062<pre>
Chris Lattner2c38d652011-08-12 17:31:02 +00002063 %T1 = type { &lt;type list&gt; } <i>; Identified normal struct type</i>
2064 %T2 = type &lt;{ &lt;type list&gt; }&gt; <i>; Identified packed struct type</i>
Bill Wendlingc39e3e02009-07-20 02:39:26 +00002065</pre>
Chris Lattner1afcace2011-07-09 17:41:24 +00002066
Chris Lattner00950542001-06-06 20:29:01 +00002067<h5>Examples:</h5>
Reid Spencerd3f876c2004-11-01 08:19:36 +00002068<table class="layout">
2069 <tr class="layout">
Jeff Cohen6f1cc772007-04-22 01:17:39 +00002070 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2071 <td class="left">A triple of three <tt>i32</tt> values</td>
Chris Lattner1afcace2011-07-09 17:41:24 +00002072 </tr>
2073 <tr class="layout">
Jeff Cohen6f1cc772007-04-22 01:17:39 +00002074 <td class="left"><tt>{&nbsp;float,&nbsp;i32&nbsp;(i32)&nbsp;*&nbsp;}</tt></td>
2075 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2076 second element is a <a href="#t_pointer">pointer</a> to a
2077 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2078 an <tt>i32</tt>.</td>
Reid Spencerd3f876c2004-11-01 08:19:36 +00002079 </tr>
Chris Lattner1afcace2011-07-09 17:41:24 +00002080 <tr class="layout">
2081 <td class="left"><tt>&lt;{ i8, i32 }&gt;</tt></td>
2082 <td class="left">A packed struct known to be 5 bytes in size.</td>
2083 </tr>
Chris Lattner00950542001-06-06 20:29:01 +00002084</table>
Dan Gohmand8791e52009-01-24 15:58:40 +00002085
Misha Brukman9d0919f2003-11-08 01:05:38 +00002086</div>
Chris Lattner1afcace2011-07-09 17:41:24 +00002087
Chris Lattner00950542001-06-06 20:29:01 +00002088<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002089<h4>
Chris Lattner628ed392011-07-23 19:59:08 +00002090 <a name="t_opaque">Opaque Structure Types</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002091</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002092
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002093<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002094
Andrew Lenharth75e10682006-12-08 17:13:00 +00002095<h5>Overview:</h5>
Chris Lattner628ed392011-07-23 19:59:08 +00002096<p>Opaque structure types are used to represent named structure types that do
2097 not have a body specified. This corresponds (for example) to the C notion of
2098 a forward declared structure.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002099
Andrew Lenharth75e10682006-12-08 17:13:00 +00002100<h5>Syntax:</h5>
Bill Wendlingc39e3e02009-07-20 02:39:26 +00002101<pre>
Chris Lattner1afcace2011-07-09 17:41:24 +00002102 %X = type opaque
2103 %52 = type opaque
Bill Wendlingc39e3e02009-07-20 02:39:26 +00002104</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002105
Andrew Lenharth75e10682006-12-08 17:13:00 +00002106<h5>Examples:</h5>
2107<table class="layout">
2108 <tr class="layout">
Chris Lattner1afcace2011-07-09 17:41:24 +00002109 <td class="left"><tt>opaque</tt></td>
2110 <td class="left">An opaque type.</td>
Andrew Lenharth75e10682006-12-08 17:13:00 +00002111 </tr>
2112</table>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002113
Andrew Lenharth75e10682006-12-08 17:13:00 +00002114</div>
2115
Chris Lattner1afcace2011-07-09 17:41:24 +00002116
2117
Andrew Lenharth75e10682006-12-08 17:13:00 +00002118<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002119<h4>
2120 <a name="t_pointer">Pointer Type</a>
2121</h4>
Chris Lattner0fd4a272009-02-08 19:53:29 +00002122
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002123<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002124
2125<h5>Overview:</h5>
Dan Gohmanff3ef322010-02-25 16:50:07 +00002126<p>The pointer type is used to specify memory locations.
2127 Pointers are commonly used to reference objects in memory.</p>
2128
2129<p>Pointer types may have an optional address space attribute defining the
2130 numbered address space where the pointed-to object resides. The default
2131 address space is number zero. The semantics of non-zero address
2132 spaces are target-specific.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002133
2134<p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2135 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
Chris Lattner0fd4a272009-02-08 19:53:29 +00002136
Chris Lattner7faa8832002-04-14 06:13:44 +00002137<h5>Syntax:</h5>
Bill Wendlingc39e3e02009-07-20 02:39:26 +00002138<pre>
2139 &lt;type&gt; *
2140</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002141
Chris Lattner7faa8832002-04-14 06:13:44 +00002142<h5>Examples:</h5>
Reid Spencerd3f876c2004-11-01 08:19:36 +00002143<table class="layout">
2144 <tr class="layout">
Dan Gohman2a08c532009-01-04 23:44:43 +00002145 <td class="left"><tt>[4 x i32]*</tt></td>
Chris Lattner23ff1f92007-12-19 05:04:11 +00002146 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2147 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2148 </tr>
2149 <tr class="layout">
Dan Gohmanfe47aae2010-05-28 17:13:49 +00002150 <td class="left"><tt>i32 (i32*) *</tt></td>
Chris Lattner23ff1f92007-12-19 05:04:11 +00002151 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
Reid Spencerca86e162006-12-31 07:07:53 +00002152 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
Chris Lattner23ff1f92007-12-19 05:04:11 +00002153 <tt>i32</tt>.</td>
2154 </tr>
2155 <tr class="layout">
2156 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2157 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2158 that resides in address space #5.</td>
Reid Spencerd3f876c2004-11-01 08:19:36 +00002159 </tr>
Misha Brukman9d0919f2003-11-08 01:05:38 +00002160</table>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002161
Misha Brukman9d0919f2003-11-08 01:05:38 +00002162</div>
Reid Spencerd3f876c2004-11-01 08:19:36 +00002163
Chris Lattnera58561b2004-08-12 19:12:28 +00002164<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002165<h4>
2166 <a name="t_vector">Vector Type</a>
2167</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002168
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002169<div>
Chris Lattner69c11bb2005-04-25 17:34:15 +00002170
Chris Lattnera58561b2004-08-12 19:12:28 +00002171<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002172<p>A vector type is a simple derived type that represents a vector of elements.
2173 Vector types are used when multiple primitive data are operated in parallel
2174 using a single instruction (SIMD). A vector type requires a size (number of
Duncan Sandsd40d14e2009-11-27 13:38:03 +00002175 elements) and an underlying primitive data type. Vector types are considered
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002176 <a href="#t_firstclass">first class</a>.</p>
Chris Lattner69c11bb2005-04-25 17:34:15 +00002177
Chris Lattnera58561b2004-08-12 19:12:28 +00002178<h5>Syntax:</h5>
Chris Lattner69c11bb2005-04-25 17:34:15 +00002179<pre>
2180 &lt; &lt;# elements&gt; x &lt;elementtype&gt; &gt;
2181</pre>
2182
Chris Lattner7d2e7be2010-10-10 18:20:35 +00002183<p>The number of elements is a constant integer value larger than 0; elementtype
2184 may be any integer or floating point type. Vectors of size zero are not
2185 allowed, and pointers are not allowed as the element type.</p>
Chris Lattner69c11bb2005-04-25 17:34:15 +00002186
Chris Lattnera58561b2004-08-12 19:12:28 +00002187<h5>Examples:</h5>
Reid Spencerd3f876c2004-11-01 08:19:36 +00002188<table class="layout">
2189 <tr class="layout">
Chris Lattner23ff1f92007-12-19 05:04:11 +00002190 <td class="left"><tt>&lt;4 x i32&gt;</tt></td>
2191 <td class="left">Vector of 4 32-bit integer values.</td>
2192 </tr>
2193 <tr class="layout">
2194 <td class="left"><tt>&lt;8 x float&gt;</tt></td>
2195 <td class="left">Vector of 8 32-bit floating-point values.</td>
2196 </tr>
2197 <tr class="layout">
2198 <td class="left"><tt>&lt;2 x i64&gt;</tt></td>
2199 <td class="left">Vector of 2 64-bit integer values.</td>
Reid Spencerd3f876c2004-11-01 08:19:36 +00002200 </tr>
2201</table>
Dan Gohmand8791e52009-01-24 15:58:40 +00002202
Misha Brukman9d0919f2003-11-08 01:05:38 +00002203</div>
2204
Bill Wendlingaf75f0c2011-07-31 06:47:33 +00002205</div>
2206
Chris Lattnerc3f59762004-12-09 17:30:23 +00002207<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002208<h2><a name="constants">Constants</a></h2>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002209<!-- *********************************************************************** -->
2210
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002211<div>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002212
2213<p>LLVM has several different basic types of constants. This section describes
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002214 them all and their syntax.</p>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002215
Chris Lattnerc3f59762004-12-09 17:30:23 +00002216<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002217<h3>
2218 <a name="simpleconstants">Simple Constants</a>
2219</h3>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002220
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002221<div>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002222
2223<dl>
2224 <dt><b>Boolean constants</b></dt>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002225 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
Nick Lewyckyec38da42009-09-27 00:45:11 +00002226 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002227
2228 <dt><b>Integer constants</b></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002229 <dd>Standard integers (such as '4') are constants of
2230 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2231 with integer types.</dd>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002232
2233 <dt><b>Floating point constants</b></dt>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002234 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002235 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2236 notation (see below). The assembler requires the exact decimal value of a
2237 floating-point constant. For example, the assembler accepts 1.25 but
2238 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2239 constants must have a <a href="#t_floating">floating point</a> type. </dd>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002240
2241 <dt><b>Null pointer constants</b></dt>
John Criswell9e2485c2004-12-10 15:51:16 +00002242 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002243 and must be of <a href="#t_pointer">pointer type</a>.</dd>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002244</dl>
2245
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002246<p>The one non-intuitive notation for constants is the hexadecimal form of
2247 floating point constants. For example, the form '<tt>double
2248 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2249 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2250 constants are required (and the only time that they are generated by the
2251 disassembler) is when a floating point constant must be emitted but it cannot
2252 be represented as a decimal floating point number in a reasonable number of
2253 digits. For example, NaN's, infinities, and other special values are
2254 represented in their IEEE hexadecimal format so that assembly and disassembly
2255 do not cause any bits to change in the constants.</p>
2256
Dale Johannesenbd5e5a82009-02-11 22:14:51 +00002257<p>When using the hexadecimal form, constants of types float and double are
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002258 represented using the 16-digit form shown above (which matches the IEEE754
2259 representation for double); float values must, however, be exactly
2260 representable as IEE754 single precision. Hexadecimal format is always used
2261 for long double, and there are three forms of long double. The 80-bit format
2262 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2263 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2264 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2265 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2266 currently supported target uses this format. Long doubles will only work if
2267 they match the long double format on your target. All hexadecimal formats
2268 are big-endian (sign bit at the left).</p>
2269
Dale Johannesen21fe99b2010-10-01 00:48:59 +00002270<p>There are no constants of type x86mmx.</p>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002271</div>
2272
2273<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002274<h3>
Bill Wendlingd9fe2982009-07-20 02:32:41 +00002275<a name="aggregateconstants"></a> <!-- old anchor -->
2276<a name="complexconstants">Complex Constants</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002277</h3>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002278
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002279<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002280
Chris Lattner70882792009-02-28 18:32:25 +00002281<p>Complex constants are a (potentially recursive) combination of simple
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002282 constants and smaller complex constants.</p>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002283
2284<dl>
2285 <dt><b>Structure constants</b></dt>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002286 <dd>Structure constants are represented with notation similar to structure
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002287 type definitions (a comma separated list of elements, surrounded by braces
2288 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2289 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2290 Structure constants must have <a href="#t_struct">structure type</a>, and
2291 the number and types of elements must match those specified by the
2292 type.</dd>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002293
2294 <dt><b>Array constants</b></dt>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002295 <dd>Array constants are represented with notation similar to array type
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002296 definitions (a comma separated list of elements, surrounded by square
2297 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2298 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2299 the number and types of elements must match those specified by the
2300 type.</dd>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002301
Reid Spencer485bad12007-02-15 03:07:05 +00002302 <dt><b>Vector constants</b></dt>
Reid Spencer485bad12007-02-15 03:07:05 +00002303 <dd>Vector constants are represented with notation similar to vector type
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002304 definitions (a comma separated list of elements, surrounded by
2305 less-than/greater-than's (<tt>&lt;&gt;</tt>)). For example: "<tt>&lt; i32
2306 42, i32 11, i32 74, i32 100 &gt;</tt>". Vector constants must
2307 have <a href="#t_vector">vector type</a>, and the number and types of
2308 elements must match those specified by the type.</dd>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002309
2310 <dt><b>Zero initialization</b></dt>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002311 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
Chris Lattnerfdfeb692010-02-12 20:49:41 +00002312 value to zero of <em>any</em> type, including scalar and
2313 <a href="#t_aggregate">aggregate</a> types.
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002314 This is often used to avoid having to print large zero initializers
2315 (e.g. for large arrays) and is always exactly equivalent to using explicit
2316 zero initializers.</dd>
Nick Lewycky21cc4462009-04-04 07:22:01 +00002317
2318 <dt><b>Metadata node</b></dt>
Nick Lewycky1e8c7a62009-05-30 16:08:30 +00002319 <dd>A metadata node is a structure-like constant with
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002320 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2321 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2322 be interpreted as part of the instruction stream, metadata is a place to
2323 attach additional information such as debug info.</dd>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002324</dl>
2325
2326</div>
2327
2328<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002329<h3>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002330 <a name="globalconstants">Global Variable and Function Addresses</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002331</h3>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002332
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002333<div>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002334
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002335<p>The addresses of <a href="#globalvars">global variables</a>
2336 and <a href="#functionstructure">functions</a> are always implicitly valid
2337 (link-time) constants. These constants are explicitly referenced when
2338 the <a href="#identifiers">identifier for the global</a> is used and always
2339 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2340 legal LLVM file:</p>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002341
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002342<pre class="doc_code">
Chris Lattnera18a4242007-06-06 18:28:13 +00002343@X = global i32 17
2344@Y = global i32 42
2345@Z = global [2 x i32*] [ i32* @X, i32* @Y ]
Chris Lattnerc3f59762004-12-09 17:30:23 +00002346</pre>
2347
2348</div>
2349
2350<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002351<h3>
2352 <a name="undefvalues">Undefined Values</a>
2353</h3>
2354
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002355<div>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002356
Chris Lattner48a109c2009-09-07 22:52:39 +00002357<p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
Benjamin Kramer8040cd32009-10-12 14:46:08 +00002358 indicates that the user of the value may receive an unspecified bit-pattern.
Bill Wendling1b383ba2010-10-27 01:07:41 +00002359 Undefined values may be of any type (other than '<tt>label</tt>'
2360 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002361
Chris Lattnerc608cb12009-09-11 01:49:31 +00002362<p>Undefined values are useful because they indicate to the compiler that the
Chris Lattner48a109c2009-09-07 22:52:39 +00002363 program is well defined no matter what value is used. This gives the
2364 compiler more freedom to optimize. Here are some examples of (potentially
2365 surprising) transformations that are valid (in pseudo IR):</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002366
Chris Lattner48a109c2009-09-07 22:52:39 +00002367
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002368<pre class="doc_code">
Chris Lattner48a109c2009-09-07 22:52:39 +00002369 %A = add %X, undef
2370 %B = sub %X, undef
2371 %C = xor %X, undef
2372Safe:
2373 %A = undef
2374 %B = undef
2375 %C = undef
2376</pre>
Chris Lattner48a109c2009-09-07 22:52:39 +00002377
2378<p>This is safe because all of the output bits are affected by the undef bits.
Bill Wendling1b383ba2010-10-27 01:07:41 +00002379 Any output bit can have a zero or one depending on the input bits.</p>
Chris Lattner48a109c2009-09-07 22:52:39 +00002380
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002381<pre class="doc_code">
Chris Lattner48a109c2009-09-07 22:52:39 +00002382 %A = or %X, undef
2383 %B = and %X, undef
2384Safe:
2385 %A = -1
2386 %B = 0
2387Unsafe:
2388 %A = undef
2389 %B = undef
2390</pre>
Chris Lattner48a109c2009-09-07 22:52:39 +00002391
2392<p>These logical operations have bits that are not always affected by the input.
Bill Wendling1b383ba2010-10-27 01:07:41 +00002393 For example, if <tt>%X</tt> has a zero bit, then the output of the
2394 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2395 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2396 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2397 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2398 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2399 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2400 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
Chris Lattner48a109c2009-09-07 22:52:39 +00002401
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002402<pre class="doc_code">
Chris Lattner48a109c2009-09-07 22:52:39 +00002403 %A = select undef, %X, %Y
2404 %B = select undef, 42, %Y
2405 %C = select %X, %Y, undef
2406Safe:
2407 %A = %X (or %Y)
2408 %B = 42 (or %Y)
2409 %C = %Y
2410Unsafe:
2411 %A = undef
2412 %B = undef
2413 %C = undef
2414</pre>
Chris Lattner48a109c2009-09-07 22:52:39 +00002415
Bill Wendling1b383ba2010-10-27 01:07:41 +00002416<p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2417 branch) conditions can go <em>either way</em>, but they have to come from one
2418 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2419 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2420 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2421 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2422 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2423 eliminated.</p>
Chris Lattner48a109c2009-09-07 22:52:39 +00002424
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002425<pre class="doc_code">
Chris Lattner48a109c2009-09-07 22:52:39 +00002426 %A = xor undef, undef
Eric Christopher6c7e8a02009-12-05 02:46:03 +00002427
Chris Lattner48a109c2009-09-07 22:52:39 +00002428 %B = undef
2429 %C = xor %B, %B
2430
2431 %D = undef
2432 %E = icmp lt %D, 4
2433 %F = icmp gte %D, 4
2434
2435Safe:
2436 %A = undef
2437 %B = undef
2438 %C = undef
2439 %D = undef
2440 %E = undef
2441 %F = undef
2442</pre>
Chris Lattner48a109c2009-09-07 22:52:39 +00002443
Bill Wendling1b383ba2010-10-27 01:07:41 +00002444<p>This example points out that two '<tt>undef</tt>' operands are not
2445 necessarily the same. This can be surprising to people (and also matches C
2446 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2447 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2448 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2449 its value over its "live range". This is true because the variable doesn't
2450 actually <em>have a live range</em>. Instead, the value is logically read
2451 from arbitrary registers that happen to be around when needed, so the value
2452 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2453 need to have the same semantics or the core LLVM "replace all uses with"
2454 concept would not hold.</p>
Chris Lattner6e9057b2009-09-07 23:33:52 +00002455
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002456<pre class="doc_code">
Chris Lattner6e9057b2009-09-07 23:33:52 +00002457 %A = fdiv undef, %X
2458 %B = fdiv %X, undef
2459Safe:
2460 %A = undef
2461b: unreachable
2462</pre>
Chris Lattner6e9057b2009-09-07 23:33:52 +00002463
2464<p>These examples show the crucial difference between an <em>undefined
Bill Wendling1b383ba2010-10-27 01:07:41 +00002465 value</em> and <em>undefined behavior</em>. An undefined value (like
2466 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2467 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2468 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2469 defined on SNaN's. However, in the second example, we can make a more
2470 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2471 arbitrary value, we are allowed to assume that it could be zero. Since a
2472 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2473 the operation does not execute at all. This allows us to delete the divide and
2474 all code after it. Because the undefined operation "can't happen", the
2475 optimizer can assume that it occurs in dead code.</p>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00002476
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002477<pre class="doc_code">
Chris Lattner6e9057b2009-09-07 23:33:52 +00002478a: store undef -> %X
2479b: store %X -> undef
2480Safe:
2481a: &lt;deleted&gt;
2482b: unreachable
2483</pre>
Chris Lattner6e9057b2009-09-07 23:33:52 +00002484
Bill Wendling1b383ba2010-10-27 01:07:41 +00002485<p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2486 undefined value can be assumed to not have any effect; we can assume that the
2487 value is overwritten with bits that happen to match what was already there.
2488 However, a store <em>to</em> an undefined location could clobber arbitrary
2489 memory, therefore, it has undefined behavior.</p>
Chris Lattner6e9057b2009-09-07 23:33:52 +00002490
Chris Lattnerc3f59762004-12-09 17:30:23 +00002491</div>
2492
2493<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002494<h3>
2495 <a name="trapvalues">Trap Values</a>
2496</h3>
2497
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002498<div>
Dan Gohmanfff6c532010-04-22 23:14:21 +00002499
Dan Gohmanc68ce062010-04-26 20:21:21 +00002500<p>Trap values are similar to <a href="#undefvalues">undef values</a>, however
Dan Gohmanfff6c532010-04-22 23:14:21 +00002501 instead of representing an unspecified bit pattern, they represent the
2502 fact that an instruction or constant expression which cannot evoke side
2503 effects has nevertheless detected a condition which results in undefined
Dan Gohmanc68ce062010-04-26 20:21:21 +00002504 behavior.</p>
Dan Gohmanfff6c532010-04-22 23:14:21 +00002505
Dan Gohman34b3d992010-04-28 00:49:41 +00002506<p>There is currently no way of representing a trap value in the IR; they
Dan Gohman855abed2010-05-03 14:51:43 +00002507 only exist when produced by operations such as
Dan Gohman34b3d992010-04-28 00:49:41 +00002508 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
Dan Gohmanae11c3f2010-04-26 23:36:52 +00002509
Dan Gohman34b3d992010-04-28 00:49:41 +00002510<p>Trap value behavior is defined in terms of value <i>dependence</i>:</p>
Dan Gohmanae11c3f2010-04-26 23:36:52 +00002511
Dan Gohman34b3d992010-04-28 00:49:41 +00002512<ul>
2513<li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2514 their operands.</li>
2515
2516<li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2517 to their dynamic predecessor basic block.</li>
2518
2519<li>Function arguments depend on the corresponding actual argument values in
2520 the dynamic callers of their functions.</li>
2521
2522<li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2523 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2524 control back to them.</li>
2525
Dan Gohmanb5328162010-05-03 14:55:22 +00002526<li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2527 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_unwind"><tt>unwind</tt></a>,
2528 or exception-throwing call instructions that dynamically transfer control
2529 back to them.</li>
2530
Dan Gohman34b3d992010-04-28 00:49:41 +00002531<li>Non-volatile loads and stores depend on the most recent stores to all of the
2532 referenced memory addresses, following the order in the IR
2533 (including loads and stores implied by intrinsics such as
2534 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2535
Dan Gohman7c24ff12010-05-03 14:59:34 +00002536<!-- TODO: In the case of multiple threads, this only applies if the store
2537 "happens-before" the load or store. -->
Dan Gohmanae11c3f2010-04-26 23:36:52 +00002538
Dan Gohman34b3d992010-04-28 00:49:41 +00002539<!-- TODO: floating-point exception state -->
Dan Gohmanae11c3f2010-04-26 23:36:52 +00002540
Dan Gohman34b3d992010-04-28 00:49:41 +00002541<li>An instruction with externally visible side effects depends on the most
2542 recent preceding instruction with externally visible side effects, following
Dan Gohmanff70fe42010-07-06 15:26:33 +00002543 the order in the IR. (This includes
2544 <a href="#volatile">volatile operations</a>.)</li>
Dan Gohmanfff6c532010-04-22 23:14:21 +00002545
Dan Gohmanb5328162010-05-03 14:55:22 +00002546<li>An instruction <i>control-depends</i> on a
2547 <a href="#terminators">terminator instruction</a>
2548 if the terminator instruction has multiple successors and the instruction
2549 is always executed when control transfers to one of the successors, and
Chris Lattner7a2bdde2011-04-15 05:18:47 +00002550 may not be executed when control is transferred to another.</li>
Dan Gohman34b3d992010-04-28 00:49:41 +00002551
Dan Gohmanca4cac42011-04-12 23:05:59 +00002552<li>Additionally, an instruction also <i>control-depends</i> on a terminator
2553 instruction if the set of instructions it otherwise depends on would be
Chris Lattner7a2bdde2011-04-15 05:18:47 +00002554 different if the terminator had transferred control to a different
Dan Gohmanca4cac42011-04-12 23:05:59 +00002555 successor.</li>
2556
Dan Gohman34b3d992010-04-28 00:49:41 +00002557<li>Dependence is transitive.</li>
2558
2559</ul>
Dan Gohman34b3d992010-04-28 00:49:41 +00002560
2561<p>Whenever a trap value is generated, all values which depend on it evaluate
2562 to trap. If they have side effects, the evoke their side effects as if each
2563 operand with a trap value were undef. If they have externally-visible side
2564 effects, the behavior is undefined.</p>
2565
2566<p>Here are some examples:</p>
Dan Gohmanc30f6e12010-04-26 20:54:53 +00002567
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002568<pre class="doc_code">
Dan Gohmanae11c3f2010-04-26 23:36:52 +00002569entry:
2570 %trap = sub nuw i32 0, 1 ; Results in a trap value.
Dan Gohman34b3d992010-04-28 00:49:41 +00002571 %still_trap = and i32 %trap, 0 ; Whereas (and i32 undef, 0) would return 0.
2572 %trap_yet_again = getelementptr i32* @h, i32 %still_trap
2573 store i32 0, i32* %trap_yet_again ; undefined behavior
2574
2575 store i32 %trap, i32* @g ; Trap value conceptually stored to memory.
2576 %trap2 = load i32* @g ; Returns a trap value, not just undef.
2577
2578 volatile store i32 %trap, i32* @g ; External observation; undefined behavior.
2579
2580 %narrowaddr = bitcast i32* @g to i16*
2581 %wideaddr = bitcast i32* @g to i64*
Nick Lewycky64f9fb12011-05-16 19:29:30 +00002582 %trap3 = load i16* %narrowaddr ; Returns a trap value.
2583 %trap4 = load i64* %wideaddr ; Returns a trap value.
Dan Gohman34b3d992010-04-28 00:49:41 +00002584
Nick Lewycky64f9fb12011-05-16 19:29:30 +00002585 %cmp = icmp slt i32 %trap, 0 ; Returns a trap value.
2586 br i1 %cmp, label %true, label %end ; Branch to either destination.
Dan Gohmanae11c3f2010-04-26 23:36:52 +00002587
2588true:
Dan Gohman34b3d992010-04-28 00:49:41 +00002589 volatile store i32 0, i32* @g ; This is control-dependent on %cmp, so
2590 ; it has undefined behavior.
Dan Gohmanae11c3f2010-04-26 23:36:52 +00002591 br label %end
2592
2593end:
2594 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2595 ; Both edges into this PHI are
2596 ; control-dependent on %cmp, so this
Dan Gohman34b3d992010-04-28 00:49:41 +00002597 ; always results in a trap value.
Dan Gohmanae11c3f2010-04-26 23:36:52 +00002598
Dan Gohmanca4cac42011-04-12 23:05:59 +00002599 volatile store i32 0, i32* @g ; This would depend on the store in %true
2600 ; if %cmp is true, or the store in %entry
2601 ; otherwise, so this is undefined behavior.
2602
Nick Lewycky64f9fb12011-05-16 19:29:30 +00002603 br i1 %cmp, label %second_true, label %second_end
Dan Gohmanca4cac42011-04-12 23:05:59 +00002604 ; The same branch again, but this time the
2605 ; true block doesn't have side effects.
2606
2607second_true:
2608 ; No side effects!
Nick Lewycky64f9fb12011-05-16 19:29:30 +00002609 ret void
Dan Gohmanca4cac42011-04-12 23:05:59 +00002610
2611second_end:
2612 volatile store i32 0, i32* @g ; This time, the instruction always depends
2613 ; on the store in %end. Also, it is
2614 ; control-equivalent to %end, so this is
Nick Lewycky64f9fb12011-05-16 19:29:30 +00002615 ; well-defined (again, ignoring earlier
Dan Gohmanae11c3f2010-04-26 23:36:52 +00002616 ; undefined behavior in this example).
Dan Gohmanae11c3f2010-04-26 23:36:52 +00002617</pre>
Dan Gohmanfff6c532010-04-22 23:14:21 +00002618
Dan Gohmanfff6c532010-04-22 23:14:21 +00002619</div>
2620
2621<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002622<h3>
2623 <a name="blockaddress">Addresses of Basic Blocks</a>
2624</h3>
2625
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002626<div>
Chris Lattnerc6f44362009-10-27 21:01:34 +00002627
Chris Lattnercdfc9402009-11-01 01:27:45 +00002628<p><b><tt>blockaddress(@function, %block)</tt></b></p>
Chris Lattnerc6f44362009-10-27 21:01:34 +00002629
2630<p>The '<tt>blockaddress</tt>' constant computes the address of the specified
Chris Lattner2dfdf2a2009-10-27 21:49:40 +00002631 basic block in the specified function, and always has an i8* type. Taking
Chris Lattnercdfc9402009-11-01 01:27:45 +00002632 the address of the entry block is illegal.</p>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00002633
Chris Lattnerc6f44362009-10-27 21:01:34 +00002634<p>This value only has defined behavior when used as an operand to the
Bill Wendling1b383ba2010-10-27 01:07:41 +00002635 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2636 comparisons against null. Pointer equality tests between labels addresses
2637 results in undefined behavior &mdash; though, again, comparison against null
2638 is ok, and no label is equal to the null pointer. This may be passed around
2639 as an opaque pointer sized value as long as the bits are not inspected. This
2640 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2641 long as the original value is reconstituted before the <tt>indirectbr</tt>
2642 instruction.</p>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00002643
Bill Wendling1b383ba2010-10-27 01:07:41 +00002644<p>Finally, some targets may provide defined semantics when using the value as
2645 the operand to an inline assembly, but that is target specific.</p>
Chris Lattnerc6f44362009-10-27 21:01:34 +00002646
2647</div>
2648
2649
2650<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002651<h3>
2652 <a name="constantexprs">Constant Expressions</a>
2653</h3>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002654
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002655<div>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002656
2657<p>Constant expressions are used to allow expressions involving other constants
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002658 to be used as constants. Constant expressions may be of
2659 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2660 operation that does not have side effects (e.g. load and call are not
Bill Wendling1b383ba2010-10-27 01:07:41 +00002661 supported). The following is the syntax for constant expressions:</p>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002662
2663<dl>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002664 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002665 <dd>Truncate a constant to another type. The bit size of CST must be larger
2666 than the bit size of TYPE. Both types must be integers.</dd>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002667
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002668 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002669 <dd>Zero extend a constant to another type. The bit size of CST must be
Duncan Sands28afd432010-07-13 12:06:14 +00002670 smaller than the bit size of TYPE. Both types must be integers.</dd>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00002671
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002672 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002673 <dd>Sign extend a constant to another type. The bit size of CST must be
Duncan Sands28afd432010-07-13 12:06:14 +00002674 smaller than the bit size of TYPE. Both types must be integers.</dd>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00002675
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002676 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002677 <dd>Truncate a floating point constant to another floating point type. The
2678 size of CST must be larger than the size of TYPE. Both types must be
2679 floating point.</dd>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00002680
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002681 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002682 <dd>Floating point extend a constant to another type. The size of CST must be
2683 smaller or equal to the size of TYPE. Both types must be floating
2684 point.</dd>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00002685
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002686 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00002687 <dd>Convert a floating point constant to the corresponding unsigned integer
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002688 constant. TYPE must be a scalar or vector integer type. CST must be of
2689 scalar or vector floating point type. Both CST and TYPE must be scalars,
2690 or vectors of the same number of elements. If the value won't fit in the
2691 integer type, the results are undefined.</dd>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00002692
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002693 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00002694 <dd>Convert a floating point constant to the corresponding signed integer
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002695 constant. TYPE must be a scalar or vector integer type. CST must be of
2696 scalar or vector floating point type. Both CST and TYPE must be scalars,
2697 or vectors of the same number of elements. If the value won't fit in the
2698 integer type, the results are undefined.</dd>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00002699
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002700 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00002701 <dd>Convert an unsigned integer constant to the corresponding floating point
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002702 constant. TYPE must be a scalar or vector floating point type. CST must be
2703 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2704 vectors of the same number of elements. If the value won't fit in the
2705 floating point type, the results are undefined.</dd>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00002706
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002707 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00002708 <dd>Convert a signed integer constant to the corresponding floating point
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002709 constant. TYPE must be a scalar or vector floating point type. CST must be
2710 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2711 vectors of the same number of elements. If the value won't fit in the
2712 floating point type, the results are undefined.</dd>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00002713
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002714 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
Reid Spencer5c0ef472006-11-11 23:08:07 +00002715 <dd>Convert a pointer typed constant to the corresponding integer constant
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002716 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2717 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2718 make it fit in <tt>TYPE</tt>.</dd>
Reid Spencer5c0ef472006-11-11 23:08:07 +00002719
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002720 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002721 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2722 type. CST must be of integer type. The CST value is zero extended,
2723 truncated, or unchanged to make it fit in a pointer size. This one is
2724 <i>really</i> dangerous!</dd>
Reid Spencer5c0ef472006-11-11 23:08:07 +00002725
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002726 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
Chris Lattner03bbad62009-02-28 18:27:03 +00002727 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2728 are the same as those for the <a href="#i_bitcast">bitcast
2729 instruction</a>.</dd>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002730
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002731 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2732 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002733 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002734 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2735 instruction, the index list may have zero or more indexes, which are
2736 required to make sense for the type of "CSTPTR".</dd>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002737
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002738 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002739 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
Reid Spencer01c42592006-12-04 19:23:19 +00002740
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002741 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
Reid Spencer01c42592006-12-04 19:23:19 +00002742 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2743
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002744 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
Reid Spencer01c42592006-12-04 19:23:19 +00002745 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
Robert Bocchino9fbe1452006-01-10 19:31:34 +00002746
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002747 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002748 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2749 constants.</dd>
Robert Bocchino9fbe1452006-01-10 19:31:34 +00002750
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002751 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002752 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2753 constants.</dd>
Chris Lattnerc1989542006-04-08 00:13:41 +00002754
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002755 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002756 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2757 constants.</dd>
Chris Lattnerc1989542006-04-08 00:13:41 +00002758
Nick Lewycky9e130ce2010-05-29 06:44:15 +00002759 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2760 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2761 constants. The index list is interpreted in a similar manner as indices in
2762 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2763 index value must be specified.</dd>
2764
2765 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2766 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2767 constants. The index list is interpreted in a similar manner as indices in
2768 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2769 index value must be specified.</dd>
2770
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00002771 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002772 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2773 be any of the <a href="#binaryops">binary</a>
2774 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2775 on operands are the same as those for the corresponding instruction
2776 (e.g. no bitwise operations on floating point values are allowed).</dd>
Chris Lattnerc3f59762004-12-09 17:30:23 +00002777</dl>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002778
Chris Lattnerc3f59762004-12-09 17:30:23 +00002779</div>
Chris Lattner9ee5d222004-03-08 16:49:10 +00002780
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002781</div>
2782
Chris Lattner00950542001-06-06 20:29:01 +00002783<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002784<h2><a name="othervalues">Other Values</a></h2>
Chris Lattnere87d6532006-01-25 23:47:57 +00002785<!-- *********************************************************************** -->
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002786<div>
Chris Lattnere87d6532006-01-25 23:47:57 +00002787<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002788<h3>
Chris Lattnere87d6532006-01-25 23:47:57 +00002789<a name="inlineasm">Inline Assembler Expressions</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002790</h3>
Chris Lattnere87d6532006-01-25 23:47:57 +00002791
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002792<div>
Chris Lattnere87d6532006-01-25 23:47:57 +00002793
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002794<p>LLVM supports inline assembler expressions (as opposed
2795 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2796 a special value. This value represents the inline assembler as a string
2797 (containing the instructions to emit), a list of operand constraints (stored
Dale Johannesen09fed252009-10-13 21:56:55 +00002798 as a string), a flag that indicates whether or not the inline asm
Dale Johannesen8ba2d5b2009-10-21 23:28:00 +00002799 expression has side effects, and a flag indicating whether the function
2800 containing the asm needs to align its stack conservatively. An example
2801 inline assembler expression is:</p>
Chris Lattnere87d6532006-01-25 23:47:57 +00002802
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002803<pre class="doc_code">
Bill Wendling2f7a8b02007-05-29 09:04:49 +00002804i32 (i32) asm "bswap $0", "=r,r"
Chris Lattnere87d6532006-01-25 23:47:57 +00002805</pre>
2806
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002807<p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2808 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2809 have:</p>
Chris Lattnere87d6532006-01-25 23:47:57 +00002810
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002811<pre class="doc_code">
Bill Wendling2f7a8b02007-05-29 09:04:49 +00002812%X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
Chris Lattnere87d6532006-01-25 23:47:57 +00002813</pre>
2814
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002815<p>Inline asms with side effects not visible in the constraint list must be
2816 marked as having side effects. This is done through the use of the
2817 '<tt>sideeffect</tt>' keyword, like so:</p>
Chris Lattnere87d6532006-01-25 23:47:57 +00002818
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002819<pre class="doc_code">
Bill Wendling2f7a8b02007-05-29 09:04:49 +00002820call void asm sideeffect "eieio", ""()
Chris Lattnere87d6532006-01-25 23:47:57 +00002821</pre>
2822
Dale Johannesen8ba2d5b2009-10-21 23:28:00 +00002823<p>In some cases inline asms will contain code that will not work unless the
2824 stack is aligned in some way, such as calls or SSE instructions on x86,
2825 yet will not contain code that does that alignment within the asm.
2826 The compiler should make conservative assumptions about what the asm might
2827 contain and should generate its usual stack alignment code in the prologue
2828 if the '<tt>alignstack</tt>' keyword is present:</p>
Dale Johannesen09fed252009-10-13 21:56:55 +00002829
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002830<pre class="doc_code">
Dale Johannesen8ba2d5b2009-10-21 23:28:00 +00002831call void asm alignstack "eieio", ""()
Dale Johannesen09fed252009-10-13 21:56:55 +00002832</pre>
Dale Johannesen09fed252009-10-13 21:56:55 +00002833
2834<p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2835 first.</p>
2836
Chris Lattnere87d6532006-01-25 23:47:57 +00002837<p>TODO: The format of the asm and constraints string still need to be
Bill Wendlinge910b4c2009-07-20 02:29:24 +00002838 documented here. Constraints on what can be done (e.g. duplication, moving,
2839 etc need to be documented). This is probably best done by reference to
2840 another document that covers inline asm from a holistic perspective.</p>
Chris Lattnercf9a4152010-04-07 05:38:05 +00002841
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002842<h4>
Chris Lattnercf9a4152010-04-07 05:38:05 +00002843<a name="inlineasm_md">Inline Asm Metadata</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002844</h4>
Chris Lattnercf9a4152010-04-07 05:38:05 +00002845
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002846<div>
Chris Lattnercf9a4152010-04-07 05:38:05 +00002847
2848<p>The call instructions that wrap inline asm nodes may have a "!srcloc" MDNode
Chris Lattnerce1b9ad2010-11-17 08:20:42 +00002849 attached to it that contains a list of constant integers. If present, the
2850 code generator will use the integer as the location cookie value when report
Chris Lattnercf9a4152010-04-07 05:38:05 +00002851 errors through the LLVMContext error reporting mechanisms. This allows a
Dan Gohman1c70c002010-04-28 00:36:01 +00002852 front-end to correlate backend errors that occur with inline asm back to the
Chris Lattnercf9a4152010-04-07 05:38:05 +00002853 source code that produced it. For example:</p>
2854
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002855<pre class="doc_code">
Chris Lattnercf9a4152010-04-07 05:38:05 +00002856call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2857...
2858!42 = !{ i32 1234567 }
2859</pre>
Chris Lattnercf9a4152010-04-07 05:38:05 +00002860
2861<p>It is up to the front-end to make sense of the magic numbers it places in the
Chris Lattnerce1b9ad2010-11-17 08:20:42 +00002862 IR. If the MDNode contains multiple constants, the code generator will use
2863 the one that corresponds to the line of the asm that the error occurs on.</p>
Chris Lattnere87d6532006-01-25 23:47:57 +00002864
2865</div>
2866
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002867</div>
2868
Chris Lattnere6a5ddd2010-01-15 21:50:19 +00002869<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002870<h3>
2871 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2872</h3>
Chris Lattnere6a5ddd2010-01-15 21:50:19 +00002873
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002874<div>
Chris Lattnere6a5ddd2010-01-15 21:50:19 +00002875
2876<p>LLVM IR allows metadata to be attached to instructions in the program that
2877 can convey extra information about the code to the optimizers and code
2878 generator. One example application of metadata is source-level debug
2879 information. There are two metadata primitives: strings and nodes. All
2880 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2881 preceding exclamation point ('<tt>!</tt>').</p>
2882
2883<p>A metadata string is a string surrounded by double quotes. It can contain
2884 any character by escaping non-printable characters with "\xx" where "xx" is
2885 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2886
2887<p>Metadata nodes are represented with notation similar to structure constants
2888 (a comma separated list of elements, surrounded by braces and preceded by an
2889 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2890 10}</tt>". Metadata nodes can have any values as their operand.</p>
2891
2892<p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2893 metadata nodes, which can be looked up in the module symbol table. For
2894 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2895
Devang Patele1d50cd2010-03-04 23:44:48 +00002896<p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002897 function is using two metadata arguments.</p>
Devang Patele1d50cd2010-03-04 23:44:48 +00002898
Bill Wendling9ff5de92011-03-02 02:17:11 +00002899<div class="doc_code">
2900<pre>
2901call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2902</pre>
2903</div>
Devang Patele1d50cd2010-03-04 23:44:48 +00002904
2905<p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00002906 attached with <tt>add</tt> instruction using <tt>!dbg</tt> identifier.</p>
Devang Patele1d50cd2010-03-04 23:44:48 +00002907
Bill Wendling9ff5de92011-03-02 02:17:11 +00002908<div class="doc_code">
2909<pre>
2910%indvar.next = add i64 %indvar, 1, !dbg !21
2911</pre>
2912</div>
2913
Chris Lattnere6a5ddd2010-01-15 21:50:19 +00002914</div>
2915
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002916</div>
Chris Lattner857755c2009-07-20 05:55:19 +00002917
2918<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002919<h2>
Chris Lattner857755c2009-07-20 05:55:19 +00002920 <a name="intrinsic_globals">Intrinsic Global Variables</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002921</h2>
Chris Lattner857755c2009-07-20 05:55:19 +00002922<!-- *********************************************************************** -->
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002923<div>
Chris Lattner857755c2009-07-20 05:55:19 +00002924<p>LLVM has a number of "magic" global variables that contain data that affect
2925code generation or other IR semantics. These are documented here. All globals
Chris Lattner401e10c2009-07-20 06:14:25 +00002926of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2927section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2928by LLVM.</p>
Chris Lattner857755c2009-07-20 05:55:19 +00002929
2930<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002931<h3>
Chris Lattner857755c2009-07-20 05:55:19 +00002932<a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002933</h3>
Chris Lattner857755c2009-07-20 05:55:19 +00002934
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002935<div>
Chris Lattner857755c2009-07-20 05:55:19 +00002936
2937<p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2938href="#linkage_appending">appending linkage</a>. This array contains a list of
2939pointers to global variables and functions which may optionally have a pointer
2940cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2941
2942<pre>
2943 @X = global i8 4
2944 @Y = global i32 123
2945
2946 @llvm.used = appending global [2 x i8*] [
2947 i8* @X,
2948 i8* bitcast (i32* @Y to i8*)
2949 ], section "llvm.metadata"
2950</pre>
2951
2952<p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2953compiler, assembler, and linker are required to treat the symbol as if there is
2954a reference to the global that it cannot see. For example, if a variable has
2955internal linkage and no references other than that from the <tt>@llvm.used</tt>
2956list, it cannot be deleted. This is commonly used to represent references from
2957inline asms and other things the compiler cannot "see", and corresponds to
2958"attribute((used))" in GNU C.</p>
2959
2960<p>On some targets, the code generator must emit a directive to the assembler or
2961object file to prevent the assembler and linker from molesting the symbol.</p>
2962
2963</div>
2964
2965<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002966<h3>
2967 <a name="intg_compiler_used">
2968 The '<tt>llvm.compiler.used</tt>' Global Variable
2969 </a>
2970</h3>
Chris Lattner401e10c2009-07-20 06:14:25 +00002971
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002972<div>
Chris Lattner401e10c2009-07-20 06:14:25 +00002973
2974<p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2975<tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2976touching the symbol. On targets that support it, this allows an intelligent
2977linker to optimize references to the symbol without being impeded as it would be
2978by <tt>@llvm.used</tt>.</p>
2979
2980<p>This is a rare construct that should only be used in rare circumstances, and
2981should not be exposed to source languages.</p>
2982
2983</div>
2984
2985<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002986<h3>
Chris Lattner857755c2009-07-20 05:55:19 +00002987<a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00002988</h3>
Chris Lattner857755c2009-07-20 05:55:19 +00002989
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00002990<div>
David Chisnalle31e9962010-04-30 19:23:49 +00002991<pre>
2992%0 = type { i32, void ()* }
David Chisnall27195a52010-04-30 19:27:35 +00002993@llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
David Chisnalle31e9962010-04-30 19:23:49 +00002994</pre>
2995<p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor functions and associated priorities. The functions referenced by this array will be called in ascending order of priority (i.e. lowest first) when the module is loaded. The order of functions with the same priority is not defined.
2996</p>
Chris Lattner857755c2009-07-20 05:55:19 +00002997
2998</div>
2999
3000<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003001<h3>
Chris Lattner857755c2009-07-20 05:55:19 +00003002<a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003003</h3>
Chris Lattner857755c2009-07-20 05:55:19 +00003004
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003005<div>
David Chisnalle31e9962010-04-30 19:23:49 +00003006<pre>
3007%0 = type { i32, void ()* }
David Chisnall27195a52010-04-30 19:27:35 +00003008@llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
David Chisnalle31e9962010-04-30 19:23:49 +00003009</pre>
Chris Lattner857755c2009-07-20 05:55:19 +00003010
David Chisnalle31e9962010-04-30 19:23:49 +00003011<p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions and associated priorities. The functions referenced by this array will be called in descending order of priority (i.e. highest first) when the module is loaded. The order of functions with the same priority is not defined.
3012</p>
Chris Lattner857755c2009-07-20 05:55:19 +00003013
3014</div>
3015
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003016</div>
Chris Lattner857755c2009-07-20 05:55:19 +00003017
Chris Lattnere87d6532006-01-25 23:47:57 +00003018<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003019<h2><a name="instref">Instruction Reference</a></h2>
Chris Lattner261efe92003-11-25 01:02:51 +00003020<!-- *********************************************************************** -->
Chris Lattnerc3f59762004-12-09 17:30:23 +00003021
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003022<div>
Chris Lattnerc3f59762004-12-09 17:30:23 +00003023
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003024<p>The LLVM instruction set consists of several different classifications of
3025 instructions: <a href="#terminators">terminator
3026 instructions</a>, <a href="#binaryops">binary instructions</a>,
3027 <a href="#bitwiseops">bitwise binary instructions</a>,
3028 <a href="#memoryops">memory instructions</a>, and
3029 <a href="#otherops">other instructions</a>.</p>
Chris Lattnerc3f59762004-12-09 17:30:23 +00003030
Chris Lattner00950542001-06-06 20:29:01 +00003031<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003032<h3>
3033 <a name="terminators">Terminator Instructions</a>
3034</h3>
Chris Lattnerc3f59762004-12-09 17:30:23 +00003035
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003036<div>
Chris Lattnerc3f59762004-12-09 17:30:23 +00003037
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003038<p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3039 in a program ends with a "Terminator" instruction, which indicates which
3040 block should be executed after the current block is finished. These
3041 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3042 control flow, not values (the one exception being the
3043 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3044
Chris Lattner6445ecb2011-08-02 20:29:13 +00003045<p>The terminator instructions are:
3046 '<a href="#i_ret"><tt>ret</tt></a>',
3047 '<a href="#i_br"><tt>br</tt></a>',
3048 '<a href="#i_switch"><tt>switch</tt></a>',
3049 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3050 '<a href="#i_invoke"><tt>invoke</tt></a>',
3051 '<a href="#i_unwind"><tt>unwind</tt></a>',
3052 '<a href="#i_resume"><tt>resume</tt></a>', and
3053 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
Chris Lattnerc3f59762004-12-09 17:30:23 +00003054
Chris Lattner00950542001-06-06 20:29:01 +00003055<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003056<h4>
3057 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3058</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003059
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003060<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003061
Chris Lattner00950542001-06-06 20:29:01 +00003062<h5>Syntax:</h5>
Dan Gohmanb1e6b962008-10-04 19:00:07 +00003063<pre>
3064 ret &lt;type&gt; &lt;value&gt; <i>; Return a value from a non-void function</i>
Chris Lattner7faa8832002-04-14 06:13:44 +00003065 ret void <i>; Return from void function</i>
Chris Lattner00950542001-06-06 20:29:01 +00003066</pre>
Chris Lattnerf4cde4e2008-04-23 04:59:35 +00003067
Chris Lattner00950542001-06-06 20:29:01 +00003068<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003069<p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3070 a value) from a function back to the caller.</p>
3071
3072<p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3073 value and then causes control flow, and one that just causes control flow to
3074 occur.</p>
Chris Lattnerf4cde4e2008-04-23 04:59:35 +00003075
Chris Lattner00950542001-06-06 20:29:01 +00003076<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003077<p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3078 return value. The type of the return value must be a
3079 '<a href="#t_firstclass">first class</a>' type.</p>
Dan Gohmanb1e6b962008-10-04 19:00:07 +00003080
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003081<p>A function is not <a href="#wellformed">well formed</a> if it it has a
3082 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3083 value or a return value with a type that does not match its type, or if it
3084 has a void return type and contains a '<tt>ret</tt>' instruction with a
3085 return value.</p>
Chris Lattnerf4cde4e2008-04-23 04:59:35 +00003086
Chris Lattner00950542001-06-06 20:29:01 +00003087<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003088<p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3089 the calling function's context. If the caller is a
3090 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3091 instruction after the call. If the caller was an
3092 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3093 the beginning of the "normal" destination block. If the instruction returns
3094 a value, that value shall set the call or invoke instruction's return
3095 value.</p>
Chris Lattnerf4cde4e2008-04-23 04:59:35 +00003096
Chris Lattner00950542001-06-06 20:29:01 +00003097<h5>Example:</h5>
Chris Lattnerf4cde4e2008-04-23 04:59:35 +00003098<pre>
3099 ret i32 5 <i>; Return an integer value of 5</i>
Chris Lattner7faa8832002-04-14 06:13:44 +00003100 ret void <i>; Return from a void function</i>
Bill Wendling0a4bbbf2009-02-28 22:12:54 +00003101 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
Chris Lattner00950542001-06-06 20:29:01 +00003102</pre>
Dan Gohmanf3e60bd2009-01-12 23:12:39 +00003103
Misha Brukman9d0919f2003-11-08 01:05:38 +00003104</div>
Chris Lattner00950542001-06-06 20:29:01 +00003105<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003106<h4>
3107 <a name="i_br">'<tt>br</tt>' Instruction</a>
3108</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003109
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003110<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003111
Chris Lattner00950542001-06-06 20:29:01 +00003112<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003113<pre>
Bill Wendlingb3aa4712011-07-26 10:41:15 +00003114 br i1 &lt;cond&gt;, label &lt;iftrue&gt;, label &lt;iffalse&gt;
3115 br label &lt;dest&gt; <i>; Unconditional branch</i>
Chris Lattner00950542001-06-06 20:29:01 +00003116</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003117
Chris Lattner00950542001-06-06 20:29:01 +00003118<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003119<p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3120 different basic block in the current function. There are two forms of this
3121 instruction, corresponding to a conditional branch and an unconditional
3122 branch.</p>
3123
Chris Lattner00950542001-06-06 20:29:01 +00003124<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003125<p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3126 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3127 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3128 target.</p>
3129
Chris Lattner00950542001-06-06 20:29:01 +00003130<h5>Semantics:</h5>
Reid Spencerc78f3372007-01-12 03:35:51 +00003131<p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003132 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3133 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3134 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3135
Chris Lattner00950542001-06-06 20:29:01 +00003136<h5>Example:</h5>
Bill Wendlingc39e3e02009-07-20 02:39:26 +00003137<pre>
3138Test:
3139 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3140 br i1 %cond, label %IfEqual, label %IfUnequal
3141IfEqual:
3142 <a href="#i_ret">ret</a> i32 1
3143IfUnequal:
3144 <a href="#i_ret">ret</a> i32 0
3145</pre>
3146
Misha Brukman9d0919f2003-11-08 01:05:38 +00003147</div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003148
Chris Lattner00950542001-06-06 20:29:01 +00003149<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003150<h4>
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003151 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003152</h4>
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003153
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003154<div>
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003155
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003156<h5>Syntax:</h5>
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003157<pre>
3158 switch &lt;intty&gt; &lt;value&gt;, label &lt;defaultdest&gt; [ &lt;intty&gt; &lt;val&gt;, label &lt;dest&gt; ... ]
3159</pre>
3160
Chris Lattner00950542001-06-06 20:29:01 +00003161<h5>Overview:</h5>
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003162<p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003163 several different places. It is a generalization of the '<tt>br</tt>'
3164 instruction, allowing a branch to occur to one of many possible
3165 destinations.</p>
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003166
Chris Lattner00950542001-06-06 20:29:01 +00003167<h5>Arguments:</h5>
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003168<p>The '<tt>switch</tt>' instruction uses three parameters: an integer
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003169 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3170 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3171 The table is not allowed to contain duplicate constant entries.</p>
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003172
Chris Lattner00950542001-06-06 20:29:01 +00003173<h5>Semantics:</h5>
Chris Lattner261efe92003-11-25 01:02:51 +00003174<p>The <tt>switch</tt> instruction specifies a table of values and
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003175 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3176 is searched for the given value. If the value is found, control flow is
Benjamin Kramer8040cd32009-10-12 14:46:08 +00003177 transferred to the corresponding destination; otherwise, control flow is
3178 transferred to the default destination.</p>
Chris Lattner00950542001-06-06 20:29:01 +00003179
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003180<h5>Implementation:</h5>
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003181<p>Depending on properties of the target machine and the particular
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003182 <tt>switch</tt> instruction, this instruction may be code generated in
3183 different ways. For example, it could be generated as a series of chained
3184 conditional branches or with a lookup table.</p>
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003185
3186<h5>Example:</h5>
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003187<pre>
3188 <i>; Emulate a conditional br instruction</i>
Reid Spencerc78f3372007-01-12 03:35:51 +00003189 %Val = <a href="#i_zext">zext</a> i1 %value to i32
Dan Gohman2a08c532009-01-04 23:44:43 +00003190 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003191
3192 <i>; Emulate an unconditional br instruction</i>
Reid Spencerca86e162006-12-31 07:07:53 +00003193 switch i32 0, label %dest [ ]
Chris Lattnerc88c17b2004-02-24 04:54:45 +00003194
3195 <i>; Implement a jump table:</i>
Dan Gohman2a08c532009-01-04 23:44:43 +00003196 switch i32 %val, label %otherwise [ i32 0, label %onzero
3197 i32 1, label %onone
3198 i32 2, label %ontwo ]
Chris Lattner00950542001-06-06 20:29:01 +00003199</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003200
Misha Brukman9d0919f2003-11-08 01:05:38 +00003201</div>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003202
Chris Lattnerf9be95f2009-10-27 19:13:16 +00003203
3204<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003205<h4>
Chris Lattnerab21db72009-10-28 00:19:10 +00003206 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003207</h4>
Chris Lattnerf9be95f2009-10-27 19:13:16 +00003208
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003209<div>
Chris Lattnerf9be95f2009-10-27 19:13:16 +00003210
3211<h5>Syntax:</h5>
3212<pre>
Chris Lattnerab21db72009-10-28 00:19:10 +00003213 indirectbr &lt;somety&gt;* &lt;address&gt;, [ label &lt;dest1&gt;, label &lt;dest2&gt;, ... ]
Chris Lattnerf9be95f2009-10-27 19:13:16 +00003214</pre>
3215
3216<h5>Overview:</h5>
3217
Chris Lattnerab21db72009-10-28 00:19:10 +00003218<p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
Chris Lattnerf9be95f2009-10-27 19:13:16 +00003219 within the current function, whose address is specified by
Chris Lattnerc6f44362009-10-27 21:01:34 +00003220 "<tt>address</tt>". Address must be derived from a <a
3221 href="#blockaddress">blockaddress</a> constant.</p>
Chris Lattnerf9be95f2009-10-27 19:13:16 +00003222
3223<h5>Arguments:</h5>
3224
3225<p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3226 rest of the arguments indicate the full set of possible destinations that the
3227 address may point to. Blocks are allowed to occur multiple times in the
3228 destination list, though this isn't particularly useful.</p>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00003229
Chris Lattnerf9be95f2009-10-27 19:13:16 +00003230<p>This destination list is required so that dataflow analysis has an accurate
3231 understanding of the CFG.</p>
3232
3233<h5>Semantics:</h5>
3234
3235<p>Control transfers to the block specified in the address argument. All
3236 possible destination blocks must be listed in the label list, otherwise this
3237 instruction has undefined behavior. This implies that jumps to labels
3238 defined in other functions have undefined behavior as well.</p>
3239
3240<h5>Implementation:</h5>
3241
3242<p>This is typically implemented with a jump through a register.</p>
3243
3244<h5>Example:</h5>
3245<pre>
Chris Lattnerab21db72009-10-28 00:19:10 +00003246 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
Chris Lattnerf9be95f2009-10-27 19:13:16 +00003247</pre>
3248
3249</div>
3250
3251
Chris Lattner00950542001-06-06 20:29:01 +00003252<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003253<h4>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003254 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003255</h4>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003256
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003257<div>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003258
Chris Lattner00950542001-06-06 20:29:01 +00003259<h5>Syntax:</h5>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003260<pre>
Devang Patel307e8ab2008-10-07 17:48:33 +00003261 &lt;result&gt; = invoke [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] &lt;ptr to function ty&gt; &lt;function ptr val&gt;(&lt;function args&gt;) [<a href="#fnattrs">fn attrs</a>]
Chris Lattner76b8a332006-05-14 18:23:06 +00003262 to label &lt;normal label&gt; unwind label &lt;exception label&gt;
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003263</pre>
3264
Chris Lattner6536cfe2002-05-06 22:08:29 +00003265<h5>Overview:</h5>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003266<p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003267 function, with the possibility of control flow transfer to either the
3268 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3269 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3270 control flow will return to the "normal" label. If the callee (or any
3271 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
3272 instruction, control is interrupted and continued at the dynamically nearest
3273 "exception" label.</p>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003274
Bill Wendlingf78faf82011-08-02 21:52:38 +00003275<p>The '<tt>exception</tt>' label is a
3276 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3277 exception. As such, '<tt>exception</tt>' label is required to have the
3278 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3279 the information about about the behavior of the program after unwinding
3280 happens, as its first non-PHI instruction. The restrictions on the
3281 "<tt>landingpad</tt>" instruction's tightly couples it to the
3282 "<tt>invoke</tt>" instruction, so that the important information contained
3283 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3284 code motion.</p>
3285
Chris Lattner00950542001-06-06 20:29:01 +00003286<h5>Arguments:</h5>
Misha Brukman9d0919f2003-11-08 01:05:38 +00003287<p>This instruction requires several arguments:</p>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003288
Chris Lattner00950542001-06-06 20:29:01 +00003289<ol>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003290 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3291 convention</a> the call should use. If none is specified, the call
3292 defaults to using C calling conventions.</li>
Devang Patelf642f472008-10-06 18:50:38 +00003293
3294 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003295 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3296 '<tt>inreg</tt>' attributes are valid here.</li>
Devang Patelf642f472008-10-06 18:50:38 +00003297
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003298 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003299 function value being invoked. In most cases, this is a direct function
3300 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3301 off an arbitrary pointer to function value.</li>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003302
3303 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003304 function to be invoked. </li>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003305
3306 <li>'<tt>function args</tt>': argument list whose types match the function
Chris Lattner0724fbd2010-03-02 06:36:51 +00003307 signature argument types and parameter attributes. All arguments must be
3308 of <a href="#t_firstclass">first class</a> type. If the function
3309 signature indicates the function accepts a variable number of arguments,
3310 the extra arguments can be specified.</li>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003311
3312 <li>'<tt>normal label</tt>': the label reached when the called function
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003313 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003314
3315 <li>'<tt>exception label</tt>': the label reached when a callee returns with
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003316 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003317
Devang Patel307e8ab2008-10-07 17:48:33 +00003318 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003319 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3320 '<tt>readnone</tt>' attributes are valid here.</li>
Chris Lattner00950542001-06-06 20:29:01 +00003321</ol>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003322
Chris Lattner00950542001-06-06 20:29:01 +00003323<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003324<p>This instruction is designed to operate as a standard
3325 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3326 primary difference is that it establishes an association with a label, which
3327 is used by the runtime library to unwind the stack.</p>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003328
3329<p>This instruction is used in languages with destructors to ensure that proper
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003330 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3331 exception. Additionally, this is important for implementation of
3332 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003333
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003334<p>For the purposes of the SSA form, the definition of the value returned by the
3335 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3336 block to the "normal" label. If the callee unwinds then no return value is
3337 available.</p>
Dan Gohmanf96a4992009-05-22 21:47:08 +00003338
Chris Lattnerdf7a6802010-01-15 18:08:37 +00003339<p>Note that the code generator does not yet completely support unwind, and
3340that the invoke/unwind semantics are likely to change in future versions.</p>
3341
Chris Lattner00950542001-06-06 20:29:01 +00003342<h5>Example:</h5>
Chris Lattnerbad10ee2005-05-06 22:57:40 +00003343<pre>
Nick Lewyckyd703f652008-03-16 07:18:12 +00003344 %retval = invoke i32 @Test(i32 15) to label %Continue
Jeff Cohen6f1cc772007-04-22 01:17:39 +00003345 unwind label %TestCleanup <i>; {i32}:retval set</i>
Nick Lewyckyd703f652008-03-16 07:18:12 +00003346 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
Jeff Cohen6f1cc772007-04-22 01:17:39 +00003347 unwind label %TestCleanup <i>; {i32}:retval set</i>
Chris Lattner00950542001-06-06 20:29:01 +00003348</pre>
Chris Lattner35eca582004-10-16 18:04:13 +00003349
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003350</div>
Chris Lattner35eca582004-10-16 18:04:13 +00003351
Chris Lattner27f71f22003-09-03 00:41:47 +00003352<!-- _______________________________________________________________________ -->
Chris Lattner35eca582004-10-16 18:04:13 +00003353
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003354<h4>
3355 <a name="i_unwind">'<tt>unwind</tt>' Instruction</a>
3356</h4>
Chris Lattner35eca582004-10-16 18:04:13 +00003357
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003358<div>
Chris Lattner35eca582004-10-16 18:04:13 +00003359
Chris Lattner27f71f22003-09-03 00:41:47 +00003360<h5>Syntax:</h5>
Chris Lattner35eca582004-10-16 18:04:13 +00003361<pre>
3362 unwind
3363</pre>
3364
Chris Lattner27f71f22003-09-03 00:41:47 +00003365<h5>Overview:</h5>
Chris Lattner35eca582004-10-16 18:04:13 +00003366<p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003367 at the first callee in the dynamic call stack which used
3368 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
3369 This is primarily used to implement exception handling.</p>
Chris Lattner35eca582004-10-16 18:04:13 +00003370
Chris Lattner27f71f22003-09-03 00:41:47 +00003371<h5>Semantics:</h5>
Chris Lattner72ed2002008-04-19 21:01:16 +00003372<p>The '<tt>unwind</tt>' instruction causes execution of the current function to
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003373 immediately halt. The dynamic call stack is then searched for the
3374 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
3375 Once found, execution continues at the "exceptional" destination block
3376 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
3377 instruction in the dynamic call chain, undefined behavior results.</p>
3378
Chris Lattnerdf7a6802010-01-15 18:08:37 +00003379<p>Note that the code generator does not yet completely support unwind, and
3380that the invoke/unwind semantics are likely to change in future versions.</p>
3381
Misha Brukman9d0919f2003-11-08 01:05:38 +00003382</div>
Chris Lattner35eca582004-10-16 18:04:13 +00003383
Bill Wendlingdccc03b2011-07-31 06:30:59 +00003384 <!-- _______________________________________________________________________ -->
3385
3386<h4>
3387 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3388</h4>
3389
3390<div>
3391
3392<h5>Syntax:</h5>
3393<pre>
3394 resume &lt;type&gt; &lt;value&gt;
3395</pre>
3396
3397<h5>Overview:</h5>
3398<p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3399 successors.</p>
3400
3401<h5>Arguments:</h5>
Bill Wendlingf78faf82011-08-02 21:52:38 +00003402<p>The '<tt>resume</tt>' instruction requires one argument, which must have the
Bill Wendlinge4ad50b2011-08-03 18:37:32 +00003403 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3404 function.</p>
Bill Wendlingdccc03b2011-07-31 06:30:59 +00003405
3406<h5>Semantics:</h5>
3407<p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3408 (in-flight) exception whose unwinding was interrupted with
Bill Wendlingf78faf82011-08-02 21:52:38 +00003409 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
Bill Wendlingdccc03b2011-07-31 06:30:59 +00003410
3411<h5>Example:</h5>
3412<pre>
Bill Wendlingf78faf82011-08-02 21:52:38 +00003413 resume { i8*, i32 } %exn
Bill Wendlingdccc03b2011-07-31 06:30:59 +00003414</pre>
3415
3416</div>
3417
Chris Lattner35eca582004-10-16 18:04:13 +00003418<!-- _______________________________________________________________________ -->
3419
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003420<h4>
3421 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3422</h4>
Chris Lattner35eca582004-10-16 18:04:13 +00003423
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003424<div>
Chris Lattner35eca582004-10-16 18:04:13 +00003425
3426<h5>Syntax:</h5>
3427<pre>
3428 unreachable
3429</pre>
3430
3431<h5>Overview:</h5>
Chris Lattner35eca582004-10-16 18:04:13 +00003432<p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003433 instruction is used to inform the optimizer that a particular portion of the
3434 code is not reachable. This can be used to indicate that the code after a
3435 no-return function cannot be reached, and other facts.</p>
Chris Lattner35eca582004-10-16 18:04:13 +00003436
3437<h5>Semantics:</h5>
Chris Lattner35eca582004-10-16 18:04:13 +00003438<p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003439
Chris Lattner35eca582004-10-16 18:04:13 +00003440</div>
3441
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003442</div>
3443
Chris Lattner00950542001-06-06 20:29:01 +00003444<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003445<h3>
3446 <a name="binaryops">Binary Operations</a>
3447</h3>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003448
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003449<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003450
3451<p>Binary operators are used to do most of the computation in a program. They
3452 require two operands of the same type, execute an operation on them, and
3453 produce a single value. The operands might represent multiple data, as is
3454 the case with the <a href="#t_vector">vector</a> data type. The result value
3455 has the same type as its operands.</p>
3456
Misha Brukman9d0919f2003-11-08 01:05:38 +00003457<p>There are several different binary operators:</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003458
Chris Lattner00950542001-06-06 20:29:01 +00003459<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003460<h4>
Chris Lattner5568e942008-05-20 20:48:21 +00003461 <a name="i_add">'<tt>add</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003462</h4>
Chris Lattner5568e942008-05-20 20:48:21 +00003463
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003464<div>
Chris Lattner5568e942008-05-20 20:48:21 +00003465
Chris Lattner00950542001-06-06 20:29:01 +00003466<h5>Syntax:</h5>
Chris Lattner5568e942008-05-20 20:48:21 +00003467<pre>
Dan Gohmancbb38f22009-07-20 22:41:19 +00003468 &lt;result&gt; = add &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanfdfca792009-09-02 17:31:42 +00003469 &lt;result&gt; = add nuw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
3470 &lt;result&gt; = add nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
3471 &lt;result&gt; = add nuw nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Chris Lattner00950542001-06-06 20:29:01 +00003472</pre>
Chris Lattner5568e942008-05-20 20:48:21 +00003473
Chris Lattner00950542001-06-06 20:29:01 +00003474<h5>Overview:</h5>
Misha Brukman9d0919f2003-11-08 01:05:38 +00003475<p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003476
Chris Lattner00950542001-06-06 20:29:01 +00003477<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003478<p>The two arguments to the '<tt>add</tt>' instruction must
3479 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3480 integer values. Both arguments must have identical types.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003481
Chris Lattner00950542001-06-06 20:29:01 +00003482<h5>Semantics:</h5>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003483<p>The value produced is the integer sum of the two operands.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003484
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003485<p>If the sum has unsigned overflow, the result returned is the mathematical
3486 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003487
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003488<p>Because LLVM integers use a two's complement representation, this instruction
3489 is appropriate for both signed and unsigned integers.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003490
Dan Gohman08d012e2009-07-22 22:44:56 +00003491<p><tt>nuw</tt> and <tt>nsw</tt> stand for &quot;No Unsigned Wrap&quot;
3492 and &quot;No Signed Wrap&quot;, respectively. If the <tt>nuw</tt> and/or
3493 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
Dan Gohmanfff6c532010-04-22 23:14:21 +00003494 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3495 respectively, occurs.</p>
Dan Gohmancbb38f22009-07-20 22:41:19 +00003496
Chris Lattner00950542001-06-06 20:29:01 +00003497<h5>Example:</h5>
Chris Lattner5568e942008-05-20 20:48:21 +00003498<pre>
3499 &lt;result&gt; = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
Chris Lattner00950542001-06-06 20:29:01 +00003500</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003501
Misha Brukman9d0919f2003-11-08 01:05:38 +00003502</div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003503
Chris Lattner00950542001-06-06 20:29:01 +00003504<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003505<h4>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003506 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003507</h4>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003508
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003509<div>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003510
3511<h5>Syntax:</h5>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003512<pre>
3513 &lt;result&gt; = fadd &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
3514</pre>
3515
3516<h5>Overview:</h5>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003517<p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3518
3519<h5>Arguments:</h5>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003520<p>The two arguments to the '<tt>fadd</tt>' instruction must be
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003521 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3522 floating point values. Both arguments must have identical types.</p>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003523
3524<h5>Semantics:</h5>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003525<p>The value produced is the floating point sum of the two operands.</p>
3526
3527<h5>Example:</h5>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003528<pre>
3529 &lt;result&gt; = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3530</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003531
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003532</div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003533
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003534<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003535<h4>
Chris Lattner5568e942008-05-20 20:48:21 +00003536 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003537</h4>
Chris Lattner5568e942008-05-20 20:48:21 +00003538
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003539<div>
Chris Lattner5568e942008-05-20 20:48:21 +00003540
Chris Lattner00950542001-06-06 20:29:01 +00003541<h5>Syntax:</h5>
Chris Lattner5568e942008-05-20 20:48:21 +00003542<pre>
Dan Gohman08d012e2009-07-22 22:44:56 +00003543 &lt;result&gt; = sub &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanfdfca792009-09-02 17:31:42 +00003544 &lt;result&gt; = sub nuw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
3545 &lt;result&gt; = sub nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
3546 &lt;result&gt; = sub nuw nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Chris Lattner00950542001-06-06 20:29:01 +00003547</pre>
Chris Lattner5568e942008-05-20 20:48:21 +00003548
Chris Lattner00950542001-06-06 20:29:01 +00003549<h5>Overview:</h5>
Misha Brukman9d0919f2003-11-08 01:05:38 +00003550<p>The '<tt>sub</tt>' instruction returns the difference of its two
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003551 operands.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003552
3553<p>Note that the '<tt>sub</tt>' instruction is used to represent the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003554 '<tt>neg</tt>' instruction present in most other intermediate
3555 representations.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003556
Chris Lattner00950542001-06-06 20:29:01 +00003557<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003558<p>The two arguments to the '<tt>sub</tt>' instruction must
3559 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3560 integer values. Both arguments must have identical types.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003561
Chris Lattner00950542001-06-06 20:29:01 +00003562<h5>Semantics:</h5>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003563<p>The value produced is the integer difference of the two operands.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003564
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003565<p>If the difference has unsigned overflow, the result returned is the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003566 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3567 result.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003568
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003569<p>Because LLVM integers use a two's complement representation, this instruction
3570 is appropriate for both signed and unsigned integers.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003571
Dan Gohman08d012e2009-07-22 22:44:56 +00003572<p><tt>nuw</tt> and <tt>nsw</tt> stand for &quot;No Unsigned Wrap&quot;
3573 and &quot;No Signed Wrap&quot;, respectively. If the <tt>nuw</tt> and/or
3574 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
Dan Gohmanfff6c532010-04-22 23:14:21 +00003575 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3576 respectively, occurs.</p>
Dan Gohmancbb38f22009-07-20 22:41:19 +00003577
Chris Lattner00950542001-06-06 20:29:01 +00003578<h5>Example:</h5>
Bill Wendlingaac388b2007-05-29 09:42:13 +00003579<pre>
3580 &lt;result&gt; = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
Reid Spencerca86e162006-12-31 07:07:53 +00003581 &lt;result&gt; = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
Chris Lattner00950542001-06-06 20:29:01 +00003582</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003583
Misha Brukman9d0919f2003-11-08 01:05:38 +00003584</div>
Chris Lattner5568e942008-05-20 20:48:21 +00003585
Chris Lattner00950542001-06-06 20:29:01 +00003586<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003587<h4>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003588 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003589</h4>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003590
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003591<div>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003592
3593<h5>Syntax:</h5>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003594<pre>
3595 &lt;result&gt; = fsub &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
3596</pre>
3597
3598<h5>Overview:</h5>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003599<p>The '<tt>fsub</tt>' instruction returns the difference of its two
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003600 operands.</p>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003601
3602<p>Note that the '<tt>fsub</tt>' instruction is used to represent the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003603 '<tt>fneg</tt>' instruction present in most other intermediate
3604 representations.</p>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003605
3606<h5>Arguments:</h5>
Bill Wendlingd9fe2982009-07-20 02:32:41 +00003607<p>The two arguments to the '<tt>fsub</tt>' instruction must be
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003608 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3609 floating point values. Both arguments must have identical types.</p>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003610
3611<h5>Semantics:</h5>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003612<p>The value produced is the floating point difference of the two operands.</p>
3613
3614<h5>Example:</h5>
3615<pre>
3616 &lt;result&gt; = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3617 &lt;result&gt; = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3618</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003619
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003620</div>
3621
3622<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003623<h4>
Chris Lattner5568e942008-05-20 20:48:21 +00003624 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003625</h4>
Chris Lattner5568e942008-05-20 20:48:21 +00003626
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003627<div>
Chris Lattner5568e942008-05-20 20:48:21 +00003628
Chris Lattner00950542001-06-06 20:29:01 +00003629<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003630<pre>
Dan Gohman08d012e2009-07-22 22:44:56 +00003631 &lt;result&gt; = mul &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanfdfca792009-09-02 17:31:42 +00003632 &lt;result&gt; = mul nuw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
3633 &lt;result&gt; = mul nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
3634 &lt;result&gt; = mul nuw nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Chris Lattner00950542001-06-06 20:29:01 +00003635</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003636
Chris Lattner00950542001-06-06 20:29:01 +00003637<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003638<p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003639
Chris Lattner00950542001-06-06 20:29:01 +00003640<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003641<p>The two arguments to the '<tt>mul</tt>' instruction must
3642 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3643 integer values. Both arguments must have identical types.</p>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00003644
Chris Lattner00950542001-06-06 20:29:01 +00003645<h5>Semantics:</h5>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003646<p>The value produced is the integer product of the two operands.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003647
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003648<p>If the result of the multiplication has unsigned overflow, the result
3649 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3650 width of the result.</p>
3651
3652<p>Because LLVM integers use a two's complement representation, and the result
3653 is the same width as the operands, this instruction returns the correct
3654 result for both signed and unsigned integers. If a full product
3655 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3656 be sign-extended or zero-extended as appropriate to the width of the full
3657 product.</p>
3658
Dan Gohman08d012e2009-07-22 22:44:56 +00003659<p><tt>nuw</tt> and <tt>nsw</tt> stand for &quot;No Unsigned Wrap&quot;
3660 and &quot;No Signed Wrap&quot;, respectively. If the <tt>nuw</tt> and/or
3661 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
Dan Gohmanfff6c532010-04-22 23:14:21 +00003662 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3663 respectively, occurs.</p>
Dan Gohmancbb38f22009-07-20 22:41:19 +00003664
Chris Lattner00950542001-06-06 20:29:01 +00003665<h5>Example:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003666<pre>
3667 &lt;result&gt; = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
Chris Lattner00950542001-06-06 20:29:01 +00003668</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003669
Misha Brukman9d0919f2003-11-08 01:05:38 +00003670</div>
Chris Lattner5568e942008-05-20 20:48:21 +00003671
Chris Lattner00950542001-06-06 20:29:01 +00003672<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003673<h4>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003674 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003675</h4>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003676
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003677<div>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003678
3679<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003680<pre>
3681 &lt;result&gt; = fmul &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003682</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003683
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003684<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003685<p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003686
3687<h5>Arguments:</h5>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003688<p>The two arguments to the '<tt>fmul</tt>' instruction must be
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003689 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3690 floating point values. Both arguments must have identical types.</p>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003691
3692<h5>Semantics:</h5>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003693<p>The value produced is the floating point product of the two operands.</p>
3694
3695<h5>Example:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003696<pre>
3697 &lt;result&gt; = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003698</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003699
Dan Gohmanae3a0be2009-06-04 22:49:04 +00003700</div>
3701
3702<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003703<h4>
3704 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
3705</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003706
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003707<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003708
Reid Spencer1628cec2006-10-26 06:15:43 +00003709<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003710<pre>
Chris Lattner35bda892011-02-06 21:44:57 +00003711 &lt;result&gt; = udiv &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
3712 &lt;result&gt; = udiv exact &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Reid Spencer1628cec2006-10-26 06:15:43 +00003713</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003714
Reid Spencer1628cec2006-10-26 06:15:43 +00003715<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003716<p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003717
Reid Spencer1628cec2006-10-26 06:15:43 +00003718<h5>Arguments:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00003719<p>The two arguments to the '<tt>udiv</tt>' instruction must be
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003720 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3721 values. Both arguments must have identical types.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003722
Reid Spencer1628cec2006-10-26 06:15:43 +00003723<h5>Semantics:</h5>
Chris Lattner5ec89832008-01-28 00:36:27 +00003724<p>The value produced is the unsigned integer quotient of the two operands.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003725
Chris Lattner5ec89832008-01-28 00:36:27 +00003726<p>Note that unsigned integer division and signed integer division are distinct
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003727 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3728
Chris Lattner5ec89832008-01-28 00:36:27 +00003729<p>Division by zero leads to undefined behavior.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003730
Chris Lattner35bda892011-02-06 21:44:57 +00003731<p>If the <tt>exact</tt> keyword is present, the result value of the
3732 <tt>udiv</tt> is a <a href="#trapvalues">trap value</a> if %op1 is not a
3733 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
3734
3735
Reid Spencer1628cec2006-10-26 06:15:43 +00003736<h5>Example:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003737<pre>
3738 &lt;result&gt; = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
Reid Spencer1628cec2006-10-26 06:15:43 +00003739</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003740
Reid Spencer1628cec2006-10-26 06:15:43 +00003741</div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003742
Reid Spencer1628cec2006-10-26 06:15:43 +00003743<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003744<h4>
3745 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
3746</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003747
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003748<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003749
Reid Spencer1628cec2006-10-26 06:15:43 +00003750<h5>Syntax:</h5>
Chris Lattner5568e942008-05-20 20:48:21 +00003751<pre>
Dan Gohmancbb38f22009-07-20 22:41:19 +00003752 &lt;result&gt; = sdiv &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanfdfca792009-09-02 17:31:42 +00003753 &lt;result&gt; = sdiv exact &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Reid Spencer1628cec2006-10-26 06:15:43 +00003754</pre>
Chris Lattner5568e942008-05-20 20:48:21 +00003755
Reid Spencer1628cec2006-10-26 06:15:43 +00003756<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003757<p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003758
Reid Spencer1628cec2006-10-26 06:15:43 +00003759<h5>Arguments:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00003760<p>The two arguments to the '<tt>sdiv</tt>' instruction must be
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003761 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3762 values. Both arguments must have identical types.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003763
Reid Spencer1628cec2006-10-26 06:15:43 +00003764<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003765<p>The value produced is the signed integer quotient of the two operands rounded
3766 towards zero.</p>
3767
Chris Lattner5ec89832008-01-28 00:36:27 +00003768<p>Note that signed integer division and unsigned integer division are distinct
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003769 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3770
Chris Lattner5ec89832008-01-28 00:36:27 +00003771<p>Division by zero leads to undefined behavior. Overflow also leads to
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003772 undefined behavior; this is a rare case, but can occur, for example, by doing
3773 a 32-bit division of -2147483648 by -1.</p>
3774
Dan Gohman9c5beed2009-07-22 00:04:19 +00003775<p>If the <tt>exact</tt> keyword is present, the result value of the
Dan Gohman27ef9972010-04-23 15:23:32 +00003776 <tt>sdiv</tt> is a <a href="#trapvalues">trap value</a> if the result would
Dan Gohman38da9272010-07-11 00:08:34 +00003777 be rounded.</p>
Dan Gohmancbb38f22009-07-20 22:41:19 +00003778
Reid Spencer1628cec2006-10-26 06:15:43 +00003779<h5>Example:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003780<pre>
3781 &lt;result&gt; = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
Reid Spencer1628cec2006-10-26 06:15:43 +00003782</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003783
Reid Spencer1628cec2006-10-26 06:15:43 +00003784</div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003785
Reid Spencer1628cec2006-10-26 06:15:43 +00003786<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003787<h4>
3788 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
3789</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003790
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003791<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003792
Chris Lattner00950542001-06-06 20:29:01 +00003793<h5>Syntax:</h5>
Chris Lattner5568e942008-05-20 20:48:21 +00003794<pre>
Gabor Greiffb224a22008-08-07 21:46:00 +00003795 &lt;result&gt; = fdiv &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Chris Lattner261efe92003-11-25 01:02:51 +00003796</pre>
Chris Lattner5568e942008-05-20 20:48:21 +00003797
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003798<h5>Overview:</h5>
3799<p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003800
Chris Lattner261efe92003-11-25 01:02:51 +00003801<h5>Arguments:</h5>
Jeff Cohen6f1cc772007-04-22 01:17:39 +00003802<p>The two arguments to the '<tt>fdiv</tt>' instruction must be
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003803 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3804 floating point values. Both arguments must have identical types.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003805
Chris Lattner261efe92003-11-25 01:02:51 +00003806<h5>Semantics:</h5>
Reid Spencer1628cec2006-10-26 06:15:43 +00003807<p>The value produced is the floating point quotient of the two operands.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003808
Chris Lattner261efe92003-11-25 01:02:51 +00003809<h5>Example:</h5>
Chris Lattner5568e942008-05-20 20:48:21 +00003810<pre>
3811 &lt;result&gt; = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
Chris Lattner261efe92003-11-25 01:02:51 +00003812</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003813
Chris Lattner261efe92003-11-25 01:02:51 +00003814</div>
Chris Lattner5568e942008-05-20 20:48:21 +00003815
Chris Lattner261efe92003-11-25 01:02:51 +00003816<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003817<h4>
3818 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3819</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003820
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003821<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003822
Reid Spencer0a783f72006-11-02 01:53:59 +00003823<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003824<pre>
3825 &lt;result&gt; = urem &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Reid Spencer0a783f72006-11-02 01:53:59 +00003826</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003827
Reid Spencer0a783f72006-11-02 01:53:59 +00003828<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003829<p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3830 division of its two arguments.</p>
3831
Reid Spencer0a783f72006-11-02 01:53:59 +00003832<h5>Arguments:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00003833<p>The two arguments to the '<tt>urem</tt>' instruction must be
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003834 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3835 values. Both arguments must have identical types.</p>
3836
Reid Spencer0a783f72006-11-02 01:53:59 +00003837<h5>Semantics:</h5>
3838<p>This instruction returns the unsigned integer <i>remainder</i> of a division.
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003839 This instruction always performs an unsigned division to get the
3840 remainder.</p>
3841
Chris Lattner5ec89832008-01-28 00:36:27 +00003842<p>Note that unsigned integer remainder and signed integer remainder are
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003843 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3844
Chris Lattner5ec89832008-01-28 00:36:27 +00003845<p>Taking the remainder of a division by zero leads to undefined behavior.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003846
Reid Spencer0a783f72006-11-02 01:53:59 +00003847<h5>Example:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003848<pre>
3849 &lt;result&gt; = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
Reid Spencer0a783f72006-11-02 01:53:59 +00003850</pre>
3851
3852</div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003853
Reid Spencer0a783f72006-11-02 01:53:59 +00003854<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003855<h4>
Chris Lattner5568e942008-05-20 20:48:21 +00003856 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003857</h4>
Chris Lattner5568e942008-05-20 20:48:21 +00003858
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003859<div>
Chris Lattner5568e942008-05-20 20:48:21 +00003860
Chris Lattner261efe92003-11-25 01:02:51 +00003861<h5>Syntax:</h5>
Chris Lattner5568e942008-05-20 20:48:21 +00003862<pre>
Gabor Greiffb224a22008-08-07 21:46:00 +00003863 &lt;result&gt; = srem &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Chris Lattner261efe92003-11-25 01:02:51 +00003864</pre>
Chris Lattner5568e942008-05-20 20:48:21 +00003865
Chris Lattner261efe92003-11-25 01:02:51 +00003866<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003867<p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3868 division of its two operands. This instruction can also take
3869 <a href="#t_vector">vector</a> versions of the values in which case the
3870 elements must be integers.</p>
Chris Lattnerc7d3ab32008-01-04 04:33:49 +00003871
Chris Lattner261efe92003-11-25 01:02:51 +00003872<h5>Arguments:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00003873<p>The two arguments to the '<tt>srem</tt>' instruction must be
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003874 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3875 values. Both arguments must have identical types.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003876
Chris Lattner261efe92003-11-25 01:02:51 +00003877<h5>Semantics:</h5>
Reid Spencer0a783f72006-11-02 01:53:59 +00003878<p>This instruction returns the <i>remainder</i> of a division (where the result
Duncan Sandsdea3a5e2011-03-07 09:12:24 +00003879 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
3880 <i>modulo</i> operator (where the result is either zero or has the same sign
3881 as the divisor, <tt>op2</tt>) of a value.
3882 For more information about the difference,
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003883 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3884 Math Forum</a>. For a table of how this is implemented in various languages,
3885 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3886 Wikipedia: modulo operation</a>.</p>
3887
Chris Lattner5ec89832008-01-28 00:36:27 +00003888<p>Note that signed integer remainder and unsigned integer remainder are
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003889 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3890
Chris Lattner5ec89832008-01-28 00:36:27 +00003891<p>Taking the remainder of a division by zero leads to undefined behavior.
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003892 Overflow also leads to undefined behavior; this is a rare case, but can
3893 occur, for example, by taking the remainder of a 32-bit division of
3894 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3895 lets srem be implemented using instructions that return both the result of
3896 the division and the remainder.)</p>
3897
Chris Lattner261efe92003-11-25 01:02:51 +00003898<h5>Example:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003899<pre>
3900 &lt;result&gt; = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
Reid Spencer0a783f72006-11-02 01:53:59 +00003901</pre>
3902
3903</div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003904
Reid Spencer0a783f72006-11-02 01:53:59 +00003905<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003906<h4>
3907 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
3908</h4>
Chris Lattner5568e942008-05-20 20:48:21 +00003909
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003910<div>
Chris Lattner5568e942008-05-20 20:48:21 +00003911
Reid Spencer0a783f72006-11-02 01:53:59 +00003912<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003913<pre>
3914 &lt;result&gt; = frem &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Reid Spencer0a783f72006-11-02 01:53:59 +00003915</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003916
Reid Spencer0a783f72006-11-02 01:53:59 +00003917<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003918<p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3919 its two operands.</p>
3920
Reid Spencer0a783f72006-11-02 01:53:59 +00003921<h5>Arguments:</h5>
3922<p>The two arguments to the '<tt>frem</tt>' instruction must be
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003923 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3924 floating point values. Both arguments must have identical types.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003925
Reid Spencer0a783f72006-11-02 01:53:59 +00003926<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003927<p>This instruction returns the <i>remainder</i> of a division. The remainder
3928 has the same sign as the dividend.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00003929
Reid Spencer0a783f72006-11-02 01:53:59 +00003930<h5>Example:</h5>
Chris Lattner5568e942008-05-20 20:48:21 +00003931<pre>
3932 &lt;result&gt; = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
Chris Lattner261efe92003-11-25 01:02:51 +00003933</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003934
Misha Brukman9d0919f2003-11-08 01:05:38 +00003935</div>
Robert Bocchino7b81c752006-02-17 21:18:08 +00003936
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003937</div>
3938
Reid Spencer8e11bf82007-02-02 13:57:07 +00003939<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003940<h3>
3941 <a name="bitwiseops">Bitwise Binary Operations</a>
3942</h3>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003943
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003944<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003945
3946<p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3947 program. They are generally very efficient instructions and can commonly be
3948 strength reduced from other instructions. They require two operands of the
3949 same type, execute an operation on them, and produce a single value. The
3950 resulting value is the same type as its operands.</p>
3951
Reid Spencer569f2fa2007-01-31 21:39:12 +00003952<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00003953<h4>
3954 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
3955</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003956
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00003957<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003958
Reid Spencer569f2fa2007-01-31 21:39:12 +00003959<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003960<pre>
Chris Lattnerf067d582011-02-07 16:40:21 +00003961 &lt;result&gt; = shl &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
3962 &lt;result&gt; = shl nuw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
3963 &lt;result&gt; = shl nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
3964 &lt;result&gt; = shl nuw nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Reid Spencer569f2fa2007-01-31 21:39:12 +00003965</pre>
Chris Lattner6ccc2d52007-10-03 21:01:14 +00003966
Reid Spencer569f2fa2007-01-31 21:39:12 +00003967<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003968<p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3969 a specified number of bits.</p>
Chris Lattner6ccc2d52007-10-03 21:01:14 +00003970
Reid Spencer569f2fa2007-01-31 21:39:12 +00003971<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003972<p>Both arguments to the '<tt>shl</tt>' instruction must be the
3973 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3974 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00003975
Reid Spencer569f2fa2007-01-31 21:39:12 +00003976<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003977<p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3978 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3979 is (statically or dynamically) negative or equal to or larger than the number
3980 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3981 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3982 shift amount in <tt>op2</tt>.</p>
Chris Lattner6ccc2d52007-10-03 21:01:14 +00003983
Chris Lattnerf067d582011-02-07 16:40:21 +00003984<p>If the <tt>nuw</tt> keyword is present, then the shift produces a
3985 <a href="#trapvalues">trap value</a> if it shifts out any non-zero bits. If
Chris Lattner66298c12011-02-09 16:44:44 +00003986 the <tt>nsw</tt> keyword is present, then the shift produces a
Chris Lattnerf067d582011-02-07 16:40:21 +00003987 <a href="#trapvalues">trap value</a> if it shifts out any bits that disagree
3988 with the resultant sign bit. As such, NUW/NSW have the same semantics as
3989 they would if the shift were expressed as a mul instruction with the same
3990 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
3991
Bill Wendlinge910b4c2009-07-20 02:29:24 +00003992<h5>Example:</h5>
3993<pre>
Reid Spencer569f2fa2007-01-31 21:39:12 +00003994 &lt;result&gt; = shl i32 4, %var <i>; yields {i32}: 4 &lt;&lt; %var</i>
3995 &lt;result&gt; = shl i32 4, 2 <i>; yields {i32}: 16</i>
3996 &lt;result&gt; = shl i32 1, 10 <i>; yields {i32}: 1024</i>
Chris Lattner6ccc2d52007-10-03 21:01:14 +00003997 &lt;result&gt; = shl i32 1, 32 <i>; undefined</i>
Mon P Wange9f10152008-12-09 05:46:39 +00003998 &lt;result&gt; = shl &lt;2 x i32&gt; &lt; i32 1, i32 1&gt;, &lt; i32 1, i32 2&gt; <i>; yields: result=&lt;2 x i32&gt; &lt; i32 2, i32 4&gt;</i>
Reid Spencer569f2fa2007-01-31 21:39:12 +00003999</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004000
Reid Spencer569f2fa2007-01-31 21:39:12 +00004001</div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004002
Reid Spencer569f2fa2007-01-31 21:39:12 +00004003<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004004<h4>
4005 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4006</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004007
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004008<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004009
Reid Spencer569f2fa2007-01-31 21:39:12 +00004010<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004011<pre>
Chris Lattnerf067d582011-02-07 16:40:21 +00004012 &lt;result&gt; = lshr &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
4013 &lt;result&gt; = lshr exact &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Reid Spencer569f2fa2007-01-31 21:39:12 +00004014</pre>
4015
4016<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004017<p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4018 operand shifted to the right a specified number of bits with zero fill.</p>
Reid Spencer569f2fa2007-01-31 21:39:12 +00004019
4020<h5>Arguments:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00004021<p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004022 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4023 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
Reid Spencer569f2fa2007-01-31 21:39:12 +00004024
4025<h5>Semantics:</h5>
4026<p>This instruction always performs a logical shift right operation. The most
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004027 significant bits of the result will be filled with zero bits after the shift.
4028 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4029 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4030 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4031 shift amount in <tt>op2</tt>.</p>
Reid Spencer569f2fa2007-01-31 21:39:12 +00004032
Chris Lattnerf067d582011-02-07 16:40:21 +00004033<p>If the <tt>exact</tt> keyword is present, the result value of the
4034 <tt>lshr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
4035 shifted out are non-zero.</p>
4036
4037
Reid Spencer569f2fa2007-01-31 21:39:12 +00004038<h5>Example:</h5>
4039<pre>
4040 &lt;result&gt; = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4041 &lt;result&gt; = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4042 &lt;result&gt; = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4043 &lt;result&gt; = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
Chris Lattner6ccc2d52007-10-03 21:01:14 +00004044 &lt;result&gt; = lshr i32 1, 32 <i>; undefined</i>
Mon P Wange9f10152008-12-09 05:46:39 +00004045 &lt;result&gt; = lshr &lt;2 x i32&gt; &lt; i32 -2, i32 4&gt;, &lt; i32 1, i32 2&gt; <i>; yields: result=&lt;2 x i32&gt; &lt; i32 0x7FFFFFFF, i32 1&gt;</i>
Reid Spencer569f2fa2007-01-31 21:39:12 +00004046</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004047
Reid Spencer569f2fa2007-01-31 21:39:12 +00004048</div>
4049
Reid Spencer8e11bf82007-02-02 13:57:07 +00004050<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004051<h4>
4052 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4053</h4>
4054
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004055<div>
Reid Spencer569f2fa2007-01-31 21:39:12 +00004056
4057<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004058<pre>
Chris Lattnerf067d582011-02-07 16:40:21 +00004059 &lt;result&gt; = ashr &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
4060 &lt;result&gt; = ashr exact &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Reid Spencer569f2fa2007-01-31 21:39:12 +00004061</pre>
4062
4063<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004064<p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4065 operand shifted to the right a specified number of bits with sign
4066 extension.</p>
Reid Spencer569f2fa2007-01-31 21:39:12 +00004067
4068<h5>Arguments:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00004069<p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004070 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4071 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
Reid Spencer569f2fa2007-01-31 21:39:12 +00004072
4073<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004074<p>This instruction always performs an arithmetic shift right operation, The
4075 most significant bits of the result will be filled with the sign bit
4076 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4077 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4078 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4079 the corresponding shift amount in <tt>op2</tt>.</p>
Reid Spencer569f2fa2007-01-31 21:39:12 +00004080
Chris Lattnerf067d582011-02-07 16:40:21 +00004081<p>If the <tt>exact</tt> keyword is present, the result value of the
4082 <tt>ashr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
4083 shifted out are non-zero.</p>
4084
Reid Spencer569f2fa2007-01-31 21:39:12 +00004085<h5>Example:</h5>
4086<pre>
4087 &lt;result&gt; = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4088 &lt;result&gt; = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4089 &lt;result&gt; = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4090 &lt;result&gt; = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
Chris Lattner6ccc2d52007-10-03 21:01:14 +00004091 &lt;result&gt; = ashr i32 1, 32 <i>; undefined</i>
Mon P Wange9f10152008-12-09 05:46:39 +00004092 &lt;result&gt; = ashr &lt;2 x i32&gt; &lt; i32 -2, i32 4&gt;, &lt; i32 1, i32 3&gt; <i>; yields: result=&lt;2 x i32&gt; &lt; i32 -1, i32 0&gt;</i>
Reid Spencer569f2fa2007-01-31 21:39:12 +00004093</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004094
Reid Spencer569f2fa2007-01-31 21:39:12 +00004095</div>
4096
Chris Lattner00950542001-06-06 20:29:01 +00004097<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004098<h4>
4099 <a name="i_and">'<tt>and</tt>' Instruction</a>
4100</h4>
Chris Lattner5568e942008-05-20 20:48:21 +00004101
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004102<div>
Chris Lattner5568e942008-05-20 20:48:21 +00004103
Chris Lattner00950542001-06-06 20:29:01 +00004104<h5>Syntax:</h5>
Chris Lattner5568e942008-05-20 20:48:21 +00004105<pre>
Gabor Greiffb224a22008-08-07 21:46:00 +00004106 &lt;result&gt; = and &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Chris Lattner00950542001-06-06 20:29:01 +00004107</pre>
Chris Lattner5568e942008-05-20 20:48:21 +00004108
Chris Lattner00950542001-06-06 20:29:01 +00004109<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004110<p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4111 operands.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00004112
Chris Lattner00950542001-06-06 20:29:01 +00004113<h5>Arguments:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00004114<p>The two arguments to the '<tt>and</tt>' instruction must be
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004115 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4116 values. Both arguments must have identical types.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00004117
Chris Lattner00950542001-06-06 20:29:01 +00004118<h5>Semantics:</h5>
Misha Brukman9d0919f2003-11-08 01:05:38 +00004119<p>The truth table used for the '<tt>and</tt>' instruction is:</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004120
Misha Brukman9d0919f2003-11-08 01:05:38 +00004121<table border="1" cellspacing="0" cellpadding="4">
Chris Lattner261efe92003-11-25 01:02:51 +00004122 <tbody>
4123 <tr>
4124 <td>In0</td>
4125 <td>In1</td>
4126 <td>Out</td>
4127 </tr>
4128 <tr>
4129 <td>0</td>
4130 <td>0</td>
4131 <td>0</td>
4132 </tr>
4133 <tr>
4134 <td>0</td>
4135 <td>1</td>
4136 <td>0</td>
4137 </tr>
4138 <tr>
4139 <td>1</td>
4140 <td>0</td>
4141 <td>0</td>
4142 </tr>
4143 <tr>
4144 <td>1</td>
4145 <td>1</td>
4146 <td>1</td>
4147 </tr>
4148 </tbody>
4149</table>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004150
Chris Lattner00950542001-06-06 20:29:01 +00004151<h5>Example:</h5>
Chris Lattner5568e942008-05-20 20:48:21 +00004152<pre>
4153 &lt;result&gt; = and i32 4, %var <i>; yields {i32}:result = 4 &amp; %var</i>
Reid Spencerca86e162006-12-31 07:07:53 +00004154 &lt;result&gt; = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4155 &lt;result&gt; = and i32 4, 8 <i>; yields {i32}:result = 0</i>
Chris Lattner00950542001-06-06 20:29:01 +00004156</pre>
Misha Brukman9d0919f2003-11-08 01:05:38 +00004157</div>
Chris Lattner00950542001-06-06 20:29:01 +00004158<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004159<h4>
4160 <a name="i_or">'<tt>or</tt>' Instruction</a>
4161</h4>
Chris Lattner5568e942008-05-20 20:48:21 +00004162
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004163<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004164
4165<h5>Syntax:</h5>
4166<pre>
4167 &lt;result&gt; = or &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
4168</pre>
4169
4170<h5>Overview:</h5>
4171<p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4172 two operands.</p>
4173
4174<h5>Arguments:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00004175<p>The two arguments to the '<tt>or</tt>' instruction must be
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004176 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4177 values. Both arguments must have identical types.</p>
4178
Chris Lattner00950542001-06-06 20:29:01 +00004179<h5>Semantics:</h5>
Misha Brukman9d0919f2003-11-08 01:05:38 +00004180<p>The truth table used for the '<tt>or</tt>' instruction is:</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004181
Chris Lattner261efe92003-11-25 01:02:51 +00004182<table border="1" cellspacing="0" cellpadding="4">
4183 <tbody>
4184 <tr>
4185 <td>In0</td>
4186 <td>In1</td>
4187 <td>Out</td>
4188 </tr>
4189 <tr>
4190 <td>0</td>
4191 <td>0</td>
4192 <td>0</td>
4193 </tr>
4194 <tr>
4195 <td>0</td>
4196 <td>1</td>
4197 <td>1</td>
4198 </tr>
4199 <tr>
4200 <td>1</td>
4201 <td>0</td>
4202 <td>1</td>
4203 </tr>
4204 <tr>
4205 <td>1</td>
4206 <td>1</td>
4207 <td>1</td>
4208 </tr>
4209 </tbody>
4210</table>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004211
Chris Lattner00950542001-06-06 20:29:01 +00004212<h5>Example:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004213<pre>
4214 &lt;result&gt; = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
Reid Spencerca86e162006-12-31 07:07:53 +00004215 &lt;result&gt; = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4216 &lt;result&gt; = or i32 4, 8 <i>; yields {i32}:result = 12</i>
Chris Lattner00950542001-06-06 20:29:01 +00004217</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004218
Misha Brukman9d0919f2003-11-08 01:05:38 +00004219</div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004220
Chris Lattner00950542001-06-06 20:29:01 +00004221<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004222<h4>
4223 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4224</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004225
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004226<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004227
Chris Lattner00950542001-06-06 20:29:01 +00004228<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004229<pre>
4230 &lt;result&gt; = xor &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Chris Lattner00950542001-06-06 20:29:01 +00004231</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004232
Chris Lattner00950542001-06-06 20:29:01 +00004233<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004234<p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4235 its two operands. The <tt>xor</tt> is used to implement the "one's
4236 complement" operation, which is the "~" operator in C.</p>
4237
Chris Lattner00950542001-06-06 20:29:01 +00004238<h5>Arguments:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00004239<p>The two arguments to the '<tt>xor</tt>' instruction must be
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004240 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4241 values. Both arguments must have identical types.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00004242
Chris Lattner00950542001-06-06 20:29:01 +00004243<h5>Semantics:</h5>
Misha Brukman9d0919f2003-11-08 01:05:38 +00004244<p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004245
Chris Lattner261efe92003-11-25 01:02:51 +00004246<table border="1" cellspacing="0" cellpadding="4">
4247 <tbody>
4248 <tr>
4249 <td>In0</td>
4250 <td>In1</td>
4251 <td>Out</td>
4252 </tr>
4253 <tr>
4254 <td>0</td>
4255 <td>0</td>
4256 <td>0</td>
4257 </tr>
4258 <tr>
4259 <td>0</td>
4260 <td>1</td>
4261 <td>1</td>
4262 </tr>
4263 <tr>
4264 <td>1</td>
4265 <td>0</td>
4266 <td>1</td>
4267 </tr>
4268 <tr>
4269 <td>1</td>
4270 <td>1</td>
4271 <td>0</td>
4272 </tr>
4273 </tbody>
4274</table>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004275
Chris Lattner00950542001-06-06 20:29:01 +00004276<h5>Example:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004277<pre>
4278 &lt;result&gt; = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
Reid Spencerca86e162006-12-31 07:07:53 +00004279 &lt;result&gt; = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4280 &lt;result&gt; = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4281 &lt;result&gt; = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
Chris Lattner00950542001-06-06 20:29:01 +00004282</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004283
Misha Brukman9d0919f2003-11-08 01:05:38 +00004284</div>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004285
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004286</div>
4287
Chris Lattner00950542001-06-06 20:29:01 +00004288<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004289<h3>
Chris Lattner3df241e2006-04-08 23:07:04 +00004290 <a name="vectorops">Vector Operations</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004291</h3>
Chris Lattner3df241e2006-04-08 23:07:04 +00004292
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004293<div>
Chris Lattner3df241e2006-04-08 23:07:04 +00004294
4295<p>LLVM supports several instructions to represent vector operations in a
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004296 target-independent manner. These instructions cover the element-access and
4297 vector-specific operations needed to process vectors effectively. While LLVM
4298 does directly support these vector operations, many sophisticated algorithms
4299 will want to use target-specific intrinsics to take full advantage of a
4300 specific target.</p>
Chris Lattner3df241e2006-04-08 23:07:04 +00004301
Chris Lattner3df241e2006-04-08 23:07:04 +00004302<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004303<h4>
Chris Lattner3df241e2006-04-08 23:07:04 +00004304 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004305</h4>
Chris Lattner3df241e2006-04-08 23:07:04 +00004306
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004307<div>
Chris Lattner3df241e2006-04-08 23:07:04 +00004308
4309<h5>Syntax:</h5>
Chris Lattner3df241e2006-04-08 23:07:04 +00004310<pre>
Reid Spencerca86e162006-12-31 07:07:53 +00004311 &lt;result&gt; = extractelement &lt;n x &lt;ty&gt;&gt; &lt;val&gt;, i32 &lt;idx&gt; <i>; yields &lt;ty&gt;</i>
Chris Lattner3df241e2006-04-08 23:07:04 +00004312</pre>
4313
4314<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004315<p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4316 from a vector at a specified index.</p>
Chris Lattner3df241e2006-04-08 23:07:04 +00004317
4318
4319<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004320<p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4321 of <a href="#t_vector">vector</a> type. The second operand is an index
4322 indicating the position from which to extract the element. The index may be
4323 a variable.</p>
Chris Lattner3df241e2006-04-08 23:07:04 +00004324
4325<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004326<p>The result is a scalar of the same type as the element type of
4327 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4328 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4329 results are undefined.</p>
Chris Lattner3df241e2006-04-08 23:07:04 +00004330
4331<h5>Example:</h5>
Chris Lattner3df241e2006-04-08 23:07:04 +00004332<pre>
Gabor Greifa5b6f452009-10-28 13:14:50 +00004333 &lt;result&gt; = extractelement &lt;4 x i32&gt; %vec, i32 0 <i>; yields i32</i>
Chris Lattner3df241e2006-04-08 23:07:04 +00004334</pre>
Chris Lattner3df241e2006-04-08 23:07:04 +00004335
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004336</div>
Chris Lattner3df241e2006-04-08 23:07:04 +00004337
4338<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004339<h4>
Chris Lattner3df241e2006-04-08 23:07:04 +00004340 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004341</h4>
Chris Lattner3df241e2006-04-08 23:07:04 +00004342
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004343<div>
Chris Lattner3df241e2006-04-08 23:07:04 +00004344
4345<h5>Syntax:</h5>
Chris Lattner3df241e2006-04-08 23:07:04 +00004346<pre>
Dan Gohmanf3480b92008-05-12 23:38:42 +00004347 &lt;result&gt; = insertelement &lt;n x &lt;ty&gt;&gt; &lt;val&gt;, &lt;ty&gt; &lt;elt&gt;, i32 &lt;idx&gt; <i>; yields &lt;n x &lt;ty&gt;&gt;</i>
Chris Lattner3df241e2006-04-08 23:07:04 +00004348</pre>
4349
4350<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004351<p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4352 vector at a specified index.</p>
Chris Lattner3df241e2006-04-08 23:07:04 +00004353
4354<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004355<p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4356 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4357 whose type must equal the element type of the first operand. The third
4358 operand is an index indicating the position at which to insert the value.
4359 The index may be a variable.</p>
Chris Lattner3df241e2006-04-08 23:07:04 +00004360
4361<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004362<p>The result is a vector of the same type as <tt>val</tt>. Its element values
4363 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4364 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4365 results are undefined.</p>
Chris Lattner3df241e2006-04-08 23:07:04 +00004366
4367<h5>Example:</h5>
Chris Lattner3df241e2006-04-08 23:07:04 +00004368<pre>
Gabor Greifa5b6f452009-10-28 13:14:50 +00004369 &lt;result&gt; = insertelement &lt;4 x i32&gt; %vec, i32 1, i32 0 <i>; yields &lt;4 x i32&gt;</i>
Chris Lattner3df241e2006-04-08 23:07:04 +00004370</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004371
Chris Lattner3df241e2006-04-08 23:07:04 +00004372</div>
4373
4374<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004375<h4>
Chris Lattner3df241e2006-04-08 23:07:04 +00004376 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004377</h4>
Chris Lattner3df241e2006-04-08 23:07:04 +00004378
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004379<div>
Chris Lattner3df241e2006-04-08 23:07:04 +00004380
4381<h5>Syntax:</h5>
Chris Lattner3df241e2006-04-08 23:07:04 +00004382<pre>
Mon P Wangaeb06d22008-11-10 04:46:22 +00004383 &lt;result&gt; = shufflevector &lt;n x &lt;ty&gt;&gt; &lt;v1&gt;, &lt;n x &lt;ty&gt;&gt; &lt;v2&gt;, &lt;m x i32&gt; &lt;mask&gt; <i>; yields &lt;m x &lt;ty&gt;&gt;</i>
Chris Lattner3df241e2006-04-08 23:07:04 +00004384</pre>
4385
4386<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004387<p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4388 from two input vectors, returning a vector with the same element type as the
4389 input and length that is the same as the shuffle mask.</p>
Chris Lattner3df241e2006-04-08 23:07:04 +00004390
4391<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004392<p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4393 with types that match each other. The third argument is a shuffle mask whose
4394 element type is always 'i32'. The result of the instruction is a vector
4395 whose length is the same as the shuffle mask and whose element type is the
4396 same as the element type of the first two operands.</p>
Chris Lattner3df241e2006-04-08 23:07:04 +00004397
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004398<p>The shuffle mask operand is required to be a constant vector with either
4399 constant integer or undef values.</p>
Chris Lattner3df241e2006-04-08 23:07:04 +00004400
4401<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004402<p>The elements of the two input vectors are numbered from left to right across
4403 both of the vectors. The shuffle mask operand specifies, for each element of
4404 the result vector, which element of the two input vectors the result element
4405 gets. The element selector may be undef (meaning "don't care") and the
4406 second operand may be undef if performing a shuffle from only one vector.</p>
Chris Lattner3df241e2006-04-08 23:07:04 +00004407
4408<h5>Example:</h5>
Chris Lattner3df241e2006-04-08 23:07:04 +00004409<pre>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00004410 &lt;result&gt; = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; %v2,
Jeff Cohen6f1cc772007-04-22 01:17:39 +00004411 &lt;4 x i32&gt; &lt;i32 0, i32 4, i32 1, i32 5&gt; <i>; yields &lt;4 x i32&gt;</i>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00004412 &lt;result&gt; = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; undef,
Reid Spencerca86e162006-12-31 07:07:53 +00004413 &lt;4 x i32&gt; &lt;i32 0, i32 1, i32 2, i32 3&gt; <i>; yields &lt;4 x i32&gt;</i> - Identity shuffle.
Eric Christopher6c7e8a02009-12-05 02:46:03 +00004414 &lt;result&gt; = shufflevector &lt;8 x i32&gt; %v1, &lt;8 x i32&gt; undef,
Mon P Wangaeb06d22008-11-10 04:46:22 +00004415 &lt;4 x i32&gt; &lt;i32 0, i32 1, i32 2, i32 3&gt; <i>; yields &lt;4 x i32&gt;</i>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00004416 &lt;result&gt; = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; %v2,
Mon P Wangaeb06d22008-11-10 04:46:22 +00004417 &lt;8 x i32&gt; &lt;i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 &gt; <i>; yields &lt;8 x i32&gt;</i>
Chris Lattner3df241e2006-04-08 23:07:04 +00004418</pre>
Chris Lattner3df241e2006-04-08 23:07:04 +00004419
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004420</div>
Tanya Lattner09474292006-04-14 19:24:33 +00004421
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004422</div>
4423
Chris Lattner3df241e2006-04-08 23:07:04 +00004424<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004425<h3>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004426 <a name="aggregateops">Aggregate Operations</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004427</h3>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004428
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004429<div>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004430
Chris Lattnerfdfeb692010-02-12 20:49:41 +00004431<p>LLVM supports several instructions for working with
4432 <a href="#t_aggregate">aggregate</a> values.</p>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004433
Dan Gohmana334d5f2008-05-12 23:51:09 +00004434<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004435<h4>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004436 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004437</h4>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004438
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004439<div>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004440
4441<h5>Syntax:</h5>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004442<pre>
4443 &lt;result&gt; = extractvalue &lt;aggregate type&gt; &lt;val&gt;, &lt;idx&gt;{, &lt;idx&gt;}*
4444</pre>
4445
4446<h5>Overview:</h5>
Chris Lattnerfdfeb692010-02-12 20:49:41 +00004447<p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4448 from an <a href="#t_aggregate">aggregate</a> value.</p>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004449
4450<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004451<p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
Chris Lattner61c70e92010-08-28 04:09:24 +00004452 of <a href="#t_struct">struct</a> or
Chris Lattnerfdfeb692010-02-12 20:49:41 +00004453 <a href="#t_array">array</a> type. The operands are constant indices to
4454 specify which value to extract in a similar manner as indices in a
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004455 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
Frits van Bommel13242892010-12-05 20:54:38 +00004456 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4457 <ul>
4458 <li>Since the value being indexed is not a pointer, the first index is
4459 omitted and assumed to be zero.</li>
4460 <li>At least one index must be specified.</li>
4461 <li>Not only struct indices but also array indices must be in
4462 bounds.</li>
4463 </ul>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004464
4465<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004466<p>The result is the value at the position in the aggregate specified by the
4467 index operands.</p>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004468
4469<h5>Example:</h5>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004470<pre>
Gabor Greifa5b6f452009-10-28 13:14:50 +00004471 &lt;result&gt; = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004472</pre>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004473
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004474</div>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004475
4476<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004477<h4>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004478 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004479</h4>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004480
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004481<div>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004482
4483<h5>Syntax:</h5>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004484<pre>
Bill Wendling194229e2011-07-26 20:42:28 +00004485 &lt;result&gt; = insertvalue &lt;aggregate type&gt; &lt;val&gt;, &lt;ty&gt; &lt;elt&gt;, &lt;idx&gt;{, &lt;idx&gt;}* <i>; yields &lt;aggregate type&gt;</i>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004486</pre>
4487
4488<h5>Overview:</h5>
Chris Lattnerfdfeb692010-02-12 20:49:41 +00004489<p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4490 in an <a href="#t_aggregate">aggregate</a> value.</p>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004491
4492<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004493<p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
Chris Lattner61c70e92010-08-28 04:09:24 +00004494 of <a href="#t_struct">struct</a> or
Chris Lattnerfdfeb692010-02-12 20:49:41 +00004495 <a href="#t_array">array</a> type. The second operand is a first-class
4496 value to insert. The following operands are constant indices indicating
4497 the position at which to insert the value in a similar manner as indices in a
Frits van Bommel13242892010-12-05 20:54:38 +00004498 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004499 value to insert must have the same type as the value identified by the
4500 indices.</p>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004501
4502<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004503<p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4504 that of <tt>val</tt> except that the value at the position specified by the
4505 indices is that of <tt>elt</tt>.</p>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004506
4507<h5>Example:</h5>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004508<pre>
Chris Lattner8645d1a2011-05-22 07:18:08 +00004509 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4510 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4511 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004512</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004513
Dan Gohmana334d5f2008-05-12 23:51:09 +00004514</div>
4515
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004516</div>
Dan Gohmana334d5f2008-05-12 23:51:09 +00004517
4518<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004519<h3>
Chris Lattner884a9702006-08-15 00:45:58 +00004520 <a name="memoryops">Memory Access and Addressing Operations</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004521</h3>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004522
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004523<div>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004524
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004525<p>A key design point of an SSA-based representation is how it represents
4526 memory. In LLVM, no memory locations are in SSA form, which makes things
Victor Hernandez2fee2942009-10-26 23:44:29 +00004527 very simple. This section describes how to read, write, and allocate
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004528 memory in LLVM.</p>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004529
Chris Lattner00950542001-06-06 20:29:01 +00004530<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004531<h4>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004532 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004533</h4>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004534
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004535<div>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004536
Chris Lattner00950542001-06-06 20:29:01 +00004537<h5>Syntax:</h5>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004538<pre>
Dan Gohmanf75a7d32010-05-28 01:14:11 +00004539 &lt;result&gt; = alloca &lt;type&gt;[, &lt;ty&gt; &lt;NumElements&gt;][, align &lt;alignment&gt;] <i>; yields {type*}:result</i>
Chris Lattner00950542001-06-06 20:29:01 +00004540</pre>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004541
Chris Lattner00950542001-06-06 20:29:01 +00004542<h5>Overview:</h5>
Jeff Cohen6f1cc772007-04-22 01:17:39 +00004543<p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004544 currently executing function, to be automatically released when this function
4545 returns to its caller. The object is always allocated in the generic address
4546 space (address space zero).</p>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004547
Chris Lattner00950542001-06-06 20:29:01 +00004548<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004549<p>The '<tt>alloca</tt>' instruction
4550 allocates <tt>sizeof(&lt;type&gt;)*NumElements</tt> bytes of memory on the
4551 runtime stack, returning a pointer of the appropriate type to the program.
4552 If "NumElements" is specified, it is the number of elements allocated,
4553 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4554 specified, the value result of the allocation is guaranteed to be aligned to
4555 at least that boundary. If not specified, or if zero, the target can choose
4556 to align the allocation on any convenient boundary compatible with the
4557 type.</p>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004558
Misha Brukman9d0919f2003-11-08 01:05:38 +00004559<p>'<tt>type</tt>' may be any sized type.</p>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004560
Chris Lattner00950542001-06-06 20:29:01 +00004561<h5>Semantics:</h5>
Bill Wendling871eb0a2009-05-08 20:49:29 +00004562<p>Memory is allocated; a pointer is returned. The operation is undefined if
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004563 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4564 memory is automatically released when the function returns. The
4565 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4566 variables that must have an address available. When the function returns
4567 (either with the <tt><a href="#i_ret">ret</a></tt>
4568 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
4569 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004570
Chris Lattner00950542001-06-06 20:29:01 +00004571<h5>Example:</h5>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004572<pre>
Dan Gohman81e21672009-01-04 23:49:44 +00004573 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4574 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4575 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4576 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
Chris Lattner00950542001-06-06 20:29:01 +00004577</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004578
Misha Brukman9d0919f2003-11-08 01:05:38 +00004579</div>
Chris Lattner2cbdc452005-11-06 08:02:57 +00004580
Chris Lattner00950542001-06-06 20:29:01 +00004581<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004582<h4>
4583 <a name="i_load">'<tt>load</tt>' Instruction</a>
4584</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004585
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004586<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004587
Chris Lattner2b7d3202002-05-06 03:03:22 +00004588<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004589<pre>
Eli Friedmanf03bb262011-08-12 22:50:01 +00004590 &lt;result&gt; = load [volatile] &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;][, !nontemporal !&lt;index&gt;]
4591 &lt;result&gt; = load atomic [volatile] &lt;ty&gt;* &lt;pointer&gt; [singlethread] &lt;ordering&gt;, align &lt;alignment&gt;
Bill Wendling7c78dbb2010-02-25 21:23:24 +00004592 !&lt;index&gt; = !{ i32 1 }
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004593</pre>
4594
Chris Lattner2b7d3202002-05-06 03:03:22 +00004595<h5>Overview:</h5>
Misha Brukman9d0919f2003-11-08 01:05:38 +00004596<p>The '<tt>load</tt>' instruction is used to read from memory.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004597
Chris Lattner2b7d3202002-05-06 03:03:22 +00004598<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004599<p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4600 from which to load. The pointer must point to
4601 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4602 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
Jeffrey Yasskin93e066d2010-04-26 21:21:24 +00004603 number or order of execution of this <tt>load</tt> with other <a
4604 href="#volatile">volatile operations</a>.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004605
Eli Friedman21006d42011-08-09 23:02:53 +00004606<p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
4607 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4608 argument. The <code>release</code> and <code>acq_rel</code> orderings are
4609 not valid on <code>load</code> instructions. Atomic loads produce <a
4610 href="#memorymodel">defined</a> results when they may see multiple atomic
4611 stores. The type of the pointee must be an integer type whose bit width
4612 is a power of two greater than or equal to eight and less than or equal
4613 to a target-specific size limit. <code>align</code> must be explicitly
4614 specified on atomic loads, and the load has undefined behavior if the
4615 alignment is not set to a value which is at least the size in bytes of
4616 the pointee. <code>!nontemporal</code> does not have any defined semantics
4617 for atomic loads.</p>
4618
Bill Wendling7c78dbb2010-02-25 21:23:24 +00004619<p>The optional constant <tt>align</tt> argument specifies the alignment of the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004620 operation (that is, the alignment of the memory address). A value of 0 or an
Bill Wendling7c78dbb2010-02-25 21:23:24 +00004621 omitted <tt>align</tt> argument means that the operation has the preferential
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004622 alignment for the target. It is the responsibility of the code emitter to
4623 ensure that the alignment information is correct. Overestimating the
Bill Wendling7c78dbb2010-02-25 21:23:24 +00004624 alignment results in undefined behavior. Underestimating the alignment may
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004625 produce less efficient code. An alignment of 1 is always safe.</p>
4626
Bill Wendling7c78dbb2010-02-25 21:23:24 +00004627<p>The optional <tt>!nontemporal</tt> metadata must reference a single
4628 metatadata name &lt;index&gt; corresponding to a metadata node with
Dan Gohmanb55a1ee2010-03-01 17:41:39 +00004629 one <tt>i32</tt> entry of value 1. The existence of
Bill Wendling7c78dbb2010-02-25 21:23:24 +00004630 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4631 and code generator that this load is not expected to be reused in the cache.
4632 The code generator may select special instructions to save cache bandwidth,
Dan Gohmanb55a1ee2010-03-01 17:41:39 +00004633 such as the <tt>MOVNT</tt> instruction on x86.</p>
David Greene8939b0d2010-02-16 20:50:18 +00004634
Chris Lattner2b7d3202002-05-06 03:03:22 +00004635<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004636<p>The location of memory pointed to is loaded. If the value being loaded is of
4637 scalar type then the number of bytes read does not exceed the minimum number
4638 of bytes needed to hold all bits of the type. For example, loading an
4639 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4640 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4641 is undefined if the value was not originally written using a store of the
4642 same type.</p>
4643
Chris Lattner2b7d3202002-05-06 03:03:22 +00004644<h5>Examples:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004645<pre>
4646 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4647 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
Reid Spencerca86e162006-12-31 07:07:53 +00004648 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
Chris Lattner2b7d3202002-05-06 03:03:22 +00004649</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004650
Misha Brukman9d0919f2003-11-08 01:05:38 +00004651</div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004652
Chris Lattner2b7d3202002-05-06 03:03:22 +00004653<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004654<h4>
4655 <a name="i_store">'<tt>store</tt>' Instruction</a>
4656</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004657
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004658<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004659
Chris Lattner2b7d3202002-05-06 03:03:22 +00004660<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004661<pre>
Eli Friedmanf03bb262011-08-12 22:50:01 +00004662 store [volatile] &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;][, !nontemporal !&lt;index&gt;] <i>; yields {void}</i>
4663 store atomic [volatile] &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt; [singlethread] &lt;ordering&gt;, align &lt;alignment&gt; <i>; yields {void}</i>
Chris Lattner2b7d3202002-05-06 03:03:22 +00004664</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004665
Chris Lattner2b7d3202002-05-06 03:03:22 +00004666<h5>Overview:</h5>
Misha Brukman9d0919f2003-11-08 01:05:38 +00004667<p>The '<tt>store</tt>' instruction is used to write to memory.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004668
Chris Lattner2b7d3202002-05-06 03:03:22 +00004669<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004670<p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4671 and an address at which to store it. The type of the
4672 '<tt>&lt;pointer&gt;</tt>' operand must be a pointer to
4673 the <a href="#t_firstclass">first class</a> type of the
Jeffrey Yasskin93e066d2010-04-26 21:21:24 +00004674 '<tt>&lt;value&gt;</tt>' operand. If the <tt>store</tt> is marked as
4675 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
4676 order of execution of this <tt>store</tt> with other <a
4677 href="#volatile">volatile operations</a>.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004678
Eli Friedman21006d42011-08-09 23:02:53 +00004679<p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
4680 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4681 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
4682 valid on <code>store</code> instructions. Atomic loads produce <a
4683 href="#memorymodel">defined</a> results when they may see multiple atomic
4684 stores. The type of the pointee must be an integer type whose bit width
4685 is a power of two greater than or equal to eight and less than or equal
4686 to a target-specific size limit. <code>align</code> must be explicitly
4687 specified on atomic stores, and the store has undefined behavior if the
4688 alignment is not set to a value which is at least the size in bytes of
4689 the pointee. <code>!nontemporal</code> does not have any defined semantics
4690 for atomic stores.</p>
4691
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004692<p>The optional constant "align" argument specifies the alignment of the
4693 operation (that is, the alignment of the memory address). A value of 0 or an
4694 omitted "align" argument means that the operation has the preferential
4695 alignment for the target. It is the responsibility of the code emitter to
4696 ensure that the alignment information is correct. Overestimating the
4697 alignment results in an undefined behavior. Underestimating the alignment may
4698 produce less efficient code. An alignment of 1 is always safe.</p>
4699
David Greene8939b0d2010-02-16 20:50:18 +00004700<p>The optional !nontemporal metadata must reference a single metatadata
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00004701 name &lt;index&gt; corresponding to a metadata node with one i32 entry of
Dan Gohmanb55a1ee2010-03-01 17:41:39 +00004702 value 1. The existence of the !nontemporal metatadata on the
David Greene8939b0d2010-02-16 20:50:18 +00004703 instruction tells the optimizer and code generator that this load is
4704 not expected to be reused in the cache. The code generator may
4705 select special instructions to save cache bandwidth, such as the
Dan Gohmanb55a1ee2010-03-01 17:41:39 +00004706 MOVNT instruction on x86.</p>
David Greene8939b0d2010-02-16 20:50:18 +00004707
4708
Chris Lattner261efe92003-11-25 01:02:51 +00004709<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004710<p>The contents of memory are updated to contain '<tt>&lt;value&gt;</tt>' at the
4711 location specified by the '<tt>&lt;pointer&gt;</tt>' operand. If
4712 '<tt>&lt;value&gt;</tt>' is of scalar type then the number of bytes written
4713 does not exceed the minimum number of bytes needed to hold all bits of the
4714 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4715 writing a value of a type like <tt>i20</tt> with a size that is not an
4716 integral number of bytes, it is unspecified what happens to the extra bits
4717 that do not belong to the type, but they will typically be overwritten.</p>
4718
Chris Lattner2b7d3202002-05-06 03:03:22 +00004719<h5>Example:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004720<pre>
4721 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
Bill Wendling8c6c72d2007-10-22 05:10:05 +00004722 store i32 3, i32* %ptr <i>; yields {void}</i>
4723 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
Chris Lattner2b7d3202002-05-06 03:03:22 +00004724</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004725
Reid Spencer47ce1792006-11-09 21:15:49 +00004726</div>
4727
Chris Lattner2b7d3202002-05-06 03:03:22 +00004728<!-- _______________________________________________________________________ -->
NAKAMURA Takumib5bb29c2011-08-12 06:17:17 +00004729<h4>
4730<a name="i_fence">'<tt>fence</tt>' Instruction</a>
4731</h4>
Eli Friedman47f35132011-07-25 23:16:38 +00004732
NAKAMURA Takumib5bb29c2011-08-12 06:17:17 +00004733<div>
Eli Friedman47f35132011-07-25 23:16:38 +00004734
4735<h5>Syntax:</h5>
4736<pre>
4737 fence [singlethread] &lt;ordering&gt; <i>; yields {void}</i>
4738</pre>
4739
4740<h5>Overview:</h5>
4741<p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
4742between operations.</p>
4743
4744<h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
4745href="#ordering">ordering</a> argument which defines what
4746<i>synchronizes-with</i> edges they add. They can only be given
4747<code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
4748<code>seq_cst</code> orderings.</p>
4749
4750<h5>Semantics:</h5>
4751<p>A fence <var>A</var> which has (at least) <code>release</code> ordering
4752semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
4753<code>acquire</code> ordering semantics if and only if there exist atomic
4754operations <var>X</var> and <var>Y</var>, both operating on some atomic object
4755<var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
4756<var>X</var> modifies <var>M</var> (either directly or through some side effect
4757of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
4758<var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
4759<i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
4760than an explicit <code>fence</code>, one (but not both) of the atomic operations
4761<var>X</var> or <var>Y</var> might provide a <code>release</code> or
4762<code>acquire</code> (resp.) ordering constraint and still
4763<i>synchronize-with</i> the explicit <code>fence</code> and establish the
4764<i>happens-before</i> edge.</p>
4765
4766<p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
4767having both <code>acquire</code> and <code>release</code> semantics specified
4768above, participates in the global program order of other <code>seq_cst</code>
4769operations and/or fences.</p>
4770
4771<p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
4772specifies that the fence only synchronizes with other fences in the same
4773thread. (This is useful for interacting with signal handlers.)</p>
4774
Eli Friedman47f35132011-07-25 23:16:38 +00004775<h5>Example:</h5>
4776<pre>
4777 fence acquire <i>; yields {void}</i>
4778 fence singlethread seq_cst <i>; yields {void}</i>
4779</pre>
4780
4781</div>
4782
4783<!-- _______________________________________________________________________ -->
NAKAMURA Takumib5bb29c2011-08-12 06:17:17 +00004784<h4>
4785<a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
4786</h4>
Eli Friedmanff030482011-07-28 21:48:00 +00004787
NAKAMURA Takumib5bb29c2011-08-12 06:17:17 +00004788<div>
Eli Friedmanff030482011-07-28 21:48:00 +00004789
4790<h5>Syntax:</h5>
4791<pre>
Eli Friedmanf03bb262011-08-12 22:50:01 +00004792 cmpxchg [volatile] &lt;ty&gt;* &lt;pointer&gt;, &lt;ty&gt; &lt;cmp&gt;, &lt;ty&gt; &lt;new&gt; [singlethread] &lt;ordering&gt; <i>; yields {ty}</i>
Eli Friedmanff030482011-07-28 21:48:00 +00004793</pre>
4794
4795<h5>Overview:</h5>
4796<p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
4797It loads a value in memory and compares it to a given value. If they are
4798equal, it stores a new value into the memory.</p>
4799
4800<h5>Arguments:</h5>
4801<p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
4802address to operate on, a value to compare to the value currently be at that
4803address, and a new value to place at that address if the compared values are
4804equal. The type of '<var>&lt;cmp&gt;</var>' must be an integer type whose
4805bit width is a power of two greater than or equal to eight and less than
4806or equal to a target-specific size limit. '<var>&lt;cmp&gt;</var>' and
4807'<var>&lt;new&gt;</var>' must have the same type, and the type of
4808'<var>&lt;pointer&gt;</var>' must be a pointer to that type. If the
4809<code>cmpxchg</code> is marked as <code>volatile</code>, then the
4810optimizer is not allowed to modify the number or order of execution
4811of this <code>cmpxchg</code> with other <a href="#volatile">volatile
4812operations</a>.</p>
4813
4814<!-- FIXME: Extend allowed types. -->
4815
4816<p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
4817<code>cmpxchg</code> synchronizes with other atomic operations.</p>
4818
4819<p>The optional "<code>singlethread</code>" argument declares that the
4820<code>cmpxchg</code> is only atomic with respect to code (usually signal
4821handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
4822cmpxchg is atomic with respect to all other code in the system.</p>
4823
4824<p>The pointer passed into cmpxchg must have alignment greater than or equal to
4825the size in memory of the operand.
4826
4827<h5>Semantics:</h5>
4828<p>The contents of memory at the location specified by the
4829'<tt>&lt;pointer&gt;</tt>' operand is read and compared to
4830'<tt>&lt;cmp&gt;</tt>'; if the read value is the equal,
4831'<tt>&lt;new&gt;</tt>' is written. The original value at the location
4832is returned.
4833
4834<p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
4835purpose of identifying <a href="#release_sequence">release sequences</a>. A
4836failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
4837parameter determined by dropping any <code>release</code> part of the
4838<code>cmpxchg</code>'s ordering.</p>
4839
4840<!--
4841FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
4842optimization work on ARM.)
4843
4844FIXME: Is a weaker ordering constraint on failure helpful in practice?
4845-->
4846
4847<h5>Example:</h5>
4848<pre>
4849entry:
4850 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
4851 <a href="#i_br">br</a> label %loop
4852
4853loop:
4854 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
4855 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
4856 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
4857 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
4858 <a href="#i_br">br</a> i1 %success, label %done, label %loop
4859
4860done:
4861 ...
4862</pre>
4863
4864</div>
4865
4866<!-- _______________________________________________________________________ -->
NAKAMURA Takumib5bb29c2011-08-12 06:17:17 +00004867<h4>
4868<a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
4869</h4>
Eli Friedmanff030482011-07-28 21:48:00 +00004870
NAKAMURA Takumib5bb29c2011-08-12 06:17:17 +00004871<div>
Eli Friedmanff030482011-07-28 21:48:00 +00004872
4873<h5>Syntax:</h5>
4874<pre>
Eli Friedmanf03bb262011-08-12 22:50:01 +00004875 atomicrmw [volatile] &lt;operation&gt; &lt;ty&gt;* &lt;pointer&gt;, &lt;ty&gt; &lt;value&gt; [singlethread] &lt;ordering&gt; <i>; yields {ty}</i>
Eli Friedmanff030482011-07-28 21:48:00 +00004876</pre>
4877
4878<h5>Overview:</h5>
4879<p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
4880
4881<h5>Arguments:</h5>
4882<p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
4883operation to apply, an address whose value to modify, an argument to the
4884operation. The operation must be one of the following keywords:</p>
4885<ul>
4886 <li>xchg</li>
4887 <li>add</li>
4888 <li>sub</li>
4889 <li>and</li>
4890 <li>nand</li>
4891 <li>or</li>
4892 <li>xor</li>
4893 <li>max</li>
4894 <li>min</li>
4895 <li>umax</li>
4896 <li>umin</li>
4897</ul>
4898
4899<p>The type of '<var>&lt;value&gt;</var>' must be an integer type whose
4900bit width is a power of two greater than or equal to eight and less than
4901or equal to a target-specific size limit. The type of the
4902'<code>&lt;pointer&gt;</code>' operand must be a pointer to that type.
4903If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
4904optimizer is not allowed to modify the number or order of execution of this
4905<code>atomicrmw</code> with other <a href="#volatile">volatile
4906 operations</a>.</p>
4907
4908<!-- FIXME: Extend allowed types. -->
4909
4910<h5>Semantics:</h5>
4911<p>The contents of memory at the location specified by the
4912'<tt>&lt;pointer&gt;</tt>' operand are atomically read, modified, and written
4913back. The original value at the location is returned. The modification is
4914specified by the <var>operation</var> argument:</p>
4915
4916<ul>
4917 <li>xchg: <code>*ptr = val</code></li>
4918 <li>add: <code>*ptr = *ptr + val</code></li>
4919 <li>sub: <code>*ptr = *ptr - val</code></li>
4920 <li>and: <code>*ptr = *ptr &amp; val</code></li>
4921 <li>nand: <code>*ptr = ~(*ptr &amp; val)</code></li>
4922 <li>or: <code>*ptr = *ptr | val</code></li>
4923 <li>xor: <code>*ptr = *ptr ^ val</code></li>
4924 <li>max: <code>*ptr = *ptr &gt; val ? *ptr : val</code> (using a signed comparison)</li>
4925 <li>min: <code>*ptr = *ptr &lt; val ? *ptr : val</code> (using a signed comparison)</li>
4926 <li>umax: <code>*ptr = *ptr &gt; val ? *ptr : val</code> (using an unsigned comparison)</li>
4927 <li>umin: <code>*ptr = *ptr &lt; val ? *ptr : val</code> (using an unsigned comparison)</li>
4928</ul>
4929
4930<h5>Example:</h5>
4931<pre>
4932 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
4933</pre>
4934
4935</div>
4936
4937<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004938<h4>
Chris Lattnerf74d5c72004-04-05 01:30:49 +00004939 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00004940</h4>
Chris Lattnerf74d5c72004-04-05 01:30:49 +00004941
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00004942<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004943
Chris Lattner7faa8832002-04-14 06:13:44 +00004944<h5>Syntax:</h5>
Chris Lattnerf74d5c72004-04-05 01:30:49 +00004945<pre>
Matthijs Kooijmane49d0bc2008-10-13 13:44:15 +00004946 &lt;result&gt; = getelementptr &lt;pty&gt;* &lt;ptrval&gt;{, &lt;ty&gt; &lt;idx&gt;}*
Dan Gohmandd8004d2009-07-27 21:53:46 +00004947 &lt;result&gt; = getelementptr inbounds &lt;pty&gt;* &lt;ptrval&gt;{, &lt;ty&gt; &lt;idx&gt;}*
Chris Lattnerf74d5c72004-04-05 01:30:49 +00004948</pre>
4949
Chris Lattner7faa8832002-04-14 06:13:44 +00004950<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004951<p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
Chris Lattnerfdfeb692010-02-12 20:49:41 +00004952 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
4953 It performs address calculation only and does not access memory.</p>
Chris Lattnerf74d5c72004-04-05 01:30:49 +00004954
Chris Lattner7faa8832002-04-14 06:13:44 +00004955<h5>Arguments:</h5>
Matthijs Kooijmane49d0bc2008-10-13 13:44:15 +00004956<p>The first argument is always a pointer, and forms the basis of the
Chris Lattnerc8eef442009-07-29 06:44:13 +00004957 calculation. The remaining arguments are indices that indicate which of the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004958 elements of the aggregate object are indexed. The interpretation of each
4959 index is dependent on the type being indexed into. The first index always
4960 indexes the pointer value given as the first argument, the second index
4961 indexes a value of the type pointed to (not necessarily the value directly
4962 pointed to, since the first index can be non-zero), etc. The first type
Chris Lattnerfdfeb692010-02-12 20:49:41 +00004963 indexed into must be a pointer value, subsequent types can be arrays,
Chris Lattner61c70e92010-08-28 04:09:24 +00004964 vectors, and structs. Note that subsequent types being indexed into
Chris Lattnerfdfeb692010-02-12 20:49:41 +00004965 can never be pointers, since that would require loading the pointer before
4966 continuing calculation.</p>
Matthijs Kooijmane49d0bc2008-10-13 13:44:15 +00004967
4968<p>The type of each index argument depends on the type it is indexing into.
Chris Lattner61c70e92010-08-28 04:09:24 +00004969 When indexing into a (optionally packed) structure, only <tt>i32</tt>
Chris Lattnerfdfeb692010-02-12 20:49:41 +00004970 integer <b>constants</b> are allowed. When indexing into an array, pointer
4971 or vector, integers of any width are allowed, and they are not required to be
Eli Friedman266246c2011-08-12 23:37:55 +00004972 constant. These integers are treated as signed values where relevant.</p>
Chris Lattnerf74d5c72004-04-05 01:30:49 +00004973
Bill Wendlinge910b4c2009-07-20 02:29:24 +00004974<p>For example, let's consider a C code fragment and how it gets compiled to
4975 LLVM:</p>
Chris Lattnerf74d5c72004-04-05 01:30:49 +00004976
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00004977<pre class="doc_code">
Bill Wendling2f7a8b02007-05-29 09:04:49 +00004978struct RT {
4979 char A;
Chris Lattnercabc8462007-05-29 15:43:56 +00004980 int B[10][20];
Bill Wendling2f7a8b02007-05-29 09:04:49 +00004981 char C;
4982};
4983struct ST {
Chris Lattnercabc8462007-05-29 15:43:56 +00004984 int X;
Bill Wendling2f7a8b02007-05-29 09:04:49 +00004985 double Y;
4986 struct RT Z;
4987};
Chris Lattnerf74d5c72004-04-05 01:30:49 +00004988
Chris Lattnercabc8462007-05-29 15:43:56 +00004989int *foo(struct ST *s) {
Bill Wendling2f7a8b02007-05-29 09:04:49 +00004990 return &amp;s[1].Z.B[5][13];
4991}
Chris Lattnerf74d5c72004-04-05 01:30:49 +00004992</pre>
4993
Misha Brukman9d0919f2003-11-08 01:05:38 +00004994<p>The LLVM code generated by the GCC frontend is:</p>
Chris Lattnerf74d5c72004-04-05 01:30:49 +00004995
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00004996<pre class="doc_code">
Chris Lattnere7886e42009-01-11 20:53:49 +00004997%RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4998%ST = <a href="#namedtypes">type</a> { i32, double, %RT }
Chris Lattnerf74d5c72004-04-05 01:30:49 +00004999
Dan Gohman4df605b2009-07-25 02:23:48 +00005000define i32* @foo(%ST* %s) {
Bill Wendling2f7a8b02007-05-29 09:04:49 +00005001entry:
5002 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
5003 ret i32* %reg
5004}
Chris Lattnerf74d5c72004-04-05 01:30:49 +00005005</pre>
5006
Chris Lattner7faa8832002-04-14 06:13:44 +00005007<h5>Semantics:</h5>
Misha Brukman9d0919f2003-11-08 01:05:38 +00005008<p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005009 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
5010 }</tt>' type, a structure. The second index indexes into the third element
5011 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
5012 i8 }</tt>' type, another structure. The third index indexes into the second
5013 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
5014 array. The two dimensions of the array are subscripted into, yielding an
5015 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
5016 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
Chris Lattnerf74d5c72004-04-05 01:30:49 +00005017
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005018<p>Note that it is perfectly legal to index partially through a structure,
5019 returning a pointer to an inner element. Because of this, the LLVM code for
5020 the given testcase is equivalent to:</p>
Chris Lattnerf74d5c72004-04-05 01:30:49 +00005021
5022<pre>
Dan Gohman4df605b2009-07-25 02:23:48 +00005023 define i32* @foo(%ST* %s) {
Reid Spencerca86e162006-12-31 07:07:53 +00005024 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
Jeff Cohen6f1cc772007-04-22 01:17:39 +00005025 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
5026 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
Reid Spencerca86e162006-12-31 07:07:53 +00005027 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5028 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5029 ret i32* %t5
Chris Lattnerf74d5c72004-04-05 01:30:49 +00005030 }
Chris Lattner6536cfe2002-05-06 22:08:29 +00005031</pre>
Chris Lattnere67a9512005-06-24 17:22:57 +00005032
Dan Gohmandd8004d2009-07-27 21:53:46 +00005033<p>If the <tt>inbounds</tt> keyword is present, the result value of the
Dan Gohman27ef9972010-04-23 15:23:32 +00005034 <tt>getelementptr</tt> is a <a href="#trapvalues">trap value</a> if the
5035 base pointer is not an <i>in bounds</i> address of an allocated object,
5036 or if any of the addresses that would be formed by successive addition of
5037 the offsets implied by the indices to the base address with infinitely
Eli Friedman266246c2011-08-12 23:37:55 +00005038 precise signed arithmetic are not an <i>in bounds</i> address of that
5039 allocated object. The <i>in bounds</i> addresses for an allocated object
5040 are all the addresses that point into the object, plus the address one
5041 byte past the end.</p>
Dan Gohmandd8004d2009-07-27 21:53:46 +00005042
5043<p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
Eli Friedman266246c2011-08-12 23:37:55 +00005044 the base address with silently-wrapping two's complement arithmetic. If the
5045 offsets have a different width from the pointer, they are sign-extended or
5046 truncated to the width of the pointer. The result value of the
5047 <tt>getelementptr</tt> may be outside the object pointed to by the base
5048 pointer. The result value may not necessarily be used to access memory
5049 though, even if it happens to point into allocated storage. See the
5050 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5051 information.</p>
Dan Gohmandd8004d2009-07-27 21:53:46 +00005052
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005053<p>The getelementptr instruction is often confusing. For some more insight into
5054 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
Chris Lattner884a9702006-08-15 00:45:58 +00005055
Chris Lattner7faa8832002-04-14 06:13:44 +00005056<h5>Example:</h5>
Chris Lattnerf74d5c72004-04-05 01:30:49 +00005057<pre>
Reid Spencerca86e162006-12-31 07:07:53 +00005058 <i>; yields [12 x i8]*:aptr</i>
Matthijs Kooijmane49d0bc2008-10-13 13:44:15 +00005059 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5060 <i>; yields i8*:vptr</i>
Dan Gohman0e451ce2008-10-14 16:51:45 +00005061 %vptr = getelementptr {i32, &lt;2 x i8&gt;}* %svptr, i64 0, i32 1, i32 1
Matthijs Kooijmane49d0bc2008-10-13 13:44:15 +00005062 <i>; yields i8*:eptr</i>
5063 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
Sanjiv Gupta9f805c22009-04-25 07:27:44 +00005064 <i>; yields i32*:iptr</i>
Sanjiv Gupta16ffa802009-04-24 16:38:13 +00005065 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
Chris Lattnerf74d5c72004-04-05 01:30:49 +00005066</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005067
Chris Lattnerf74d5c72004-04-05 01:30:49 +00005068</div>
Reid Spencer47ce1792006-11-09 21:15:49 +00005069
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005070</div>
5071
Chris Lattner00950542001-06-06 20:29:01 +00005072<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005073<h3>
5074 <a name="convertops">Conversion Operations</a>
5075</h3>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005076
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005077<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005078
Reid Spencer2fd21e62006-11-08 01:18:52 +00005079<p>The instructions in this category are the conversion instructions (casting)
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005080 which all take a single operand and a type. They perform various bit
5081 conversions on the operand.</p>
5082
Chris Lattner6536cfe2002-05-06 22:08:29 +00005083<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005084<h4>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005085 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005086</h4>
5087
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005088<div>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005089
5090<h5>Syntax:</h5>
5091<pre>
5092 &lt;result&gt; = trunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
5093</pre>
5094
5095<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005096<p>The '<tt>trunc</tt>' instruction truncates its operand to the
5097 type <tt>ty2</tt>.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005098
5099<h5>Arguments:</h5>
Nadav Rotem8c20ec52011-02-24 21:01:34 +00005100<p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5101 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5102 of the same number of integers.
5103 The bit size of the <tt>value</tt> must be larger than
5104 the bit size of the destination type, <tt>ty2</tt>.
5105 Equal sized types are not allowed.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005106
5107<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005108<p>The '<tt>trunc</tt>' instruction truncates the high order bits
5109 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5110 source size must be larger than the destination size, <tt>trunc</tt> cannot
5111 be a <i>no-op cast</i>. It will always truncate bits.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005112
5113<h5>Example:</h5>
5114<pre>
Nadav Rotem8c20ec52011-02-24 21:01:34 +00005115 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5116 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5117 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5118 %W = trunc &lt;2 x i16&gt; &lt;i16 8, i16 7&gt; to &lt;2 x i8&gt; <i>; yields &lt;i8 8, i8 7&gt;</i>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005119</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005120
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005121</div>
5122
5123<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005124<h4>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005125 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005126</h4>
5127
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005128<div>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005129
5130<h5>Syntax:</h5>
5131<pre>
5132 &lt;result&gt; = zext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
5133</pre>
5134
5135<h5>Overview:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005136<p>The '<tt>zext</tt>' instruction zero extends its operand to type
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005137 <tt>ty2</tt>.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005138
5139
5140<h5>Arguments:</h5>
Nadav Rotemed9b9342011-02-20 12:37:50 +00005141<p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5142 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5143 of the same number of integers.
5144 The bit size of the <tt>value</tt> must be smaller than
5145 the bit size of the destination type,
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005146 <tt>ty2</tt>.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005147
5148<h5>Semantics:</h5>
5149<p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005150 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005151
Reid Spencerb5929522007-01-12 15:46:11 +00005152<p>When zero extending from i1, the result will always be either 0 or 1.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005153
5154<h5>Example:</h5>
5155<pre>
Reid Spencerca86e162006-12-31 07:07:53 +00005156 %X = zext i32 257 to i64 <i>; yields i64:257</i>
Reid Spencerc78f3372007-01-12 03:35:51 +00005157 %Y = zext i1 true to i32 <i>; yields i32:1</i>
Nadav Rotemed9b9342011-02-20 12:37:50 +00005158 %Z = zext &lt;2 x i16&gt; &lt;i16 8, i16 7&gt; to &lt;2 x i32&gt; <i>; yields &lt;i32 8, i32 7&gt;</i>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005159</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005160
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005161</div>
5162
5163<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005164<h4>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005165 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005166</h4>
5167
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005168<div>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005169
5170<h5>Syntax:</h5>
5171<pre>
5172 &lt;result&gt; = sext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
5173</pre>
5174
5175<h5>Overview:</h5>
5176<p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5177
5178<h5>Arguments:</h5>
Nadav Rotem8c20ec52011-02-24 21:01:34 +00005179<p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5180 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5181 of the same number of integers.
5182 The bit size of the <tt>value</tt> must be smaller than
5183 the bit size of the destination type,
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005184 <tt>ty2</tt>.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005185
5186<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005187<p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5188 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5189 of the type <tt>ty2</tt>.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005190
Reid Spencerc78f3372007-01-12 03:35:51 +00005191<p>When sign extending from i1, the extension always results in -1 or 0.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005192
5193<h5>Example:</h5>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005194<pre>
Reid Spencerca86e162006-12-31 07:07:53 +00005195 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
Reid Spencerc78f3372007-01-12 03:35:51 +00005196 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
Nadav Rotem8c20ec52011-02-24 21:01:34 +00005197 %Z = sext &lt;2 x i16&gt; &lt;i16 8, i16 7&gt; to &lt;2 x i32&gt; <i>; yields &lt;i32 8, i32 7&gt;</i>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005198</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005199
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005200</div>
5201
5202<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005203<h4>
Reid Spencer3fa91b02006-11-09 21:48:10 +00005204 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005205</h4>
Reid Spencer3fa91b02006-11-09 21:48:10 +00005206
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005207<div>
Reid Spencer3fa91b02006-11-09 21:48:10 +00005208
5209<h5>Syntax:</h5>
Reid Spencer3fa91b02006-11-09 21:48:10 +00005210<pre>
5211 &lt;result&gt; = fptrunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
5212</pre>
5213
5214<h5>Overview:</h5>
5215<p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005216 <tt>ty2</tt>.</p>
Reid Spencer3fa91b02006-11-09 21:48:10 +00005217
5218<h5>Arguments:</h5>
5219<p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005220 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5221 to cast it to. The size of <tt>value</tt> must be larger than the size of
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005222 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005223 <i>no-op cast</i>.</p>
Reid Spencer3fa91b02006-11-09 21:48:10 +00005224
5225<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005226<p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005227 <a href="#t_floating">floating point</a> type to a smaller
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005228 <a href="#t_floating">floating point</a> type. If the value cannot fit
5229 within the destination type, <tt>ty2</tt>, then the results are
5230 undefined.</p>
Reid Spencer3fa91b02006-11-09 21:48:10 +00005231
5232<h5>Example:</h5>
5233<pre>
5234 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5235 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5236</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005237
Reid Spencer3fa91b02006-11-09 21:48:10 +00005238</div>
5239
5240<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005241<h4>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005242 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005243</h4>
5244
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005245<div>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005246
5247<h5>Syntax:</h5>
5248<pre>
5249 &lt;result&gt; = fpext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
5250</pre>
5251
5252<h5>Overview:</h5>
5253<p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005254 floating point value.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005255
5256<h5>Arguments:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005257<p>The '<tt>fpext</tt>' instruction takes a
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005258 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5259 a <a href="#t_floating">floating point</a> type to cast it to. The source
5260 type must be smaller than the destination type.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005261
5262<h5>Semantics:</h5>
Reid Spencerd4448792006-11-09 23:03:26 +00005263<p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005264 <a href="#t_floating">floating point</a> type to a larger
5265 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5266 used to make a <i>no-op cast</i> because it always changes bits. Use
5267 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005268
5269<h5>Example:</h5>
5270<pre>
Nick Lewycky5bb3ece2011-03-31 18:20:19 +00005271 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5272 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005273</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005274
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005275</div>
5276
5277<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005278<h4>
Reid Spencer24d6da52007-01-21 00:29:26 +00005279 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005280</h4>
5281
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005282<div>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005283
5284<h5>Syntax:</h5>
5285<pre>
Reid Spencer1539a1c2007-07-31 14:40:14 +00005286 &lt;result&gt; = fptoui &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005287</pre>
5288
5289<h5>Overview:</h5>
Reid Spencer1539a1c2007-07-31 14:40:14 +00005290<p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005291 unsigned integer equivalent of type <tt>ty2</tt>.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005292
5293<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005294<p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5295 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5296 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5297 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5298 vector integer type with the same number of elements as <tt>ty</tt></p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005299
5300<h5>Semantics:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005301<p>The '<tt>fptoui</tt>' instruction converts its
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005302 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5303 towards zero) unsigned integer value. If the value cannot fit
5304 in <tt>ty2</tt>, the results are undefined.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005305
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005306<h5>Example:</h5>
5307<pre>
Reid Spencer1539a1c2007-07-31 14:40:14 +00005308 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
Chris Lattner88519042007-09-22 03:17:52 +00005309 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
Gabor Greif6a292012009-10-28 09:21:30 +00005310 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005311</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005312
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005313</div>
5314
5315<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005316<h4>
Reid Spencerd4448792006-11-09 23:03:26 +00005317 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005318</h4>
5319
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005320<div>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005321
5322<h5>Syntax:</h5>
5323<pre>
Reid Spencerd4448792006-11-09 23:03:26 +00005324 &lt;result&gt; = fptosi &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005325</pre>
5326
5327<h5>Overview:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005328<p>The '<tt>fptosi</tt>' instruction converts
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005329 <a href="#t_floating">floating point</a> <tt>value</tt> to
5330 type <tt>ty2</tt>.</p>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005331
Chris Lattner6536cfe2002-05-06 22:08:29 +00005332<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005333<p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5334 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5335 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5336 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5337 vector integer type with the same number of elements as <tt>ty</tt></p>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005338
Chris Lattner6536cfe2002-05-06 22:08:29 +00005339<h5>Semantics:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005340<p>The '<tt>fptosi</tt>' instruction converts its
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005341 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5342 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5343 the results are undefined.</p>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005344
Chris Lattner33ba0d92001-07-09 00:26:23 +00005345<h5>Example:</h5>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005346<pre>
Reid Spencerc78f3372007-01-12 03:35:51 +00005347 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
Chris Lattner88519042007-09-22 03:17:52 +00005348 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
Gabor Greif6a292012009-10-28 09:21:30 +00005349 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005350</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005351
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005352</div>
5353
5354<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005355<h4>
Reid Spencerd4448792006-11-09 23:03:26 +00005356 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005357</h4>
5358
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005359<div>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005360
5361<h5>Syntax:</h5>
5362<pre>
Reid Spencerd4448792006-11-09 23:03:26 +00005363 &lt;result&gt; = uitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005364</pre>
5365
5366<h5>Overview:</h5>
Reid Spencerd4448792006-11-09 23:03:26 +00005367<p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005368 integer and converts that value to the <tt>ty2</tt> type.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005369
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005370<h5>Arguments:</h5>
Nate Begemanb348d182007-11-17 03:58:34 +00005371<p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005372 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5373 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5374 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5375 floating point type with the same number of elements as <tt>ty</tt></p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005376
5377<h5>Semantics:</h5>
Reid Spencerd4448792006-11-09 23:03:26 +00005378<p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005379 integer quantity and converts it to the corresponding floating point
5380 value. If the value cannot fit in the floating point value, the results are
5381 undefined.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005382
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005383<h5>Example:</h5>
5384<pre>
Reid Spencerca86e162006-12-31 07:07:53 +00005385 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
Dan Gohman0e451ce2008-10-14 16:51:45 +00005386 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005387</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005388
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005389</div>
5390
5391<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005392<h4>
Reid Spencerd4448792006-11-09 23:03:26 +00005393 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005394</h4>
5395
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005396<div>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005397
5398<h5>Syntax:</h5>
5399<pre>
Reid Spencerd4448792006-11-09 23:03:26 +00005400 &lt;result&gt; = sitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005401</pre>
5402
5403<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005404<p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5405 and converts that value to the <tt>ty2</tt> type.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005406
5407<h5>Arguments:</h5>
Nate Begemanb348d182007-11-17 03:58:34 +00005408<p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005409 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5410 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5411 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5412 floating point type with the same number of elements as <tt>ty</tt></p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005413
5414<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005415<p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5416 quantity and converts it to the corresponding floating point value. If the
5417 value cannot fit in the floating point value, the results are undefined.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005418
5419<h5>Example:</h5>
5420<pre>
Reid Spencerca86e162006-12-31 07:07:53 +00005421 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
Dan Gohman0e451ce2008-10-14 16:51:45 +00005422 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005423</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005424
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005425</div>
5426
5427<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005428<h4>
Reid Spencer72679252006-11-11 21:00:47 +00005429 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005430</h4>
5431
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005432<div>
Reid Spencer72679252006-11-11 21:00:47 +00005433
5434<h5>Syntax:</h5>
5435<pre>
5436 &lt;result&gt; = ptrtoint &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
5437</pre>
5438
5439<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005440<p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
5441 the integer type <tt>ty2</tt>.</p>
Reid Spencer72679252006-11-11 21:00:47 +00005442
5443<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005444<p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5445 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
5446 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
Reid Spencer72679252006-11-11 21:00:47 +00005447
5448<h5>Semantics:</h5>
5449<p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005450 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5451 truncating or zero extending that value to the size of the integer type. If
5452 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5453 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5454 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5455 change.</p>
Reid Spencer72679252006-11-11 21:00:47 +00005456
5457<h5>Example:</h5>
5458<pre>
Jeff Cohenb627eab2007-04-29 01:07:00 +00005459 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
5460 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
Reid Spencer72679252006-11-11 21:00:47 +00005461</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005462
Reid Spencer72679252006-11-11 21:00:47 +00005463</div>
5464
5465<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005466<h4>
Reid Spencer72679252006-11-11 21:00:47 +00005467 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005468</h4>
5469
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005470<div>
Reid Spencer72679252006-11-11 21:00:47 +00005471
5472<h5>Syntax:</h5>
5473<pre>
5474 &lt;result&gt; = inttoptr &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
5475</pre>
5476
5477<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005478<p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5479 pointer type, <tt>ty2</tt>.</p>
Reid Spencer72679252006-11-11 21:00:47 +00005480
5481<h5>Arguments:</h5>
Duncan Sands8036ca42007-03-30 12:22:09 +00005482<p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005483 value to cast, and a type to cast it to, which must be a
5484 <a href="#t_pointer">pointer</a> type.</p>
Reid Spencer72679252006-11-11 21:00:47 +00005485
5486<h5>Semantics:</h5>
5487<p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005488 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5489 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5490 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5491 than the size of a pointer then a zero extension is done. If they are the
5492 same size, nothing is done (<i>no-op cast</i>).</p>
Reid Spencer72679252006-11-11 21:00:47 +00005493
5494<h5>Example:</h5>
5495<pre>
Jeff Cohenb627eab2007-04-29 01:07:00 +00005496 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
Gabor Greif6a292012009-10-28 09:21:30 +00005497 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5498 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
Reid Spencer72679252006-11-11 21:00:47 +00005499</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005500
Reid Spencer72679252006-11-11 21:00:47 +00005501</div>
5502
5503<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005504<h4>
Reid Spencer5c0ef472006-11-11 23:08:07 +00005505 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005506</h4>
5507
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005508<div>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005509
5510<h5>Syntax:</h5>
5511<pre>
Reid Spencer5c0ef472006-11-11 23:08:07 +00005512 &lt;result&gt; = bitcast &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005513</pre>
5514
5515<h5>Overview:</h5>
Reid Spencer5c0ef472006-11-11 23:08:07 +00005516<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005517 <tt>ty2</tt> without changing any bits.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005518
5519<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005520<p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5521 non-aggregate first class value, and a type to cast it to, which must also be
5522 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5523 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5524 identical. If the source type is a pointer, the destination type must also be
5525 a pointer. This instruction supports bitwise conversion of vectors to
5526 integers and to vectors of other types (as long as they have the same
5527 size).</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005528
5529<h5>Semantics:</h5>
Reid Spencer5c0ef472006-11-11 23:08:07 +00005530<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005531 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5532 this conversion. The conversion is done as if the <tt>value</tt> had been
5533 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
5534 be converted to other pointer types with this instruction. To convert
5535 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5536 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
Reid Spencer9dee3ac2006-11-08 01:11:31 +00005537
5538<h5>Example:</h5>
5539<pre>
Jeff Cohenb627eab2007-04-29 01:07:00 +00005540 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
Reid Spencerca86e162006-12-31 07:07:53 +00005541 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005542 %Z = bitcast &lt;2 x int&gt; %V to i64; <i>; yields i64: %V</i>
Chris Lattner33ba0d92001-07-09 00:26:23 +00005543</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005544
Misha Brukman9d0919f2003-11-08 01:05:38 +00005545</div>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005546
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005547</div>
5548
Reid Spencer2fd21e62006-11-08 01:18:52 +00005549<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005550<h3>
5551 <a name="otherops">Other Operations</a>
5552</h3>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005553
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005554<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005555
5556<p>The instructions in this category are the "miscellaneous" instructions, which
5557 defy better classification.</p>
5558
Reid Spencerf3a70a62006-11-18 21:50:54 +00005559<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005560<h4>
5561 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5562</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005563
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005564<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005565
Reid Spencerf3a70a62006-11-18 21:50:54 +00005566<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005567<pre>
5568 &lt;result&gt; = icmp &lt;cond&gt; &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {i1} or {&lt;N x i1&gt;}:result</i>
Reid Spencerf3a70a62006-11-18 21:50:54 +00005569</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005570
Reid Spencerf3a70a62006-11-18 21:50:54 +00005571<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005572<p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5573 boolean values based on comparison of its two integer, integer vector, or
5574 pointer operands.</p>
5575
Reid Spencerf3a70a62006-11-18 21:50:54 +00005576<h5>Arguments:</h5>
5577<p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005578 the condition code indicating the kind of comparison to perform. It is not a
5579 value, just a keyword. The possible condition code are:</p>
5580
Reid Spencerf3a70a62006-11-18 21:50:54 +00005581<ol>
5582 <li><tt>eq</tt>: equal</li>
5583 <li><tt>ne</tt>: not equal </li>
5584 <li><tt>ugt</tt>: unsigned greater than</li>
5585 <li><tt>uge</tt>: unsigned greater or equal</li>
5586 <li><tt>ult</tt>: unsigned less than</li>
5587 <li><tt>ule</tt>: unsigned less or equal</li>
5588 <li><tt>sgt</tt>: signed greater than</li>
5589 <li><tt>sge</tt>: signed greater or equal</li>
5590 <li><tt>slt</tt>: signed less than</li>
5591 <li><tt>sle</tt>: signed less or equal</li>
5592</ol>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005593
Chris Lattner3b19d652007-01-15 01:54:13 +00005594<p>The remaining two arguments must be <a href="#t_integer">integer</a> or
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005595 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5596 typed. They must also be identical types.</p>
5597
Reid Spencerf3a70a62006-11-18 21:50:54 +00005598<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005599<p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5600 condition code given as <tt>cond</tt>. The comparison performed always yields
Nick Lewyckyec38da42009-09-27 00:45:11 +00005601 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005602 result, as follows:</p>
5603
Reid Spencerf3a70a62006-11-18 21:50:54 +00005604<ol>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005605 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005606 <tt>false</tt> otherwise. No sign interpretation is necessary or
5607 performed.</li>
5608
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005609 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005610 <tt>false</tt> otherwise. No sign interpretation is necessary or
5611 performed.</li>
5612
Reid Spencerf3a70a62006-11-18 21:50:54 +00005613 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005614 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5615
Reid Spencerf3a70a62006-11-18 21:50:54 +00005616 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005617 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5618 to <tt>op2</tt>.</li>
5619
Reid Spencerf3a70a62006-11-18 21:50:54 +00005620 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005621 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5622
Reid Spencerf3a70a62006-11-18 21:50:54 +00005623 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005624 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5625
Reid Spencerf3a70a62006-11-18 21:50:54 +00005626 <li><tt>sgt</tt>: interprets the operands as signed values and yields
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005627 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5628
Reid Spencerf3a70a62006-11-18 21:50:54 +00005629 <li><tt>sge</tt>: interprets the operands as signed values and yields
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005630 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5631 to <tt>op2</tt>.</li>
5632
Reid Spencerf3a70a62006-11-18 21:50:54 +00005633 <li><tt>slt</tt>: interprets the operands as signed values and yields
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005634 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5635
Reid Spencerf3a70a62006-11-18 21:50:54 +00005636 <li><tt>sle</tt>: interprets the operands as signed values and yields
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005637 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
Reid Spencerf3a70a62006-11-18 21:50:54 +00005638</ol>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005639
Reid Spencerf3a70a62006-11-18 21:50:54 +00005640<p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005641 values are compared as if they were integers.</p>
5642
5643<p>If the operands are integer vectors, then they are compared element by
5644 element. The result is an <tt>i1</tt> vector with the same number of elements
5645 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
Reid Spencerf3a70a62006-11-18 21:50:54 +00005646
5647<h5>Example:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005648<pre>
5649 &lt;result&gt; = icmp eq i32 4, 5 <i>; yields: result=false</i>
Reid Spencerca86e162006-12-31 07:07:53 +00005650 &lt;result&gt; = icmp ne float* %X, %X <i>; yields: result=false</i>
5651 &lt;result&gt; = icmp ult i16 4, 5 <i>; yields: result=true</i>
5652 &lt;result&gt; = icmp sgt i16 4, 5 <i>; yields: result=false</i>
5653 &lt;result&gt; = icmp ule i16 -4, 5 <i>; yields: result=false</i>
5654 &lt;result&gt; = icmp sge i16 4, 5 <i>; yields: result=false</i>
Reid Spencerf3a70a62006-11-18 21:50:54 +00005655</pre>
Dan Gohmanc766f722009-01-22 01:39:38 +00005656
5657<p>Note that the code generator does not yet support vector types with
5658 the <tt>icmp</tt> instruction.</p>
5659
Reid Spencerf3a70a62006-11-18 21:50:54 +00005660</div>
5661
5662<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005663<h4>
5664 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
5665</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005666
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005667<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005668
Reid Spencerf3a70a62006-11-18 21:50:54 +00005669<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005670<pre>
5671 &lt;result&gt; = fcmp &lt;cond&gt; &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {i1} or {&lt;N x i1&gt;}:result</i>
Reid Spencerf3a70a62006-11-18 21:50:54 +00005672</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005673
Reid Spencerf3a70a62006-11-18 21:50:54 +00005674<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005675<p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
5676 values based on comparison of its operands.</p>
5677
5678<p>If the operands are floating point scalars, then the result type is a boolean
Nick Lewyckyec38da42009-09-27 00:45:11 +00005679(<a href="#t_integer"><tt>i1</tt></a>).</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005680
5681<p>If the operands are floating point vectors, then the result type is a vector
5682 of boolean with the same number of elements as the operands being
5683 compared.</p>
5684
Reid Spencerf3a70a62006-11-18 21:50:54 +00005685<h5>Arguments:</h5>
5686<p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005687 the condition code indicating the kind of comparison to perform. It is not a
5688 value, just a keyword. The possible condition code are:</p>
5689
Reid Spencerf3a70a62006-11-18 21:50:54 +00005690<ol>
Reid Spencerb7f26282006-11-19 03:00:14 +00005691 <li><tt>false</tt>: no comparison, always returns false</li>
Reid Spencerf3a70a62006-11-18 21:50:54 +00005692 <li><tt>oeq</tt>: ordered and equal</li>
5693 <li><tt>ogt</tt>: ordered and greater than </li>
5694 <li><tt>oge</tt>: ordered and greater than or equal</li>
5695 <li><tt>olt</tt>: ordered and less than </li>
5696 <li><tt>ole</tt>: ordered and less than or equal</li>
5697 <li><tt>one</tt>: ordered and not equal</li>
5698 <li><tt>ord</tt>: ordered (no nans)</li>
5699 <li><tt>ueq</tt>: unordered or equal</li>
5700 <li><tt>ugt</tt>: unordered or greater than </li>
5701 <li><tt>uge</tt>: unordered or greater than or equal</li>
5702 <li><tt>ult</tt>: unordered or less than </li>
5703 <li><tt>ule</tt>: unordered or less than or equal</li>
5704 <li><tt>une</tt>: unordered or not equal</li>
5705 <li><tt>uno</tt>: unordered (either nans)</li>
Reid Spencerb7f26282006-11-19 03:00:14 +00005706 <li><tt>true</tt>: no comparison, always returns true</li>
Reid Spencerf3a70a62006-11-18 21:50:54 +00005707</ol>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005708
Jeff Cohenb627eab2007-04-29 01:07:00 +00005709<p><i>Ordered</i> means that neither operand is a QNAN while
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005710 <i>unordered</i> means that either operand may be a QNAN.</p>
5711
5712<p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
5713 a <a href="#t_floating">floating point</a> type or
5714 a <a href="#t_vector">vector</a> of floating point type. They must have
5715 identical types.</p>
5716
Reid Spencerf3a70a62006-11-18 21:50:54 +00005717<h5>Semantics:</h5>
Gabor Greiffb224a22008-08-07 21:46:00 +00005718<p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005719 according to the condition code given as <tt>cond</tt>. If the operands are
5720 vectors, then the vectors are compared element by element. Each comparison
Nick Lewyckyec38da42009-09-27 00:45:11 +00005721 performed always yields an <a href="#t_integer">i1</a> result, as
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005722 follows:</p>
5723
Reid Spencerf3a70a62006-11-18 21:50:54 +00005724<ol>
5725 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005726
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005727 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005728 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5729
Reid Spencerb7f26282006-11-19 03:00:14 +00005730 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
Dan Gohmanb55a1ee2010-03-01 17:41:39 +00005731 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005732
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005733 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005734 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5735
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005736 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005737 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5738
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005739 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005740 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5741
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005742 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005743 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5744
Reid Spencerb7f26282006-11-19 03:00:14 +00005745 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005746
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005747 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005748 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5749
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005750 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005751 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5752
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005753 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005754 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5755
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005756 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005757 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5758
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005759 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005760 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5761
Eric Christopher6c7e8a02009-12-05 02:46:03 +00005762 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005763 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5764
Reid Spencerb7f26282006-11-19 03:00:14 +00005765 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005766
Reid Spencerf3a70a62006-11-18 21:50:54 +00005767 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5768</ol>
Reid Spencerf3a70a62006-11-18 21:50:54 +00005769
5770<h5>Example:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005771<pre>
5772 &lt;result&gt; = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
Dan Gohmanf72fb672008-09-09 01:02:47 +00005773 &lt;result&gt; = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5774 &lt;result&gt; = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5775 &lt;result&gt; = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
Reid Spencerf3a70a62006-11-18 21:50:54 +00005776</pre>
Dan Gohmanc766f722009-01-22 01:39:38 +00005777
5778<p>Note that the code generator does not yet support vector types with
5779 the <tt>fcmp</tt> instruction.</p>
5780
Reid Spencerf3a70a62006-11-18 21:50:54 +00005781</div>
5782
Reid Spencer2fd21e62006-11-08 01:18:52 +00005783<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005784<h4>
Chris Lattner5568e942008-05-20 20:48:21 +00005785 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005786</h4>
Chris Lattner5568e942008-05-20 20:48:21 +00005787
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005788<div>
Chris Lattner5568e942008-05-20 20:48:21 +00005789
Reid Spencer2fd21e62006-11-08 01:18:52 +00005790<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005791<pre>
5792 &lt;result&gt; = phi &lt;ty&gt; [ &lt;val0&gt;, &lt;label0&gt;], ...
5793</pre>
Chris Lattner5568e942008-05-20 20:48:21 +00005794
Reid Spencer2fd21e62006-11-08 01:18:52 +00005795<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005796<p>The '<tt>phi</tt>' instruction is used to implement the &#966; node in the
5797 SSA graph representing the function.</p>
5798
Reid Spencer2fd21e62006-11-08 01:18:52 +00005799<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005800<p>The type of the incoming values is specified with the first type field. After
5801 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5802 one pair for each predecessor basic block of the current block. Only values
5803 of <a href="#t_firstclass">first class</a> type may be used as the value
5804 arguments to the PHI node. Only labels may be used as the label
5805 arguments.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00005806
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005807<p>There must be no non-phi instructions between the start of a basic block and
5808 the PHI instructions: i.e. PHI instructions must be first in a basic
5809 block.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00005810
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005811<p>For the purposes of the SSA form, the use of each incoming value is deemed to
5812 occur on the edge from the corresponding predecessor block to the current
5813 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5814 value on the same edge).</p>
Jay Foadd2449092009-06-03 10:20:10 +00005815
Reid Spencer2fd21e62006-11-08 01:18:52 +00005816<h5>Semantics:</h5>
Jeff Cohenb627eab2007-04-29 01:07:00 +00005817<p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005818 specified by the pair corresponding to the predecessor basic block that
5819 executed just prior to the current block.</p>
Chris Lattner5568e942008-05-20 20:48:21 +00005820
Reid Spencer2fd21e62006-11-08 01:18:52 +00005821<h5>Example:</h5>
Chris Lattner5568e942008-05-20 20:48:21 +00005822<pre>
5823Loop: ; Infinite loop that counts from 0 on up...
5824 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5825 %nextindvar = add i32 %indvar, 1
5826 br label %Loop
5827</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005828
Reid Spencer2fd21e62006-11-08 01:18:52 +00005829</div>
5830
Chris Lattnercc37aae2004-03-12 05:50:16 +00005831<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005832<h4>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005833 <a name="i_select">'<tt>select</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005834</h4>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005835
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005836<div>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005837
5838<h5>Syntax:</h5>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005839<pre>
Dan Gohmanf72fb672008-09-09 01:02:47 +00005840 &lt;result&gt; = select <i>selty</i> &lt;cond&gt;, &lt;ty&gt; &lt;val1&gt;, &lt;ty&gt; &lt;val2&gt; <i>; yields ty</i>
5841
Dan Gohman0e451ce2008-10-14 16:51:45 +00005842 <i>selty</i> is either i1 or {&lt;N x i1&gt;}
Chris Lattnercc37aae2004-03-12 05:50:16 +00005843</pre>
5844
5845<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005846<p>The '<tt>select</tt>' instruction is used to choose one value based on a
5847 condition, without branching.</p>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005848
5849
5850<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005851<p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5852 values indicating the condition, and two values of the
5853 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
5854 vectors and the condition is a scalar, then entire vectors are selected, not
5855 individual elements.</p>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005856
5857<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005858<p>If the condition is an i1 and it evaluates to 1, the instruction returns the
5859 first value argument; otherwise, it returns the second value argument.</p>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005860
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005861<p>If the condition is a vector of i1, then the value arguments must be vectors
5862 of the same size, and the selection is done element by element.</p>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005863
5864<h5>Example:</h5>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005865<pre>
Reid Spencerc78f3372007-01-12 03:35:51 +00005866 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
Chris Lattnercc37aae2004-03-12 05:50:16 +00005867</pre>
Dan Gohmanc766f722009-01-22 01:39:38 +00005868
5869<p>Note that the code generator does not yet support conditions
5870 with vector type.</p>
5871
Chris Lattnercc37aae2004-03-12 05:50:16 +00005872</div>
5873
Robert Bocchino05ccd702006-01-15 20:48:27 +00005874<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005875<h4>
Chris Lattner2bff5242005-05-06 05:47:36 +00005876 <a name="i_call">'<tt>call</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005877</h4>
Chris Lattner2bff5242005-05-06 05:47:36 +00005878
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005879<div>
Chris Lattner2bff5242005-05-06 05:47:36 +00005880
Chris Lattner00950542001-06-06 20:29:01 +00005881<h5>Syntax:</h5>
Chris Lattner2bff5242005-05-06 05:47:36 +00005882<pre>
Devang Patel307e8ab2008-10-07 17:48:33 +00005883 &lt;result&gt; = [tail] call [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] &lt;ty&gt; [&lt;fnty&gt;*] &lt;fnptrval&gt;(&lt;function args&gt;) [<a href="#fnattrs">fn attrs</a>]
Chris Lattner2bff5242005-05-06 05:47:36 +00005884</pre>
5885
Chris Lattner00950542001-06-06 20:29:01 +00005886<h5>Overview:</h5>
Misha Brukman9d0919f2003-11-08 01:05:38 +00005887<p>The '<tt>call</tt>' instruction represents a simple function call.</p>
Chris Lattner2bff5242005-05-06 05:47:36 +00005888
Chris Lattner00950542001-06-06 20:29:01 +00005889<h5>Arguments:</h5>
Misha Brukman9d0919f2003-11-08 01:05:38 +00005890<p>This instruction requires several arguments:</p>
Chris Lattner2bff5242005-05-06 05:47:36 +00005891
Chris Lattner6536cfe2002-05-06 22:08:29 +00005892<ol>
Jeffrey Yasskin95fa80a2010-01-09 19:44:16 +00005893 <li>The optional "tail" marker indicates that the callee function does not
5894 access any allocas or varargs in the caller. Note that calls may be
5895 marked "tail" even if they do not occur before
5896 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
5897 present, the function call is eligible for tail call optimization,
5898 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
Evan Chengdc444e92010-03-08 21:05:02 +00005899 optimized into a jump</a>. The code generator may optimize calls marked
5900 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
5901 sibling call optimization</a> when the caller and callee have
5902 matching signatures, or 2) forced tail call optimization when the
5903 following extra requirements are met:
Jeffrey Yasskin95fa80a2010-01-09 19:44:16 +00005904 <ul>
5905 <li>Caller and callee both have the calling
5906 convention <tt>fastcc</tt>.</li>
5907 <li>The call is in tail position (ret immediately follows call and ret
5908 uses value of call or is void).</li>
5909 <li>Option <tt>-tailcallopt</tt> is enabled,
Dan Gohmanfbbee8d2010-03-02 01:08:11 +00005910 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
Jeffrey Yasskin95fa80a2010-01-09 19:44:16 +00005911 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
5912 constraints are met.</a></li>
5913 </ul>
5914 </li>
Devang Patelf642f472008-10-06 18:50:38 +00005915
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005916 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5917 convention</a> the call should use. If none is specified, the call
Jeffrey Yasskin95fa80a2010-01-09 19:44:16 +00005918 defaults to using C calling conventions. The calling convention of the
5919 call must match the calling convention of the target function, or else the
5920 behavior is undefined.</li>
Devang Patelf642f472008-10-06 18:50:38 +00005921
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005922 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5923 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5924 '<tt>inreg</tt>' attributes are valid here.</li>
5925
5926 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5927 type of the return value. Functions that return no value are marked
5928 <tt><a href="#t_void">void</a></tt>.</li>
5929
5930 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5931 being invoked. The argument types must match the types implied by this
5932 signature. This type can be omitted if the function is not varargs and if
5933 the function type does not return a pointer to a function.</li>
5934
5935 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5936 be invoked. In most cases, this is a direct function invocation, but
5937 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5938 to function value.</li>
5939
5940 <li>'<tt>function args</tt>': argument list whose types match the function
Chris Lattner0724fbd2010-03-02 06:36:51 +00005941 signature argument types and parameter attributes. All arguments must be
5942 of <a href="#t_firstclass">first class</a> type. If the function
5943 signature indicates the function accepts a variable number of arguments,
5944 the extra arguments can be specified.</li>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005945
5946 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5947 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5948 '<tt>readnone</tt>' attributes are valid here.</li>
Chris Lattner6536cfe2002-05-06 22:08:29 +00005949</ol>
Chris Lattner2bff5242005-05-06 05:47:36 +00005950
Chris Lattner00950542001-06-06 20:29:01 +00005951<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005952<p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5953 a specified function, with its incoming arguments bound to the specified
5954 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5955 function, control flow continues with the instruction after the function
5956 call, and the return value of the function is bound to the result
5957 argument.</p>
Chris Lattner2bff5242005-05-06 05:47:36 +00005958
Chris Lattner00950542001-06-06 20:29:01 +00005959<h5>Example:</h5>
Chris Lattner2bff5242005-05-06 05:47:36 +00005960<pre>
Nick Lewyckydb7e3c92007-09-08 13:57:50 +00005961 %retval = call i32 @test(i32 %argc)
Dan Gohmanfe47aae2010-05-28 17:13:49 +00005962 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
Chris Lattner772fccf2008-03-21 17:24:17 +00005963 %X = tail call i32 @foo() <i>; yields i32</i>
5964 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5965 call void %foo(i8 97 signext)
Devang Patelc3fc6df2008-03-10 20:49:15 +00005966
5967 %struct.A = type { i32, i8 }
Devang Patelf642f472008-10-06 18:50:38 +00005968 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
Dan Gohmanb1e6b962008-10-04 19:00:07 +00005969 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5970 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
Chris Lattner85a350f2008-10-08 06:26:11 +00005971 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
Matthijs Kooijmancb73d192008-10-07 10:03:45 +00005972 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
Chris Lattner2bff5242005-05-06 05:47:36 +00005973</pre>
5974
Dale Johannesen07de8d12009-09-24 18:38:21 +00005975<p>llvm treats calls to some functions with names and arguments that match the
Dale Johannesen9f8380b2009-09-25 17:04:42 +00005976standard C99 library as being the C99 library functions, and may perform
5977optimizations or generate code for them under that assumption. This is
5978something we'd like to change in the future to provide better support for
Dan Gohmanb55a1ee2010-03-01 17:41:39 +00005979freestanding environments and non-C-based languages.</p>
Dale Johannesen07de8d12009-09-24 18:38:21 +00005980
Misha Brukman9d0919f2003-11-08 01:05:38 +00005981</div>
Chris Lattnere19d7a72004-09-27 21:51:25 +00005982
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00005983<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005984<h4>
Chris Lattnerfb6977d2006-01-13 23:26:01 +00005985 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00005986</h4>
Chris Lattnere19d7a72004-09-27 21:51:25 +00005987
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00005988<div>
Chris Lattnere19d7a72004-09-27 21:51:25 +00005989
Chris Lattner8d1a81d2003-10-18 05:51:36 +00005990<h5>Syntax:</h5>
Chris Lattnere19d7a72004-09-27 21:51:25 +00005991<pre>
Andrew Lenharth8bf607a2005-06-18 18:28:17 +00005992 &lt;resultval&gt; = va_arg &lt;va_list*&gt; &lt;arglist&gt;, &lt;argty&gt;
Chris Lattnere19d7a72004-09-27 21:51:25 +00005993</pre>
5994
Chris Lattner8d1a81d2003-10-18 05:51:36 +00005995<h5>Overview:</h5>
Andrew Lenharth8bf607a2005-06-18 18:28:17 +00005996<p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
Bill Wendlinge910b4c2009-07-20 02:29:24 +00005997 the "variable argument" area of a function call. It is used to implement the
5998 <tt>va_arg</tt> macro in C.</p>
Chris Lattnere19d7a72004-09-27 21:51:25 +00005999
Chris Lattner8d1a81d2003-10-18 05:51:36 +00006000<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006001<p>This instruction takes a <tt>va_list*</tt> value and the type of the
6002 argument. It returns a value of the specified argument type and increments
6003 the <tt>va_list</tt> to point to the next argument. The actual type
6004 of <tt>va_list</tt> is target specific.</p>
Chris Lattnere19d7a72004-09-27 21:51:25 +00006005
Chris Lattner8d1a81d2003-10-18 05:51:36 +00006006<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006007<p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6008 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6009 to the next argument. For more information, see the variable argument
6010 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
Chris Lattnere19d7a72004-09-27 21:51:25 +00006011
6012<p>It is legal for this instruction to be called in a function which does not
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006013 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6014 function.</p>
Chris Lattnere19d7a72004-09-27 21:51:25 +00006015
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006016<p><tt>va_arg</tt> is an LLVM instruction instead of
6017 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6018 argument.</p>
Chris Lattnere19d7a72004-09-27 21:51:25 +00006019
Chris Lattner8d1a81d2003-10-18 05:51:36 +00006020<h5>Example:</h5>
Chris Lattnere19d7a72004-09-27 21:51:25 +00006021<p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6022
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006023<p>Note that the code generator does not yet fully support va_arg on many
6024 targets. Also, it does not currently support va_arg with aggregate types on
6025 any target.</p>
Dan Gohmanf3e60bd2009-01-12 23:12:39 +00006026
Misha Brukman9d0919f2003-11-08 01:05:38 +00006027</div>
Chris Lattner8ff75902004-01-06 05:31:32 +00006028
Bill Wendlingf78faf82011-08-02 21:52:38 +00006029<!-- _______________________________________________________________________ -->
6030<h4>
6031 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6032</h4>
6033
6034<div>
6035
6036<h5>Syntax:</h5>
6037<pre>
Bill Wendlingbf13ee12011-08-08 08:06:05 +00006038 &lt;resultval&gt; = landingpad &lt;somety&gt; personality &lt;type&gt; &lt;pers_fn&gt; &lt;clause&gt;+
6039 &lt;resultval&gt; = landingpad &lt;somety&gt; personality &lt;type&gt; &lt;pers_fn&gt; cleanup &lt;clause&gt;*
6040
Bill Wendlingf78faf82011-08-02 21:52:38 +00006041 &lt;clause&gt; := catch &lt;type&gt; &lt;value&gt;
Bill Wendlinge6e88262011-08-12 20:24:12 +00006042 &lt;clause&gt; := filter &lt;array constant type&gt; &lt;array constant&gt;
Bill Wendlingf78faf82011-08-02 21:52:38 +00006043</pre>
6044
6045<h5>Overview:</h5>
6046<p>The '<tt>landingpad</tt>' instruction is used by
6047 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6048 system</a> to specify that a basic block is a landing pad &mdash; one where
6049 the exception lands, and corresponds to the code found in the
6050 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6051 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6052 re-entry to the function. The <tt>resultval</tt> has the
6053 type <tt>somety</tt>.</p>
6054
6055<h5>Arguments:</h5>
6056<p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6057 function associated with the unwinding mechanism. The optional
6058 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6059
6060<p>A <tt>clause</tt> begins with the clause type &mdash; <tt>catch</tt>
Bill Wendlinge6e88262011-08-12 20:24:12 +00006061 or <tt>filter</tt> &mdash; and contains the global variable representing the
6062 "type" that may be caught or filtered respectively. Unlike the
6063 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6064 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6065 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
Bill Wendlingf78faf82011-08-02 21:52:38 +00006066 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6067
6068<h5>Semantics:</h5>
6069<p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6070 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6071 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6072 calling conventions, how the personality function results are represented in
6073 LLVM IR is target specific.</p>
6074
Bill Wendlingb7a01352011-08-03 17:17:06 +00006075<p>The clauses are applied in order from top to bottom. If two
6076 <tt>landingpad</tt> instructions are merged together through inlining, the
Bill Wendling2905c322011-08-08 07:58:58 +00006077 clauses from the calling function are appended to the list of clauses.</p>
Bill Wendlingb7a01352011-08-03 17:17:06 +00006078
Bill Wendlingf78faf82011-08-02 21:52:38 +00006079<p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6080
6081<ul>
6082 <li>A landing pad block is a basic block which is the unwind destination of an
6083 '<tt>invoke</tt>' instruction.</li>
6084 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6085 first non-PHI instruction.</li>
6086 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6087 pad block.</li>
6088 <li>A basic block that is not a landing pad block may not include a
6089 '<tt>landingpad</tt>' instruction.</li>
6090 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6091 personality function.</li>
6092</ul>
6093
6094<h5>Example:</h5>
6095<pre>
6096 ;; A landing pad which can catch an integer.
6097 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6098 catch i8** @_ZTIi
6099 ;; A landing pad that is a cleanup.
6100 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
Bill Wendlinge6e88262011-08-12 20:24:12 +00006101 cleanup
Bill Wendlingf78faf82011-08-02 21:52:38 +00006102 ;; A landing pad which can catch an integer and can only throw a double.
6103 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6104 catch i8** @_ZTIi
Bill Wendlinge6e88262011-08-12 20:24:12 +00006105 filter [1 x i8**] [@_ZTId]
Bill Wendlingf78faf82011-08-02 21:52:38 +00006106</pre>
6107
6108</div>
6109
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006110</div>
6111
6112</div>
6113
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006114<!-- *********************************************************************** -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006115<h2><a name="intrinsics">Intrinsic Functions</a></h2>
Chris Lattner261efe92003-11-25 01:02:51 +00006116<!-- *********************************************************************** -->
Chris Lattner8ff75902004-01-06 05:31:32 +00006117
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006118<div>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006119
6120<p>LLVM supports the notion of an "intrinsic function". These functions have
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006121 well known names and semantics and are required to follow certain
6122 restrictions. Overall, these intrinsics represent an extension mechanism for
6123 the LLVM language that does not require changing all of the transformations
6124 in LLVM when adding to the language (or the bitcode reader/writer, the
6125 parser, etc...).</p>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006126
John Criswellfc6b8952005-05-16 16:17:45 +00006127<p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006128 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6129 begin with this prefix. Intrinsic functions must always be external
6130 functions: you cannot define the body of intrinsic functions. Intrinsic
6131 functions may only be used in call or invoke instructions: it is illegal to
6132 take the address of an intrinsic function. Additionally, because intrinsic
6133 functions are part of the LLVM language, it is required if any are added that
6134 they be documented here.</p>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006135
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006136<p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6137 family of functions that perform the same operation but on different data
6138 types. Because LLVM can represent over 8 million different integer types,
6139 overloading is used commonly to allow an intrinsic function to operate on any
6140 integer type. One or more of the argument types or the result type can be
6141 overloaded to accept any integer type. Argument types may also be defined as
6142 exactly matching a previous argument's type or the result type. This allows
6143 an intrinsic function which accepts multiple arguments, but needs all of them
6144 to be of the same type, to only be overloaded with respect to a single
6145 argument or the result.</p>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006146
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006147<p>Overloaded intrinsics will have the names of its overloaded argument types
6148 encoded into its function name, each preceded by a period. Only those types
6149 which are overloaded result in a name suffix. Arguments whose type is matched
6150 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6151 can take an integer of any width and returns an integer of exactly the same
6152 integer width. This leads to a family of functions such as
6153 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6154 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6155 suffix is required. Because the argument's type is matched against the return
6156 type, it does not require its own name suffix.</p>
Reid Spencer409e28f2007-04-01 08:04:23 +00006157
Eric Christopher6c7e8a02009-12-05 02:46:03 +00006158<p>To learn how to add an intrinsic function, please see the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006159 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006160
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006161<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006162<h3>
Chris Lattner8ff75902004-01-06 05:31:32 +00006163 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006164</h3>
Chris Lattner8ff75902004-01-06 05:31:32 +00006165
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006166<div>
Chris Lattnerd7923912004-05-23 21:06:01 +00006167
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006168<p>Variable argument support is defined in LLVM with
6169 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6170 intrinsic functions. These functions are related to the similarly named
6171 macros defined in the <tt>&lt;stdarg.h&gt;</tt> header file.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006172
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006173<p>All of these functions operate on arguments that use a target-specific value
6174 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6175 not define what this type is, so all transformations should be prepared to
6176 handle these functions regardless of the type used.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006177
Chris Lattner374ab302006-05-15 17:26:46 +00006178<p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006179 instruction and the variable argument handling intrinsic functions are
6180 used.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006181
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00006182<pre class="doc_code">
Anton Korobeynikov5d522f32007-03-21 23:58:04 +00006183define i32 @test(i32 %X, ...) {
Chris Lattner33aec9e2004-02-12 17:01:32 +00006184 ; Initialize variable argument processing
Jeff Cohenb627eab2007-04-29 01:07:00 +00006185 %ap = alloca i8*
Chris Lattnerb75137d2007-01-08 07:55:15 +00006186 %ap2 = bitcast i8** %ap to i8*
Anton Korobeynikov5d522f32007-03-21 23:58:04 +00006187 call void @llvm.va_start(i8* %ap2)
Chris Lattner33aec9e2004-02-12 17:01:32 +00006188
6189 ; Read a single integer argument
Jeff Cohenb627eab2007-04-29 01:07:00 +00006190 %tmp = va_arg i8** %ap, i32
Chris Lattner33aec9e2004-02-12 17:01:32 +00006191
6192 ; Demonstrate usage of llvm.va_copy and llvm.va_end
Jeff Cohenb627eab2007-04-29 01:07:00 +00006193 %aq = alloca i8*
Chris Lattnerb75137d2007-01-08 07:55:15 +00006194 %aq2 = bitcast i8** %aq to i8*
Jeff Cohenb627eab2007-04-29 01:07:00 +00006195 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
Anton Korobeynikov5d522f32007-03-21 23:58:04 +00006196 call void @llvm.va_end(i8* %aq2)
Chris Lattner33aec9e2004-02-12 17:01:32 +00006197
6198 ; Stop processing of arguments.
Anton Korobeynikov5d522f32007-03-21 23:58:04 +00006199 call void @llvm.va_end(i8* %ap2)
Reid Spencerca86e162006-12-31 07:07:53 +00006200 ret i32 %tmp
Chris Lattner33aec9e2004-02-12 17:01:32 +00006201}
Anton Korobeynikov5d522f32007-03-21 23:58:04 +00006202
6203declare void @llvm.va_start(i8*)
6204declare void @llvm.va_copy(i8*, i8*)
6205declare void @llvm.va_end(i8*)
Chris Lattner33aec9e2004-02-12 17:01:32 +00006206</pre>
Chris Lattner8ff75902004-01-06 05:31:32 +00006207
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006208<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006209<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006210 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006211</h4>
Chris Lattner8ff75902004-01-06 05:31:32 +00006212
6213
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006214<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006215
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006216<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006217<pre>
6218 declare void %llvm.va_start(i8* &lt;arglist&gt;)
6219</pre>
6220
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006221<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006222<p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*&lt;arglist&gt;</tt>
6223 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
Andrew Lenharth8bf607a2005-06-18 18:28:17 +00006224
6225<h5>Arguments:</h5>
Dan Gohman0e451ce2008-10-14 16:51:45 +00006226<p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
Andrew Lenharth8bf607a2005-06-18 18:28:17 +00006227
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006228<h5>Semantics:</h5>
Dan Gohman0e451ce2008-10-14 16:51:45 +00006229<p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006230 macro available in C. In a target-dependent way, it initializes
6231 the <tt>va_list</tt> element to which the argument points, so that the next
6232 call to <tt>va_arg</tt> will produce the first variable argument passed to
6233 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6234 need to know the last argument of the function as the compiler can figure
6235 that out.</p>
Andrew Lenharth8bf607a2005-06-18 18:28:17 +00006236
Misha Brukman9d0919f2003-11-08 01:05:38 +00006237</div>
Chris Lattner8ff75902004-01-06 05:31:32 +00006238
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006239<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006240<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006241 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006242</h4>
Chris Lattner8ff75902004-01-06 05:31:32 +00006243
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006244<div>
Chris Lattnerb75137d2007-01-08 07:55:15 +00006245
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006246<h5>Syntax:</h5>
6247<pre>
6248 declare void @llvm.va_end(i8* &lt;arglist&gt;)
6249</pre>
6250
6251<h5>Overview:</h5>
Jeff Cohenb627eab2007-04-29 01:07:00 +00006252<p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*&lt;arglist&gt;</tt>,
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006253 which has been initialized previously
6254 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6255 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
Chris Lattnerb75137d2007-01-08 07:55:15 +00006256
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006257<h5>Arguments:</h5>
Jeff Cohenb627eab2007-04-29 01:07:00 +00006258<p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
Chris Lattnerb75137d2007-01-08 07:55:15 +00006259
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006260<h5>Semantics:</h5>
Misha Brukman9d0919f2003-11-08 01:05:38 +00006261<p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006262 macro available in C. In a target-dependent way, it destroys
6263 the <tt>va_list</tt> element to which the argument points. Calls
6264 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6265 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6266 with calls to <tt>llvm.va_end</tt>.</p>
Chris Lattnerb75137d2007-01-08 07:55:15 +00006267
Misha Brukman9d0919f2003-11-08 01:05:38 +00006268</div>
Chris Lattner8ff75902004-01-06 05:31:32 +00006269
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006270<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006271<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006272 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006273</h4>
Chris Lattner8ff75902004-01-06 05:31:32 +00006274
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006275<div>
Chris Lattnerd7923912004-05-23 21:06:01 +00006276
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006277<h5>Syntax:</h5>
Chris Lattnerd7923912004-05-23 21:06:01 +00006278<pre>
Anton Korobeynikovec43a062007-03-22 00:02:17 +00006279 declare void @llvm.va_copy(i8* &lt;destarglist&gt;, i8* &lt;srcarglist&gt;)
Chris Lattnerd7923912004-05-23 21:06:01 +00006280</pre>
6281
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006282<h5>Overview:</h5>
Jeff Cohenb627eab2007-04-29 01:07:00 +00006283<p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006284 from the source argument list to the destination argument list.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006285
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006286<h5>Arguments:</h5>
Andrew Lenharth8bf607a2005-06-18 18:28:17 +00006287<p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006288 The second argument is a pointer to a <tt>va_list</tt> element to copy
6289 from.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006290
Chris Lattnerd9ad5b32003-05-08 04:57:36 +00006291<h5>Semantics:</h5>
Jeff Cohenb627eab2007-04-29 01:07:00 +00006292<p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006293 macro available in C. In a target-dependent way, it copies the
6294 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6295 element. This intrinsic is necessary because
6296 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6297 arbitrarily complex and require, for example, memory allocation.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006298
Misha Brukman9d0919f2003-11-08 01:05:38 +00006299</div>
Chris Lattner8ff75902004-01-06 05:31:32 +00006300
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006301</div>
6302
Bill Wendling0246bb72011-07-31 06:45:03 +00006303</div>
6304
Chris Lattner33aec9e2004-02-12 17:01:32 +00006305<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006306<h3>
Chris Lattnerd7923912004-05-23 21:06:01 +00006307 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006308</h3>
Chris Lattnerd7923912004-05-23 21:06:01 +00006309
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006310<div>
Chris Lattnerd7923912004-05-23 21:06:01 +00006311
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006312<p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
Chris Lattnerd3eda892008-08-05 18:29:16 +00006313Collection</a> (GC) requires the implementation and generation of these
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006314intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6315roots on the stack</a>, as well as garbage collector implementations that
6316require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6317barriers. Front-ends for type-safe garbage collected languages should generate
6318these intrinsics to make use of the LLVM garbage collectors. For more details,
6319see <a href="GarbageCollection.html">Accurate Garbage Collection with
6320LLVM</a>.</p>
Christopher Lamb303dae92007-12-17 01:00:21 +00006321
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006322<p>The garbage collection intrinsics only operate on objects in the generic
6323 address space (address space zero).</p>
Christopher Lamb303dae92007-12-17 01:00:21 +00006324
Chris Lattnerd7923912004-05-23 21:06:01 +00006325<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006326<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006327 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006328</h4>
Chris Lattnerd7923912004-05-23 21:06:01 +00006329
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006330<div>
Chris Lattnerd7923912004-05-23 21:06:01 +00006331
6332<h5>Syntax:</h5>
Chris Lattnerd7923912004-05-23 21:06:01 +00006333<pre>
Chris Lattner1df4f752007-09-21 17:30:40 +00006334 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
Chris Lattnerd7923912004-05-23 21:06:01 +00006335</pre>
6336
6337<h5>Overview:</h5>
John Criswell9e2485c2004-12-10 15:51:16 +00006338<p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006339 the code generator, and allows some metadata to be associated with it.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006340
6341<h5>Arguments:</h5>
Chris Lattnerd7923912004-05-23 21:06:01 +00006342<p>The first argument specifies the address of a stack object that contains the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006343 root pointer. The second pointer (which must be either a constant or a
6344 global value address) contains the meta-data to be associated with the
6345 root.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006346
6347<h5>Semantics:</h5>
Chris Lattner05d67092008-04-24 05:59:56 +00006348<p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006349 location. At compile-time, the code generator generates information to allow
6350 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6351 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6352 algorithm</a>.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006353
6354</div>
6355
Chris Lattnerd7923912004-05-23 21:06:01 +00006356<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006357<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006358 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006359</h4>
Chris Lattnerd7923912004-05-23 21:06:01 +00006360
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006361<div>
Chris Lattnerd7923912004-05-23 21:06:01 +00006362
6363<h5>Syntax:</h5>
Chris Lattnerd7923912004-05-23 21:06:01 +00006364<pre>
Chris Lattner1df4f752007-09-21 17:30:40 +00006365 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
Chris Lattnerd7923912004-05-23 21:06:01 +00006366</pre>
6367
6368<h5>Overview:</h5>
Chris Lattnerd7923912004-05-23 21:06:01 +00006369<p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006370 locations, allowing garbage collector implementations that require read
6371 barriers.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006372
6373<h5>Arguments:</h5>
Chris Lattner80626e92006-03-14 20:02:51 +00006374<p>The second argument is the address to read from, which should be an address
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006375 allocated from the garbage collector. The first object is a pointer to the
6376 start of the referenced object, if needed by the language runtime (otherwise
6377 null).</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006378
6379<h5>Semantics:</h5>
Chris Lattnerd7923912004-05-23 21:06:01 +00006380<p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006381 instruction, but may be replaced with substantially more complex code by the
6382 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6383 may only be used in a function which <a href="#gc">specifies a GC
6384 algorithm</a>.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006385
6386</div>
6387
Chris Lattnerd7923912004-05-23 21:06:01 +00006388<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006389<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006390 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006391</h4>
Chris Lattnerd7923912004-05-23 21:06:01 +00006392
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006393<div>
Chris Lattnerd7923912004-05-23 21:06:01 +00006394
6395<h5>Syntax:</h5>
Chris Lattnerd7923912004-05-23 21:06:01 +00006396<pre>
Chris Lattner1df4f752007-09-21 17:30:40 +00006397 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
Chris Lattnerd7923912004-05-23 21:06:01 +00006398</pre>
6399
6400<h5>Overview:</h5>
Chris Lattnerd7923912004-05-23 21:06:01 +00006401<p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006402 locations, allowing garbage collector implementations that require write
6403 barriers (such as generational or reference counting collectors).</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006404
6405<h5>Arguments:</h5>
Chris Lattner80626e92006-03-14 20:02:51 +00006406<p>The first argument is the reference to store, the second is the start of the
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006407 object to store it to, and the third is the address of the field of Obj to
6408 store to. If the runtime does not require a pointer to the object, Obj may
6409 be null.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006410
6411<h5>Semantics:</h5>
Chris Lattnerd7923912004-05-23 21:06:01 +00006412<p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006413 instruction, but may be replaced with substantially more complex code by the
6414 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6415 may only be used in a function which <a href="#gc">specifies a GC
6416 algorithm</a>.</p>
Chris Lattnerd7923912004-05-23 21:06:01 +00006417
6418</div>
6419
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006420</div>
6421
Chris Lattnerd7923912004-05-23 21:06:01 +00006422<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006423<h3>
Chris Lattner10610642004-02-14 04:08:35 +00006424 <a name="int_codegen">Code Generator Intrinsics</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006425</h3>
Chris Lattner10610642004-02-14 04:08:35 +00006426
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006427<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006428
6429<p>These intrinsics are provided by LLVM to expose special features that may
6430 only be implemented with code generator support.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006431
Chris Lattner10610642004-02-14 04:08:35 +00006432<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006433<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006434 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006435</h4>
Chris Lattner10610642004-02-14 04:08:35 +00006436
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006437<div>
Chris Lattner10610642004-02-14 04:08:35 +00006438
6439<h5>Syntax:</h5>
6440<pre>
Anton Korobeynikovec43a062007-03-22 00:02:17 +00006441 declare i8 *@llvm.returnaddress(i32 &lt;level&gt;)
Chris Lattner10610642004-02-14 04:08:35 +00006442</pre>
6443
6444<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006445<p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6446 target-specific value indicating the return address of the current function
6447 or one of its callers.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006448
6449<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006450<p>The argument to this intrinsic indicates which function to return the address
6451 for. Zero indicates the calling function, one indicates its caller, etc.
6452 The argument is <b>required</b> to be a constant integer value.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006453
6454<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006455<p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6456 indicating the return address of the specified call frame, or zero if it
6457 cannot be identified. The value returned by this intrinsic is likely to be
6458 incorrect or 0 for arguments other than zero, so it should only be used for
6459 debugging purposes.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006460
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006461<p>Note that calling this intrinsic does not prevent function inlining or other
6462 aggressive transformations, so the value returned may not be that of the
6463 obvious source-language caller.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006464
Chris Lattner10610642004-02-14 04:08:35 +00006465</div>
6466
Chris Lattner10610642004-02-14 04:08:35 +00006467<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006468<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006469 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006470</h4>
Chris Lattner10610642004-02-14 04:08:35 +00006471
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006472<div>
Chris Lattner10610642004-02-14 04:08:35 +00006473
6474<h5>Syntax:</h5>
6475<pre>
Dan Gohmanfe47aae2010-05-28 17:13:49 +00006476 declare i8* @llvm.frameaddress(i32 &lt;level&gt;)
Chris Lattner10610642004-02-14 04:08:35 +00006477</pre>
6478
6479<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006480<p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6481 target-specific frame pointer value for the specified stack frame.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006482
6483<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006484<p>The argument to this intrinsic indicates which function to return the frame
6485 pointer for. Zero indicates the calling function, one indicates its caller,
6486 etc. The argument is <b>required</b> to be a constant integer value.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006487
6488<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006489<p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6490 indicating the frame address of the specified call frame, or zero if it
6491 cannot be identified. The value returned by this intrinsic is likely to be
6492 incorrect or 0 for arguments other than zero, so it should only be used for
6493 debugging purposes.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006494
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006495<p>Note that calling this intrinsic does not prevent function inlining or other
6496 aggressive transformations, so the value returned may not be that of the
6497 obvious source-language caller.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006498
Chris Lattner10610642004-02-14 04:08:35 +00006499</div>
6500
Chris Lattner9a9d7ac2005-02-28 19:24:19 +00006501<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006502<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006503 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006504</h4>
Chris Lattner57e1f392006-01-13 02:03:13 +00006505
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006506<div>
Chris Lattner57e1f392006-01-13 02:03:13 +00006507
6508<h5>Syntax:</h5>
6509<pre>
Dan Gohmanfe47aae2010-05-28 17:13:49 +00006510 declare i8* @llvm.stacksave()
Chris Lattner57e1f392006-01-13 02:03:13 +00006511</pre>
6512
6513<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006514<p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6515 of the function stack, for use
6516 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6517 useful for implementing language features like scoped automatic variable
6518 sized arrays in C99.</p>
Chris Lattner57e1f392006-01-13 02:03:13 +00006519
6520<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006521<p>This intrinsic returns a opaque pointer value that can be passed
6522 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6523 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6524 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6525 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6526 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6527 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
Chris Lattner57e1f392006-01-13 02:03:13 +00006528
6529</div>
6530
6531<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006532<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006533 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006534</h4>
Chris Lattner57e1f392006-01-13 02:03:13 +00006535
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006536<div>
Chris Lattner57e1f392006-01-13 02:03:13 +00006537
6538<h5>Syntax:</h5>
6539<pre>
Dan Gohmanfe47aae2010-05-28 17:13:49 +00006540 declare void @llvm.stackrestore(i8* %ptr)
Chris Lattner57e1f392006-01-13 02:03:13 +00006541</pre>
6542
6543<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006544<p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6545 the function stack to the state it was in when the
6546 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6547 executed. This is useful for implementing language features like scoped
6548 automatic variable sized arrays in C99.</p>
Chris Lattner57e1f392006-01-13 02:03:13 +00006549
6550<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006551<p>See the description
6552 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
Chris Lattner57e1f392006-01-13 02:03:13 +00006553
6554</div>
6555
Chris Lattner57e1f392006-01-13 02:03:13 +00006556<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006557<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006558 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006559</h4>
Chris Lattner9a9d7ac2005-02-28 19:24:19 +00006560
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006561<div>
Chris Lattner9a9d7ac2005-02-28 19:24:19 +00006562
6563<h5>Syntax:</h5>
6564<pre>
Bruno Cardoso Lopes9a767332011-06-14 04:58:37 +00006565 declare void @llvm.prefetch(i8* &lt;address&gt;, i32 &lt;rw&gt;, i32 &lt;locality&gt;, i32 &lt;cache type&gt;)
Chris Lattner9a9d7ac2005-02-28 19:24:19 +00006566</pre>
6567
6568<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006569<p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6570 insert a prefetch instruction if supported; otherwise, it is a noop.
6571 Prefetches have no effect on the behavior of the program but can change its
6572 performance characteristics.</p>
Chris Lattner9a9d7ac2005-02-28 19:24:19 +00006573
6574<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006575<p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6576 specifier determining if the fetch should be for a read (0) or write (1),
6577 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
Bruno Cardoso Lopes9a767332011-06-14 04:58:37 +00006578 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6579 specifies whether the prefetch is performed on the data (1) or instruction (0)
6580 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6581 must be constant integers.</p>
Chris Lattner9a9d7ac2005-02-28 19:24:19 +00006582
6583<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006584<p>This intrinsic does not modify the behavior of the program. In particular,
6585 prefetches cannot trap and do not produce a value. On targets that support
6586 this intrinsic, the prefetch can provide hints to the processor cache for
6587 better performance.</p>
Chris Lattner9a9d7ac2005-02-28 19:24:19 +00006588
6589</div>
6590
Andrew Lenharth7f4ec3b2005-03-28 20:05:49 +00006591<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006592<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006593 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006594</h4>
Andrew Lenharth7f4ec3b2005-03-28 20:05:49 +00006595
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006596<div>
Andrew Lenharth7f4ec3b2005-03-28 20:05:49 +00006597
6598<h5>Syntax:</h5>
6599<pre>
Chris Lattner1df4f752007-09-21 17:30:40 +00006600 declare void @llvm.pcmarker(i32 &lt;id&gt;)
Andrew Lenharth7f4ec3b2005-03-28 20:05:49 +00006601</pre>
6602
6603<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006604<p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
6605 Counter (PC) in a region of code to simulators and other tools. The method
6606 is target specific, but it is expected that the marker will use exported
6607 symbols to transmit the PC of the marker. The marker makes no guarantees
6608 that it will remain with any specific instruction after optimizations. It is
6609 possible that the presence of a marker will inhibit optimizations. The
6610 intended use is to be inserted after optimizations to allow correlations of
6611 simulation runs.</p>
Andrew Lenharth7f4ec3b2005-03-28 20:05:49 +00006612
6613<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006614<p><tt>id</tt> is a numerical id identifying the marker.</p>
Andrew Lenharth7f4ec3b2005-03-28 20:05:49 +00006615
6616<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006617<p>This intrinsic does not modify the behavior of the program. Backends that do
Dan Gohmanb55a1ee2010-03-01 17:41:39 +00006618 not support this intrinsic may ignore it.</p>
Andrew Lenharth7f4ec3b2005-03-28 20:05:49 +00006619
6620</div>
6621
Andrew Lenharth51b8d542005-11-11 16:47:30 +00006622<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006623<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006624 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006625</h4>
Andrew Lenharth51b8d542005-11-11 16:47:30 +00006626
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006627<div>
Andrew Lenharth51b8d542005-11-11 16:47:30 +00006628
6629<h5>Syntax:</h5>
6630<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00006631 declare i64 @llvm.readcyclecounter()
Andrew Lenharth51b8d542005-11-11 16:47:30 +00006632</pre>
6633
6634<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006635<p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
6636 counter register (or similar low latency, high accuracy clocks) on those
6637 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6638 should map to RPCC. As the backing counters overflow quickly (on the order
6639 of 9 seconds on alpha), this should only be used for small timings.</p>
Andrew Lenharth51b8d542005-11-11 16:47:30 +00006640
6641<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006642<p>When directly supported, reading the cycle counter should not modify any
6643 memory. Implementations are allowed to either return a application specific
6644 value or a system wide value. On backends without support, this is lowered
6645 to a constant 0.</p>
Andrew Lenharth51b8d542005-11-11 16:47:30 +00006646
6647</div>
6648
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006649</div>
6650
Chris Lattner10610642004-02-14 04:08:35 +00006651<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006652<h3>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006653 <a name="int_libc">Standard C Library Intrinsics</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006654</h3>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006655
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006656<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006657
6658<p>LLVM provides intrinsics for a few important standard C library functions.
6659 These intrinsics allow source-language front-ends to pass information about
6660 the alignment of the pointer arguments to the code generator, providing
6661 opportunity for more efficient code generation.</p>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006662
Chris Lattner33aec9e2004-02-12 17:01:32 +00006663<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006664<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006665 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006666</h4>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006667
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006668<div>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006669
6670<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006671<p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
Mon P Wange88909b2010-04-07 06:35:53 +00006672 integer bit width and for different address spaces. Not all targets support
6673 all bit widths however.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006674
Chris Lattner33aec9e2004-02-12 17:01:32 +00006675<pre>
Dan Gohmanfe47aae2010-05-28 17:13:49 +00006676 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* &lt;dest&gt;, i8* &lt;src&gt;,
Chris Lattner9f636de2010-04-08 00:53:57 +00006677 i32 &lt;len&gt;, i32 &lt;align&gt;, i1 &lt;isvolatile&gt;)
Dan Gohmanfe47aae2010-05-28 17:13:49 +00006678 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* &lt;dest&gt;, i8* &lt;src&gt;,
Chris Lattner9f636de2010-04-08 00:53:57 +00006679 i64 &lt;len&gt;, i32 &lt;align&gt;, i1 &lt;isvolatile&gt;)
Chris Lattner33aec9e2004-02-12 17:01:32 +00006680</pre>
6681
6682<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006683<p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6684 source location to the destination location.</p>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006685
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006686<p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
Chris Lattner9f636de2010-04-08 00:53:57 +00006687 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6688 and the pointers can be in specified address spaces.</p>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006689
6690<h5>Arguments:</h5>
Chris Lattner9f636de2010-04-08 00:53:57 +00006691
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006692<p>The first argument is a pointer to the destination, the second is a pointer
6693 to the source. The third argument is an integer argument specifying the
Chris Lattner9f636de2010-04-08 00:53:57 +00006694 number of bytes to copy, the fourth argument is the alignment of the
6695 source and destination locations, and the fifth is a boolean indicating a
6696 volatile access.</p>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006697
Dan Gohmanb55a1ee2010-03-01 17:41:39 +00006698<p>If the call to this intrinsic has an alignment value that is not 0 or 1,
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006699 then the caller guarantees that both the source and destination pointers are
6700 aligned to that boundary.</p>
Chris Lattner3301ced2004-02-12 21:18:15 +00006701
Jeffrey Yasskin93e066d2010-04-26 21:21:24 +00006702<p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6703 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
6704 The detailed access behavior is not very cleanly specified and it is unwise
6705 to depend on it.</p>
Chris Lattner9f636de2010-04-08 00:53:57 +00006706
Chris Lattner33aec9e2004-02-12 17:01:32 +00006707<h5>Semantics:</h5>
Chris Lattner9f636de2010-04-08 00:53:57 +00006708
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006709<p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6710 source location to the destination location, which are not allowed to
6711 overlap. It copies "len" bytes of memory over. If the argument is known to
6712 be aligned to some boundary, this can be specified as the fourth argument,
6713 otherwise it should be set to 0 or 1.</p>
Chris Lattner33aec9e2004-02-12 17:01:32 +00006714
Chris Lattner33aec9e2004-02-12 17:01:32 +00006715</div>
6716
Chris Lattner0eb51b42004-02-12 18:10:10 +00006717<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006718<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006719 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006720</h4>
Chris Lattner0eb51b42004-02-12 18:10:10 +00006721
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006722<div>
Chris Lattner0eb51b42004-02-12 18:10:10 +00006723
6724<h5>Syntax:</h5>
Chris Lattner824b9582008-11-21 16:42:48 +00006725<p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
Mon P Wange88909b2010-04-07 06:35:53 +00006726 width and for different address space. Not all targets support all bit
6727 widths however.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006728
Chris Lattner0eb51b42004-02-12 18:10:10 +00006729<pre>
Dan Gohmanfe47aae2010-05-28 17:13:49 +00006730 declare void @llvm.memmove.p0i8.p0i8.i32(i8* &lt;dest&gt;, i8* &lt;src&gt;,
Chris Lattner9f636de2010-04-08 00:53:57 +00006731 i32 &lt;len&gt;, i32 &lt;align&gt;, i1 &lt;isvolatile&gt;)
Dan Gohmanfe47aae2010-05-28 17:13:49 +00006732 declare void @llvm.memmove.p0i8.p0i8.i64(i8* &lt;dest&gt;, i8* &lt;src&gt;,
Chris Lattner9f636de2010-04-08 00:53:57 +00006733 i64 &lt;len&gt;, i32 &lt;align&gt;, i1 &lt;isvolatile&gt;)
Chris Lattner0eb51b42004-02-12 18:10:10 +00006734</pre>
6735
6736<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006737<p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
6738 source location to the destination location. It is similar to the
6739 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
6740 overlap.</p>
Chris Lattner0eb51b42004-02-12 18:10:10 +00006741
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006742<p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
Chris Lattner9f636de2010-04-08 00:53:57 +00006743 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6744 and the pointers can be in specified address spaces.</p>
Chris Lattner0eb51b42004-02-12 18:10:10 +00006745
6746<h5>Arguments:</h5>
Chris Lattner9f636de2010-04-08 00:53:57 +00006747
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006748<p>The first argument is a pointer to the destination, the second is a pointer
6749 to the source. The third argument is an integer argument specifying the
Chris Lattner9f636de2010-04-08 00:53:57 +00006750 number of bytes to copy, the fourth argument is the alignment of the
6751 source and destination locations, and the fifth is a boolean indicating a
6752 volatile access.</p>
Chris Lattner0eb51b42004-02-12 18:10:10 +00006753
Dan Gohmanb55a1ee2010-03-01 17:41:39 +00006754<p>If the call to this intrinsic has an alignment value that is not 0 or 1,
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006755 then the caller guarantees that the source and destination pointers are
6756 aligned to that boundary.</p>
Chris Lattner3301ced2004-02-12 21:18:15 +00006757
Jeffrey Yasskin93e066d2010-04-26 21:21:24 +00006758<p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6759 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
6760 The detailed access behavior is not very cleanly specified and it is unwise
6761 to depend on it.</p>
Chris Lattner9f636de2010-04-08 00:53:57 +00006762
Chris Lattner0eb51b42004-02-12 18:10:10 +00006763<h5>Semantics:</h5>
Chris Lattner9f636de2010-04-08 00:53:57 +00006764
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006765<p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
6766 source location to the destination location, which may overlap. It copies
6767 "len" bytes of memory over. If the argument is known to be aligned to some
6768 boundary, this can be specified as the fourth argument, otherwise it should
6769 be set to 0 or 1.</p>
Chris Lattner0eb51b42004-02-12 18:10:10 +00006770
Chris Lattner0eb51b42004-02-12 18:10:10 +00006771</div>
6772
Chris Lattner10610642004-02-14 04:08:35 +00006773<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006774<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006775 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006776</h4>
Chris Lattner10610642004-02-14 04:08:35 +00006777
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006778<div>
Chris Lattner10610642004-02-14 04:08:35 +00006779
6780<h5>Syntax:</h5>
Chris Lattner824b9582008-11-21 16:42:48 +00006781<p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
John Criswellcdcbbfc2010-07-30 16:30:28 +00006782 width and for different address spaces. However, not all targets support all
6783 bit widths.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006784
Chris Lattner10610642004-02-14 04:08:35 +00006785<pre>
Dan Gohmanfe47aae2010-05-28 17:13:49 +00006786 declare void @llvm.memset.p0i8.i32(i8* &lt;dest&gt;, i8 &lt;val&gt;,
Chris Lattnerff35c3f2010-04-08 00:54:34 +00006787 i32 &lt;len&gt;, i32 &lt;align&gt;, i1 &lt;isvolatile&gt;)
Dan Gohmanfe47aae2010-05-28 17:13:49 +00006788 declare void @llvm.memset.p0i8.i64(i8* &lt;dest&gt;, i8 &lt;val&gt;,
Chris Lattnerff35c3f2010-04-08 00:54:34 +00006789 i64 &lt;len&gt;, i32 &lt;align&gt;, i1 &lt;isvolatile&gt;)
Chris Lattner10610642004-02-14 04:08:35 +00006790</pre>
6791
6792<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006793<p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
6794 particular byte value.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006795
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006796<p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
John Criswellcdcbbfc2010-07-30 16:30:28 +00006797 intrinsic does not return a value and takes extra alignment/volatile
6798 arguments. Also, the destination can be in an arbitrary address space.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006799
6800<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006801<p>The first argument is a pointer to the destination to fill, the second is the
John Criswellcdcbbfc2010-07-30 16:30:28 +00006802 byte value with which to fill it, the third argument is an integer argument
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006803 specifying the number of bytes to fill, and the fourth argument is the known
John Criswellcdcbbfc2010-07-30 16:30:28 +00006804 alignment of the destination location.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006805
Dan Gohmanb55a1ee2010-03-01 17:41:39 +00006806<p>If the call to this intrinsic has an alignment value that is not 0 or 1,
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006807 then the caller guarantees that the destination pointer is aligned to that
6808 boundary.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006809
Jeffrey Yasskin93e066d2010-04-26 21:21:24 +00006810<p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6811 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
6812 The detailed access behavior is not very cleanly specified and it is unwise
6813 to depend on it.</p>
Chris Lattner9f636de2010-04-08 00:53:57 +00006814
Chris Lattner10610642004-02-14 04:08:35 +00006815<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006816<p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
6817 at the destination location. If the argument is known to be aligned to some
6818 boundary, this can be specified as the fourth argument, otherwise it should
6819 be set to 0 or 1.</p>
Chris Lattner10610642004-02-14 04:08:35 +00006820
Chris Lattner10610642004-02-14 04:08:35 +00006821</div>
6822
Chris Lattner32006282004-06-11 02:28:03 +00006823<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006824<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006825 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006826</h4>
Chris Lattnera4d74142005-07-21 01:29:16 +00006827
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006828<div>
Chris Lattnera4d74142005-07-21 01:29:16 +00006829
6830<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006831<p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
6832 floating point or vector of floating point type. Not all targets support all
6833 types however.</p>
6834
Chris Lattnera4d74142005-07-21 01:29:16 +00006835<pre>
Dale Johannesen408f9c12007-10-02 17:47:38 +00006836 declare float @llvm.sqrt.f32(float %Val)
6837 declare double @llvm.sqrt.f64(double %Val)
6838 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6839 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6840 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
Chris Lattnera4d74142005-07-21 01:29:16 +00006841</pre>
6842
6843<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006844<p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
6845 returning the same value as the libm '<tt>sqrt</tt>' functions would.
6846 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
6847 behavior for negative numbers other than -0.0 (which allows for better
6848 optimization, because there is no need to worry about errno being
6849 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
Chris Lattnera4d74142005-07-21 01:29:16 +00006850
6851<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006852<p>The argument and return value are floating point numbers of the same
6853 type.</p>
Chris Lattnera4d74142005-07-21 01:29:16 +00006854
6855<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006856<p>This function returns the sqrt of the specified operand if it is a
6857 nonnegative floating point number.</p>
Chris Lattnera4d74142005-07-21 01:29:16 +00006858
Chris Lattnera4d74142005-07-21 01:29:16 +00006859</div>
6860
Chris Lattnerf4d252d2006-09-08 06:34:02 +00006861<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006862<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00006863 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006864</h4>
Chris Lattnerf4d252d2006-09-08 06:34:02 +00006865
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006866<div>
Chris Lattnerf4d252d2006-09-08 06:34:02 +00006867
6868<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006869<p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
6870 floating point or vector of floating point type. Not all targets support all
6871 types however.</p>
6872
Chris Lattnerf4d252d2006-09-08 06:34:02 +00006873<pre>
Dale Johannesen408f9c12007-10-02 17:47:38 +00006874 declare float @llvm.powi.f32(float %Val, i32 %power)
6875 declare double @llvm.powi.f64(double %Val, i32 %power)
6876 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6877 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6878 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
Chris Lattnerf4d252d2006-09-08 06:34:02 +00006879</pre>
6880
6881<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006882<p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
6883 specified (positive or negative) power. The order of evaluation of
6884 multiplications is not defined. When a vector of floating point type is
6885 used, the second argument remains a scalar integer value.</p>
Chris Lattnerf4d252d2006-09-08 06:34:02 +00006886
6887<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006888<p>The second argument is an integer power, and the first is a value to raise to
6889 that power.</p>
Chris Lattnerf4d252d2006-09-08 06:34:02 +00006890
6891<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006892<p>This function returns the first value raised to the second power with an
6893 unspecified sequence of rounding operations.</p>
Chris Lattnerf4d252d2006-09-08 06:34:02 +00006894
Chris Lattnerf4d252d2006-09-08 06:34:02 +00006895</div>
6896
Dan Gohman91c284c2007-10-15 20:30:11 +00006897<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006898<h4>
Dan Gohman91c284c2007-10-15 20:30:11 +00006899 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006900</h4>
Dan Gohman91c284c2007-10-15 20:30:11 +00006901
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006902<div>
Dan Gohman91c284c2007-10-15 20:30:11 +00006903
6904<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006905<p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
6906 floating point or vector of floating point type. Not all targets support all
6907 types however.</p>
6908
Dan Gohman91c284c2007-10-15 20:30:11 +00006909<pre>
6910 declare float @llvm.sin.f32(float %Val)
6911 declare double @llvm.sin.f64(double %Val)
6912 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6913 declare fp128 @llvm.sin.f128(fp128 %Val)
6914 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6915</pre>
6916
6917<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006918<p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
Dan Gohman91c284c2007-10-15 20:30:11 +00006919
6920<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006921<p>The argument and return value are floating point numbers of the same
6922 type.</p>
Dan Gohman91c284c2007-10-15 20:30:11 +00006923
6924<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006925<p>This function returns the sine of the specified operand, returning the same
6926 values as the libm <tt>sin</tt> functions would, and handles error conditions
6927 in the same way.</p>
Dan Gohman91c284c2007-10-15 20:30:11 +00006928
Dan Gohman91c284c2007-10-15 20:30:11 +00006929</div>
6930
6931<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006932<h4>
Dan Gohman91c284c2007-10-15 20:30:11 +00006933 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006934</h4>
Dan Gohman91c284c2007-10-15 20:30:11 +00006935
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006936<div>
Dan Gohman91c284c2007-10-15 20:30:11 +00006937
6938<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006939<p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
6940 floating point or vector of floating point type. Not all targets support all
6941 types however.</p>
6942
Dan Gohman91c284c2007-10-15 20:30:11 +00006943<pre>
6944 declare float @llvm.cos.f32(float %Val)
6945 declare double @llvm.cos.f64(double %Val)
6946 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6947 declare fp128 @llvm.cos.f128(fp128 %Val)
6948 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6949</pre>
6950
6951<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006952<p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
Dan Gohman91c284c2007-10-15 20:30:11 +00006953
6954<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006955<p>The argument and return value are floating point numbers of the same
6956 type.</p>
Dan Gohman91c284c2007-10-15 20:30:11 +00006957
6958<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006959<p>This function returns the cosine of the specified operand, returning the same
6960 values as the libm <tt>cos</tt> functions would, and handles error conditions
6961 in the same way.</p>
Dan Gohman91c284c2007-10-15 20:30:11 +00006962
Dan Gohman91c284c2007-10-15 20:30:11 +00006963</div>
6964
6965<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006966<h4>
Dan Gohman91c284c2007-10-15 20:30:11 +00006967 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00006968</h4>
Dan Gohman91c284c2007-10-15 20:30:11 +00006969
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00006970<div>
Dan Gohman91c284c2007-10-15 20:30:11 +00006971
6972<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006973<p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
6974 floating point or vector of floating point type. Not all targets support all
6975 types however.</p>
6976
Dan Gohman91c284c2007-10-15 20:30:11 +00006977<pre>
6978 declare float @llvm.pow.f32(float %Val, float %Power)
6979 declare double @llvm.pow.f64(double %Val, double %Power)
6980 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6981 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6982 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6983</pre>
6984
6985<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006986<p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
6987 specified (positive or negative) power.</p>
Dan Gohman91c284c2007-10-15 20:30:11 +00006988
6989<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006990<p>The second argument is a floating point power, and the first is a value to
6991 raise to that power.</p>
Dan Gohman91c284c2007-10-15 20:30:11 +00006992
6993<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00006994<p>This function returns the first value raised to the second power, returning
6995 the same values as the libm <tt>pow</tt> functions would, and handles error
6996 conditions in the same way.</p>
Dan Gohman91c284c2007-10-15 20:30:11 +00006997
Dan Gohman91c284c2007-10-15 20:30:11 +00006998</div>
6999
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007000</div>
7001
Dan Gohman4e9011c2011-05-23 21:13:03 +00007002<!-- _______________________________________________________________________ -->
7003<h4>
7004 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7005</h4>
7006
7007<div>
7008
7009<h5>Syntax:</h5>
7010<p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7011 floating point or vector of floating point type. Not all targets support all
7012 types however.</p>
7013
7014<pre>
7015 declare float @llvm.exp.f32(float %Val)
7016 declare double @llvm.exp.f64(double %Val)
7017 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7018 declare fp128 @llvm.exp.f128(fp128 %Val)
7019 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7020</pre>
7021
7022<h5>Overview:</h5>
7023<p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7024
7025<h5>Arguments:</h5>
7026<p>The argument and return value are floating point numbers of the same
7027 type.</p>
7028
7029<h5>Semantics:</h5>
7030<p>This function returns the same values as the libm <tt>exp</tt> functions
7031 would, and handles error conditions in the same way.</p>
7032
7033</div>
7034
7035<!-- _______________________________________________________________________ -->
7036<h4>
7037 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7038</h4>
7039
7040<div>
7041
7042<h5>Syntax:</h5>
7043<p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7044 floating point or vector of floating point type. Not all targets support all
7045 types however.</p>
7046
7047<pre>
7048 declare float @llvm.log.f32(float %Val)
7049 declare double @llvm.log.f64(double %Val)
7050 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7051 declare fp128 @llvm.log.f128(fp128 %Val)
7052 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7053</pre>
7054
7055<h5>Overview:</h5>
7056<p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7057
7058<h5>Arguments:</h5>
7059<p>The argument and return value are floating point numbers of the same
7060 type.</p>
7061
7062<h5>Semantics:</h5>
7063<p>This function returns the same values as the libm <tt>log</tt> functions
7064 would, and handles error conditions in the same way.</p>
7065
Cameron Zwarich33390842011-07-08 21:39:21 +00007066<h4>
7067 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7068</h4>
7069
7070<div>
7071
7072<h5>Syntax:</h5>
7073<p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7074 floating point or vector of floating point type. Not all targets support all
7075 types however.</p>
7076
7077<pre>
7078 declare float @llvm.fma.f32(float %a, float %b, float %c)
7079 declare double @llvm.fma.f64(double %a, double %b, double %c)
7080 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7081 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7082 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7083</pre>
7084
7085<h5>Overview:</h5>
Cameron Zwarichabc43e62011-07-08 22:13:55 +00007086<p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
Cameron Zwarich33390842011-07-08 21:39:21 +00007087 operation.</p>
7088
7089<h5>Arguments:</h5>
7090<p>The argument and return value are floating point numbers of the same
7091 type.</p>
7092
7093<h5>Semantics:</h5>
7094<p>This function returns the same values as the libm <tt>fma</tt> functions
7095 would.</p>
7096
Dan Gohman4e9011c2011-05-23 21:13:03 +00007097</div>
7098
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007099<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007100<h3>
Nate Begeman7e36c472006-01-13 23:26:38 +00007101 <a name="int_manip">Bit Manipulation Intrinsics</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007102</h3>
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007103
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007104<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007105
7106<p>LLVM provides intrinsics for a few important bit manipulation operations.
7107 These allow efficient code generation for some algorithms.</p>
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007108
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007109<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007110<h4>
Reid Spencera3e435f2007-04-04 02:42:35 +00007111 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007112</h4>
Nate Begeman7e36c472006-01-13 23:26:38 +00007113
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007114<div>
Nate Begeman7e36c472006-01-13 23:26:38 +00007115
7116<h5>Syntax:</h5>
Reid Spencer409e28f2007-04-01 08:04:23 +00007117<p>This is an overloaded intrinsic function. You can use bswap on any integer
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007118 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7119
Nate Begeman7e36c472006-01-13 23:26:38 +00007120<pre>
Chandler Carruth69940402007-08-04 01:51:18 +00007121 declare i16 @llvm.bswap.i16(i16 &lt;id&gt;)
7122 declare i32 @llvm.bswap.i32(i32 &lt;id&gt;)
7123 declare i64 @llvm.bswap.i64(i64 &lt;id&gt;)
Nate Begeman7e36c472006-01-13 23:26:38 +00007124</pre>
7125
7126<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007127<p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7128 values with an even number of bytes (positive multiple of 16 bits). These
7129 are useful for performing operations on data that is not in the target's
7130 native byte order.</p>
Nate Begeman7e36c472006-01-13 23:26:38 +00007131
7132<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007133<p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7134 and low byte of the input i16 swapped. Similarly,
7135 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7136 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7137 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7138 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7139 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7140 more, respectively).</p>
Nate Begeman7e36c472006-01-13 23:26:38 +00007141
7142</div>
7143
7144<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007145<h4>
Reid Spencer0b118202006-01-16 21:12:35 +00007146 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007147</h4>
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007148
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007149<div>
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007150
7151<h5>Syntax:</h5>
Reid Spencer409e28f2007-04-01 08:04:23 +00007152<p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
Owen Andersonf1ac4652011-07-01 21:52:38 +00007153 width, or on any vector with integer elements. Not all targets support all
7154 bit widths or vector types, however.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007155
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007156<pre>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007157 declare i8 @llvm.ctpop.i8(i8 &lt;src&gt;)
Chandler Carruth69940402007-08-04 01:51:18 +00007158 declare i16 @llvm.ctpop.i16(i16 &lt;src&gt;)
Anton Korobeynikovec43a062007-03-22 00:02:17 +00007159 declare i32 @llvm.ctpop.i32(i32 &lt;src&gt;)
Chandler Carruth69940402007-08-04 01:51:18 +00007160 declare i64 @llvm.ctpop.i64(i64 &lt;src&gt;)
7161 declare i256 @llvm.ctpop.i256(i256 &lt;src&gt;)
Owen Andersonf1ac4652011-07-01 21:52:38 +00007162 declare &lt;2 x i32&gt; @llvm.ctpop.v2i32(&lt;2 x i32&gt; &lt;src&gt;)
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007163</pre>
7164
7165<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007166<p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7167 in a value.</p>
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007168
7169<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007170<p>The only argument is the value to be counted. The argument may be of any
Owen Andersonf1ac4652011-07-01 21:52:38 +00007171 integer type, or a vector with integer elements.
7172 The return type must match the argument type.</p>
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007173
7174<h5>Semantics:</h5>
Owen Andersonf1ac4652011-07-01 21:52:38 +00007175<p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7176 element of a vector.</p>
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007177
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007178</div>
7179
7180<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007181<h4>
Chris Lattner8a886be2006-01-16 22:34:14 +00007182 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007183</h4>
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007184
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007185<div>
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007186
7187<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007188<p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
Owen Andersonf1ac4652011-07-01 21:52:38 +00007189 integer bit width, or any vector whose elements are integers. Not all
7190 targets support all bit widths or vector types, however.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007191
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007192<pre>
Chandler Carruth69940402007-08-04 01:51:18 +00007193 declare i8 @llvm.ctlz.i8 (i8 &lt;src&gt;)
7194 declare i16 @llvm.ctlz.i16(i16 &lt;src&gt;)
Anton Korobeynikovec43a062007-03-22 00:02:17 +00007195 declare i32 @llvm.ctlz.i32(i32 &lt;src&gt;)
Chandler Carruth69940402007-08-04 01:51:18 +00007196 declare i64 @llvm.ctlz.i64(i64 &lt;src&gt;)
7197 declare i256 @llvm.ctlz.i256(i256 &lt;src&gt;)
Owen Andersonf1ac4652011-07-01 21:52:38 +00007198 declare &lt;2 x i32&gt; @llvm.ctlz.v2i32(&lt;2 x i32&gt; &lt;src;gt)
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007199</pre>
7200
7201<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007202<p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7203 leading zeros in a variable.</p>
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007204
7205<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007206<p>The only argument is the value to be counted. The argument may be of any
Owen Andersonf1ac4652011-07-01 21:52:38 +00007207 integer type, or any vector type with integer element type.
7208 The return type must match the argument type.</p>
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007209
7210<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007211<p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
Owen Andersonf1ac4652011-07-01 21:52:38 +00007212 zeros in a variable, or within each element of the vector if the operation
7213 is of vector type. If the src == 0 then the result is the size in bits of
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007214 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007215
Andrew Lenharthec370fd2005-05-03 18:01:48 +00007216</div>
Chris Lattner32006282004-06-11 02:28:03 +00007217
Chris Lattnereff29ab2005-05-15 19:39:26 +00007218<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007219<h4>
Chris Lattner8a886be2006-01-16 22:34:14 +00007220 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007221</h4>
Chris Lattnereff29ab2005-05-15 19:39:26 +00007222
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007223<div>
Chris Lattnereff29ab2005-05-15 19:39:26 +00007224
7225<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007226<p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
Owen Andersonf1ac4652011-07-01 21:52:38 +00007227 integer bit width, or any vector of integer elements. Not all targets
7228 support all bit widths or vector types, however.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007229
Chris Lattnereff29ab2005-05-15 19:39:26 +00007230<pre>
Chandler Carruth69940402007-08-04 01:51:18 +00007231 declare i8 @llvm.cttz.i8 (i8 &lt;src&gt;)
7232 declare i16 @llvm.cttz.i16(i16 &lt;src&gt;)
Anton Korobeynikovec43a062007-03-22 00:02:17 +00007233 declare i32 @llvm.cttz.i32(i32 &lt;src&gt;)
Chandler Carruth69940402007-08-04 01:51:18 +00007234 declare i64 @llvm.cttz.i64(i64 &lt;src&gt;)
7235 declare i256 @llvm.cttz.i256(i256 &lt;src&gt;)
Owen Andersonf1ac4652011-07-01 21:52:38 +00007236 declase &lt;2 x i32&gt; @llvm.cttz.v2i32(&lt;2 x i32&gt; &lt;src&gt;)
Chris Lattnereff29ab2005-05-15 19:39:26 +00007237</pre>
7238
7239<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007240<p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7241 trailing zeros.</p>
Chris Lattnereff29ab2005-05-15 19:39:26 +00007242
7243<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007244<p>The only argument is the value to be counted. The argument may be of any
Owen Andersonf1ac4652011-07-01 21:52:38 +00007245 integer type, or a vectory with integer element type.. The return type
7246 must match the argument type.</p>
Chris Lattnereff29ab2005-05-15 19:39:26 +00007247
7248<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007249<p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
Owen Andersonf1ac4652011-07-01 21:52:38 +00007250 zeros in a variable, or within each element of a vector.
7251 If the src == 0 then the result is the size in bits of
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007252 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
Chris Lattnereff29ab2005-05-15 19:39:26 +00007253
Chris Lattnereff29ab2005-05-15 19:39:26 +00007254</div>
7255
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007256</div>
7257
Bill Wendlingda01af72009-02-08 04:04:40 +00007258<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007259<h3>
Bill Wendlingda01af72009-02-08 04:04:40 +00007260 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007261</h3>
Bill Wendlingda01af72009-02-08 04:04:40 +00007262
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007263<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007264
7265<p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
Bill Wendlingda01af72009-02-08 04:04:40 +00007266
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007267<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007268<h4>
7269 <a name="int_sadd_overflow">
7270 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7271 </a>
7272</h4>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007273
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007274<div>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007275
7276<h5>Syntax:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007277<p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007278 on any integer bit width.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007279
7280<pre>
7281 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7282 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7283 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7284</pre>
7285
7286<h5>Overview:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007287<p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007288 a signed addition of the two arguments, and indicate whether an overflow
7289 occurred during the signed summation.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007290
7291<h5>Arguments:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007292<p>The arguments (%a and %b) and the first element of the result structure may
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007293 be of integer types of any bit width, but they must have the same bit
7294 width. The second element of the result structure must be of
7295 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7296 undergo signed addition.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007297
7298<h5>Semantics:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007299<p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007300 a signed addition of the two variables. They return a structure &mdash; the
7301 first element of which is the signed summation, and the second element of
7302 which is a bit specifying if the signed summation resulted in an
7303 overflow.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007304
7305<h5>Examples:</h5>
7306<pre>
7307 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7308 %sum = extractvalue {i32, i1} %res, 0
7309 %obit = extractvalue {i32, i1} %res, 1
7310 br i1 %obit, label %overflow, label %normal
7311</pre>
7312
7313</div>
7314
7315<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007316<h4>
7317 <a name="int_uadd_overflow">
7318 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7319 </a>
7320</h4>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007321
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007322<div>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007323
7324<h5>Syntax:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007325<p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007326 on any integer bit width.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007327
7328<pre>
7329 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7330 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7331 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7332</pre>
7333
7334<h5>Overview:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007335<p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007336 an unsigned addition of the two arguments, and indicate whether a carry
7337 occurred during the unsigned summation.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007338
7339<h5>Arguments:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007340<p>The arguments (%a and %b) and the first element of the result structure may
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007341 be of integer types of any bit width, but they must have the same bit
7342 width. The second element of the result structure must be of
7343 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7344 undergo unsigned addition.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007345
7346<h5>Semantics:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007347<p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007348 an unsigned addition of the two arguments. They return a structure &mdash;
7349 the first element of which is the sum, and the second element of which is a
7350 bit specifying if the unsigned summation resulted in a carry.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007351
7352<h5>Examples:</h5>
7353<pre>
7354 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7355 %sum = extractvalue {i32, i1} %res, 0
7356 %obit = extractvalue {i32, i1} %res, 1
7357 br i1 %obit, label %carry, label %normal
7358</pre>
7359
7360</div>
7361
7362<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007363<h4>
7364 <a name="int_ssub_overflow">
7365 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7366 </a>
7367</h4>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007368
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007369<div>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007370
7371<h5>Syntax:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007372<p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007373 on any integer bit width.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007374
7375<pre>
7376 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7377 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7378 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7379</pre>
7380
7381<h5>Overview:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007382<p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007383 a signed subtraction of the two arguments, and indicate whether an overflow
7384 occurred during the signed subtraction.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007385
7386<h5>Arguments:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007387<p>The arguments (%a and %b) and the first element of the result structure may
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007388 be of integer types of any bit width, but they must have the same bit
7389 width. The second element of the result structure must be of
7390 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7391 undergo signed subtraction.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007392
7393<h5>Semantics:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007394<p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007395 a signed subtraction of the two arguments. They return a structure &mdash;
7396 the first element of which is the subtraction, and the second element of
7397 which is a bit specifying if the signed subtraction resulted in an
7398 overflow.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007399
7400<h5>Examples:</h5>
7401<pre>
7402 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7403 %sum = extractvalue {i32, i1} %res, 0
7404 %obit = extractvalue {i32, i1} %res, 1
7405 br i1 %obit, label %overflow, label %normal
7406</pre>
7407
7408</div>
7409
7410<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007411<h4>
7412 <a name="int_usub_overflow">
7413 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7414 </a>
7415</h4>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007416
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007417<div>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007418
7419<h5>Syntax:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007420<p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007421 on any integer bit width.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007422
7423<pre>
7424 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7425 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7426 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7427</pre>
7428
7429<h5>Overview:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007430<p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007431 an unsigned subtraction of the two arguments, and indicate whether an
7432 overflow occurred during the unsigned subtraction.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007433
7434<h5>Arguments:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007435<p>The arguments (%a and %b) and the first element of the result structure may
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007436 be of integer types of any bit width, but they must have the same bit
7437 width. The second element of the result structure must be of
7438 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7439 undergo unsigned subtraction.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007440
7441<h5>Semantics:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007442<p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007443 an unsigned subtraction of the two arguments. They return a structure &mdash;
7444 the first element of which is the subtraction, and the second element of
7445 which is a bit specifying if the unsigned subtraction resulted in an
7446 overflow.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007447
7448<h5>Examples:</h5>
7449<pre>
7450 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7451 %sum = extractvalue {i32, i1} %res, 0
7452 %obit = extractvalue {i32, i1} %res, 1
7453 br i1 %obit, label %overflow, label %normal
7454</pre>
7455
7456</div>
7457
7458<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007459<h4>
7460 <a name="int_smul_overflow">
7461 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7462 </a>
7463</h4>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007464
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007465<div>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007466
7467<h5>Syntax:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007468<p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007469 on any integer bit width.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007470
7471<pre>
7472 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7473 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7474 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7475</pre>
7476
7477<h5>Overview:</h5>
7478
7479<p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007480 a signed multiplication of the two arguments, and indicate whether an
7481 overflow occurred during the signed multiplication.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007482
7483<h5>Arguments:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007484<p>The arguments (%a and %b) and the first element of the result structure may
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007485 be of integer types of any bit width, but they must have the same bit
7486 width. The second element of the result structure must be of
7487 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7488 undergo signed multiplication.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007489
7490<h5>Semantics:</h5>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007491<p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007492 a signed multiplication of the two arguments. They return a structure &mdash;
7493 the first element of which is the multiplication, and the second element of
7494 which is a bit specifying if the signed multiplication resulted in an
7495 overflow.</p>
Bill Wendlingac1df8e2009-02-08 01:40:31 +00007496
7497<h5>Examples:</h5>
7498<pre>
7499 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7500 %sum = extractvalue {i32, i1} %res, 0
7501 %obit = extractvalue {i32, i1} %res, 1
7502 br i1 %obit, label %overflow, label %normal
7503</pre>
7504
Reid Spencerf86037f2007-04-11 23:23:49 +00007505</div>
7506
Bill Wendling41b485c2009-02-08 23:00:09 +00007507<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007508<h4>
7509 <a name="int_umul_overflow">
7510 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7511 </a>
7512</h4>
Bill Wendling41b485c2009-02-08 23:00:09 +00007513
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007514<div>
Bill Wendling41b485c2009-02-08 23:00:09 +00007515
7516<h5>Syntax:</h5>
Bill Wendling41b485c2009-02-08 23:00:09 +00007517<p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007518 on any integer bit width.</p>
Bill Wendling41b485c2009-02-08 23:00:09 +00007519
7520<pre>
7521 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7522 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7523 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7524</pre>
7525
7526<h5>Overview:</h5>
Bill Wendling41b485c2009-02-08 23:00:09 +00007527<p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007528 a unsigned multiplication of the two arguments, and indicate whether an
7529 overflow occurred during the unsigned multiplication.</p>
Bill Wendling41b485c2009-02-08 23:00:09 +00007530
7531<h5>Arguments:</h5>
Bill Wendling41b485c2009-02-08 23:00:09 +00007532<p>The arguments (%a and %b) and the first element of the result structure may
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007533 be of integer types of any bit width, but they must have the same bit
7534 width. The second element of the result structure must be of
7535 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7536 undergo unsigned multiplication.</p>
Bill Wendling41b485c2009-02-08 23:00:09 +00007537
7538<h5>Semantics:</h5>
Bill Wendling41b485c2009-02-08 23:00:09 +00007539<p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007540 an unsigned multiplication of the two arguments. They return a structure
7541 &mdash; the first element of which is the multiplication, and the second
7542 element of which is a bit specifying if the unsigned multiplication resulted
7543 in an overflow.</p>
Bill Wendling41b485c2009-02-08 23:00:09 +00007544
7545<h5>Examples:</h5>
7546<pre>
7547 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7548 %sum = extractvalue {i32, i1} %res, 0
7549 %obit = extractvalue {i32, i1} %res, 1
7550 br i1 %obit, label %overflow, label %normal
7551</pre>
7552
7553</div>
7554
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007555</div>
7556
Chris Lattner8ff75902004-01-06 05:31:32 +00007557<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007558<h3>
Anton Korobeynikovf02e7302010-03-14 18:42:47 +00007559 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007560</h3>
Anton Korobeynikovf02e7302010-03-14 18:42:47 +00007561
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007562<div>
Anton Korobeynikovf02e7302010-03-14 18:42:47 +00007563
Chris Lattner0cec9c82010-03-15 04:12:21 +00007564<p>Half precision floating point is a storage-only format. This means that it is
7565 a dense encoding (in memory) but does not support computation in the
7566 format.</p>
Chris Lattner82c3dc62010-03-14 23:03:31 +00007567
Chris Lattner0cec9c82010-03-15 04:12:21 +00007568<p>This means that code must first load the half-precision floating point
Chris Lattner82c3dc62010-03-14 23:03:31 +00007569 value as an i16, then convert it to float with <a
7570 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
7571 Computation can then be performed on the float value (including extending to
Chris Lattner0cec9c82010-03-15 04:12:21 +00007572 double etc). To store the value back to memory, it is first converted to
7573 float if needed, then converted to i16 with
Chris Lattner82c3dc62010-03-14 23:03:31 +00007574 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
7575 storing as an i16 value.</p>
Anton Korobeynikovf02e7302010-03-14 18:42:47 +00007576
7577<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007578<h4>
7579 <a name="int_convert_to_fp16">
7580 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
7581 </a>
7582</h4>
Anton Korobeynikovf02e7302010-03-14 18:42:47 +00007583
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007584<div>
Anton Korobeynikovf02e7302010-03-14 18:42:47 +00007585
7586<h5>Syntax:</h5>
7587<pre>
7588 declare i16 @llvm.convert.to.fp16(f32 %a)
7589</pre>
7590
7591<h5>Overview:</h5>
7592<p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7593 a conversion from single precision floating point format to half precision
7594 floating point format.</p>
7595
7596<h5>Arguments:</h5>
7597<p>The intrinsic function contains single argument - the value to be
7598 converted.</p>
7599
7600<h5>Semantics:</h5>
7601<p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7602 a conversion from single precision floating point format to half precision
Chris Lattner0cec9c82010-03-15 04:12:21 +00007603 floating point format. The return value is an <tt>i16</tt> which
Chris Lattner82c3dc62010-03-14 23:03:31 +00007604 contains the converted number.</p>
Anton Korobeynikovf02e7302010-03-14 18:42:47 +00007605
7606<h5>Examples:</h5>
7607<pre>
7608 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7609 store i16 %res, i16* @x, align 2
7610</pre>
7611
7612</div>
7613
7614<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007615<h4>
7616 <a name="int_convert_from_fp16">
7617 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
7618 </a>
7619</h4>
Anton Korobeynikovf02e7302010-03-14 18:42:47 +00007620
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007621<div>
Anton Korobeynikovf02e7302010-03-14 18:42:47 +00007622
7623<h5>Syntax:</h5>
7624<pre>
7625 declare f32 @llvm.convert.from.fp16(i16 %a)
7626</pre>
7627
7628<h5>Overview:</h5>
7629<p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
7630 a conversion from half precision floating point format to single precision
7631 floating point format.</p>
7632
7633<h5>Arguments:</h5>
7634<p>The intrinsic function contains single argument - the value to be
7635 converted.</p>
7636
7637<h5>Semantics:</h5>
7638<p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
Chris Lattner0cec9c82010-03-15 04:12:21 +00007639 conversion from half single precision floating point format to single
Chris Lattner82c3dc62010-03-14 23:03:31 +00007640 precision floating point format. The input half-float value is represented by
7641 an <tt>i16</tt> value.</p>
Anton Korobeynikovf02e7302010-03-14 18:42:47 +00007642
7643<h5>Examples:</h5>
7644<pre>
7645 %a = load i16* @x, align 2
7646 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7647</pre>
7648
7649</div>
7650
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007651</div>
7652
Anton Korobeynikovf02e7302010-03-14 18:42:47 +00007653<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007654<h3>
Chris Lattner8ff75902004-01-06 05:31:32 +00007655 <a name="int_debugger">Debugger Intrinsics</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007656</h3>
Chris Lattner8ff75902004-01-06 05:31:32 +00007657
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007658<div>
Chris Lattner8ff75902004-01-06 05:31:32 +00007659
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007660<p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
7661 prefix), are described in
7662 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
7663 Level Debugging</a> document.</p>
7664
7665</div>
Chris Lattner8ff75902004-01-06 05:31:32 +00007666
Jim Laskeydd4ef1b2007-03-14 19:31:19 +00007667<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007668<h3>
Jim Laskeydd4ef1b2007-03-14 19:31:19 +00007669 <a name="int_eh">Exception Handling Intrinsics</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007670</h3>
Jim Laskeydd4ef1b2007-03-14 19:31:19 +00007671
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007672<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007673
7674<p>The LLVM exception handling intrinsics (which all start with
7675 <tt>llvm.eh.</tt> prefix), are described in
7676 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
7677 Handling</a> document.</p>
7678
Jim Laskeydd4ef1b2007-03-14 19:31:19 +00007679</div>
7680
Tanya Lattner6d806e92007-06-15 20:50:54 +00007681<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007682<h3>
Duncan Sandsf7331b32007-09-11 14:10:23 +00007683 <a name="int_trampoline">Trampoline Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007684</h3>
Duncan Sands36397f52007-07-27 12:58:54 +00007685
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007686<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007687
7688<p>This intrinsic makes it possible to excise one parameter, marked with
Dan Gohmanff235352010-07-02 23:18:08 +00007689 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
7690 The result is a callable
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007691 function pointer lacking the nest parameter - the caller does not need to
7692 provide a value for it. Instead, the value to use is stored in advance in a
7693 "trampoline", a block of memory usually allocated on the stack, which also
7694 contains code to splice the nest value into the argument list. This is used
7695 to implement the GCC nested function address extension.</p>
7696
7697<p>For example, if the function is
7698 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
7699 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
7700 follows:</p>
7701
Benjamin Kramer26fe25f2010-07-13 12:26:09 +00007702<pre class="doc_code">
Duncan Sandsf7331b32007-09-11 14:10:23 +00007703 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7704 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
Rafael Espindola3395fe12011-08-23 18:26:56 +00007705 %p = call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
Duncan Sandsf7331b32007-09-11 14:10:23 +00007706 %fp = bitcast i8* %p to i32 (i32, i32)*
Duncan Sands36397f52007-07-27 12:58:54 +00007707</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007708
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00007709<p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
7710 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007711
Duncan Sands36397f52007-07-27 12:58:54 +00007712<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007713<h4>
7714 <a name="int_it">
7715 '<tt>llvm.init.trampoline</tt>' Intrinsic
7716 </a>
7717</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007718
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007719<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007720
Duncan Sands36397f52007-07-27 12:58:54 +00007721<h5>Syntax:</h5>
7722<pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007723 declare i8* @llvm.init.trampoline(i8* &lt;tramp&gt;, i8* &lt;func&gt;, i8* &lt;nval&gt;)
Duncan Sands36397f52007-07-27 12:58:54 +00007724</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007725
Duncan Sands36397f52007-07-27 12:58:54 +00007726<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007727<p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
7728 function pointer suitable for executing it.</p>
7729
Duncan Sands36397f52007-07-27 12:58:54 +00007730<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007731<p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
7732 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
7733 sufficiently aligned block of memory; this memory is written to by the
7734 intrinsic. Note that the size and the alignment are target-specific - LLVM
7735 currently provides no portable way of determining them, so a front-end that
7736 generates this intrinsic needs to have some target-specific knowledge.
7737 The <tt>func</tt> argument must hold a function bitcast to
7738 an <tt>i8*</tt>.</p>
7739
Duncan Sands36397f52007-07-27 12:58:54 +00007740<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007741<p>The block of memory pointed to by <tt>tramp</tt> is filled with target
7742 dependent code, turning it into a function. A pointer to this function is
7743 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
7744 function pointer type</a> before being called. The new function's signature
7745 is the same as that of <tt>func</tt> with any arguments marked with
7746 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
7747 is allowed, and it must be of pointer type. Calling the new function is
7748 equivalent to calling <tt>func</tt> with the same argument list, but
7749 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
7750 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
7751 by <tt>tramp</tt> is modified, then the effect of any later call to the
7752 returned function pointer is undefined.</p>
7753
Duncan Sands36397f52007-07-27 12:58:54 +00007754</div>
7755
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007756</div>
7757
Duncan Sands36397f52007-07-27 12:58:54 +00007758<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007759<h3>
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007760 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007761</h3>
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007762
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007763<div>
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007764
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007765<p>These intrinsic functions expand the "universal IR" of LLVM to represent
7766 hardware constructs for atomic operations and memory synchronization. This
7767 provides an interface to the hardware, not an interface to the programmer. It
7768 is aimed at a low enough level to allow any programming models or APIs
7769 (Application Programming Interfaces) which need atomic behaviors to map
7770 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
7771 hardware provides a "universal IR" for source languages, it also provides a
7772 starting point for developing a "universal" atomic operation and
7773 synchronization IR.</p>
7774
7775<p>These do <em>not</em> form an API such as high-level threading libraries,
7776 software transaction memory systems, atomic primitives, and intrinsic
7777 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
7778 application libraries. The hardware interface provided by LLVM should allow
7779 a clean implementation of all of these APIs and parallel programming models.
7780 No one model or paradigm should be selected above others unless the hardware
7781 itself ubiquitously does so.</p>
7782
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007783<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007784<h4>
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007785 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007786</h4>
7787
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007788<div>
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007789<h5>Syntax:</h5>
7790<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00007791 declare void @llvm.memory.barrier(i1 &lt;ll&gt;, i1 &lt;ls&gt;, i1 &lt;sl&gt;, i1 &lt;ss&gt;, i1 &lt;device&gt;)
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007792</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007793
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007794<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007795<p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
7796 specific pairs of memory access types.</p>
7797
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007798<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007799<p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
7800 The first four arguments enables a specific barrier as listed below. The
Dan Gohmanb55a1ee2010-03-01 17:41:39 +00007801 fifth argument specifies that the barrier applies to io or device or uncached
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007802 memory.</p>
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007803
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007804<ul>
7805 <li><tt>ll</tt>: load-load barrier</li>
7806 <li><tt>ls</tt>: load-store barrier</li>
7807 <li><tt>sl</tt>: store-load barrier</li>
7808 <li><tt>ss</tt>: store-store barrier</li>
7809 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
7810</ul>
7811
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007812<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007813<p>This intrinsic causes the system to enforce some ordering constraints upon
7814 the loads and stores of the program. This barrier does not
7815 indicate <em>when</em> any events will occur, it only enforces
7816 an <em>order</em> in which they occur. For any of the specified pairs of load
7817 and store operations (f.ex. load-load, or store-load), all of the first
7818 operations preceding the barrier will complete before any of the second
7819 operations succeeding the barrier begin. Specifically the semantics for each
7820 pairing is as follows:</p>
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007821
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007822<ul>
7823 <li><tt>ll</tt>: All loads before the barrier must complete before any load
7824 after the barrier begins.</li>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00007825 <li><tt>ls</tt>: All loads before the barrier must complete before any
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007826 store after the barrier begins.</li>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00007827 <li><tt>ss</tt>: All stores before the barrier must complete before any
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007828 store after the barrier begins.</li>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00007829 <li><tt>sl</tt>: All stores before the barrier must complete before any
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007830 load after the barrier begins.</li>
7831</ul>
7832
7833<p>These semantics are applied with a logical "and" behavior when more than one
7834 is enabled in a single memory barrier intrinsic.</p>
7835
7836<p>Backends may implement stronger barriers than those requested when they do
7837 not support as fine grained a barrier as requested. Some architectures do
7838 not need all types of barriers and on such architectures, these become
7839 noops.</p>
7840
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007841<h5>Example:</h5>
7842<pre>
Victor Hernandez2fee2942009-10-26 23:44:29 +00007843%mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7844%ptr = bitcast i8* %mallocP to i32*
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007845 store i32 4, %ptr
7846
7847%result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
Evan Cheng0b0669a2011-06-29 17:14:00 +00007848 call void @llvm.memory.barrier(i1 false, i1 true, i1 false, i1 false, i1 true)
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007849 <i>; guarantee the above finishes</i>
7850 store i32 8, %ptr <i>; before this begins</i>
7851</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007852
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00007853</div>
7854
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007855<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007856<h4>
Mon P Wang28873102008-06-25 08:15:39 +00007857 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007858</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007859
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007860<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007861
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007862<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007863<p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
7864 any integer bit width and for different address spaces. Not all targets
7865 support all bit widths however.</p>
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007866
7867<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00007868 declare i8 @llvm.atomic.cmp.swap.i8.p0i8(i8* &lt;ptr&gt;, i8 &lt;cmp&gt;, i8 &lt;val&gt;)
7869 declare i16 @llvm.atomic.cmp.swap.i16.p0i16(i16* &lt;ptr&gt;, i16 &lt;cmp&gt;, i16 &lt;val&gt;)
7870 declare i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* &lt;ptr&gt;, i32 &lt;cmp&gt;, i32 &lt;val&gt;)
7871 declare i64 @llvm.atomic.cmp.swap.i64.p0i64(i64* &lt;ptr&gt;, i64 &lt;cmp&gt;, i64 &lt;val&gt;)
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007872</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007873
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007874<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007875<p>This loads a value in memory and compares it to a given value. If they are
7876 equal, it stores a new value into the memory.</p>
7877
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007878<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007879<p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
7880 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
7881 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
7882 this integer type. While any bit width integer may be used, targets may only
7883 lower representations they support in hardware.</p>
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007884
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007885<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007886<p>This entire intrinsic must be executed atomically. It first loads the value
7887 in memory pointed to by <tt>ptr</tt> and compares it with the
7888 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
7889 memory. The loaded value is yielded in all cases. This provides the
7890 equivalent of an atomic compare-and-swap operation within the SSA
7891 framework.</p>
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007892
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007893<h5>Examples:</h5>
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007894<pre>
Victor Hernandez2fee2942009-10-26 23:44:29 +00007895%mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7896%ptr = bitcast i8* %mallocP to i32*
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007897 store i32 4, %ptr
7898
7899%val1 = add i32 4, 4
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00007900%result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 4, %val1)
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007901 <i>; yields {i32}:result1 = 4</i>
7902%stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7903%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7904
7905%val2 = add i32 1, 1
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00007906%result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 5, %val2)
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007907 <i>; yields {i32}:result2 = 8</i>
7908%stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
7909
7910%memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
7911</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007912
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007913</div>
7914
7915<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007916<h4>
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007917 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007918</h4>
7919
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007920<div>
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007921<h5>Syntax:</h5>
7922
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007923<p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
7924 integer bit width. Not all targets support all bit widths however.</p>
7925
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007926<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00007927 declare i8 @llvm.atomic.swap.i8.p0i8(i8* &lt;ptr&gt;, i8 &lt;val&gt;)
7928 declare i16 @llvm.atomic.swap.i16.p0i16(i16* &lt;ptr&gt;, i16 &lt;val&gt;)
7929 declare i32 @llvm.atomic.swap.i32.p0i32(i32* &lt;ptr&gt;, i32 &lt;val&gt;)
7930 declare i64 @llvm.atomic.swap.i64.p0i64(i64* &lt;ptr&gt;, i64 &lt;val&gt;)
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007931</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007932
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007933<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007934<p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
7935 the value from memory. It then stores the value in <tt>val</tt> in the memory
7936 at <tt>ptr</tt>.</p>
7937
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007938<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007939<p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
7940 the <tt>val</tt> argument and the result must be integers of the same bit
7941 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
7942 integer type. The targets may only lower integer representations they
7943 support.</p>
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007944
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007945<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007946<p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
7947 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
7948 equivalent of an atomic swap operation within the SSA framework.</p>
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007949
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007950<h5>Examples:</h5>
7951<pre>
Victor Hernandez2fee2942009-10-26 23:44:29 +00007952%mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7953%ptr = bitcast i8* %mallocP to i32*
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007954 store i32 4, %ptr
7955
7956%val1 = add i32 4, 4
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00007957%result1 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val1)
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007958 <i>; yields {i32}:result1 = 4</i>
7959%stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7960%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7961
7962%val2 = add i32 1, 1
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00007963%result2 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val2)
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007964 <i>; yields {i32}:result2 = 8</i>
7965
7966%stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
7967%memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
7968</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007969
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007970</div>
7971
7972<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007973<h4>
Mon P Wang28873102008-06-25 08:15:39 +00007974 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00007975</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007976
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00007977<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007978
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007979<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007980<p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
7981 any integer bit width. Not all targets support all bit widths however.</p>
7982
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007983<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00007984 declare i8 @llvm.atomic.load.add.i8.p0i8(i8* &lt;ptr&gt;, i8 &lt;delta&gt;)
7985 declare i16 @llvm.atomic.load.add.i16.p0i16(i16* &lt;ptr&gt;, i16 &lt;delta&gt;)
7986 declare i32 @llvm.atomic.load.add.i32.p0i32(i32* &lt;ptr&gt;, i32 &lt;delta&gt;)
7987 declare i64 @llvm.atomic.load.add.i64.p0i64(i64* &lt;ptr&gt;, i64 &lt;delta&gt;)
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007988</pre>
Andrew Lenharthab0b9492008-02-21 06:45:13 +00007989
Bill Wendlinge910b4c2009-07-20 02:29:24 +00007990<h5>Overview:</h5>
7991<p>This intrinsic adds <tt>delta</tt> to the value stored in memory
7992 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7993
7994<h5>Arguments:</h5>
7995<p>The intrinsic takes two arguments, the first a pointer to an integer value
7996 and the second an integer value. The result is also an integer value. These
7997 integer types can have any bit width, but they must all have the same bit
7998 width. The targets may only lower integer representations they support.</p>
7999
Andrew Lenharthab0b9492008-02-21 06:45:13 +00008000<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008001<p>This intrinsic does a series of operations atomically. It first loads the
8002 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
8003 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
Andrew Lenharthab0b9492008-02-21 06:45:13 +00008004
8005<h5>Examples:</h5>
8006<pre>
Victor Hernandez2fee2942009-10-26 23:44:29 +00008007%mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8008%ptr = bitcast i8* %mallocP to i32*
8009 store i32 4, %ptr
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008010%result1 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 4)
Andrew Lenharthab0b9492008-02-21 06:45:13 +00008011 <i>; yields {i32}:result1 = 4</i>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008012%result2 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 2)
Andrew Lenharthab0b9492008-02-21 06:45:13 +00008013 <i>; yields {i32}:result2 = 8</i>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008014%result3 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 5)
Andrew Lenharthab0b9492008-02-21 06:45:13 +00008015 <i>; yields {i32}:result3 = 10</i>
Mon P Wang28873102008-06-25 08:15:39 +00008016%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
Andrew Lenharthab0b9492008-02-21 06:45:13 +00008017</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008018
Andrew Lenharthab0b9492008-02-21 06:45:13 +00008019</div>
8020
Mon P Wang28873102008-06-25 08:15:39 +00008021<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008022<h4>
Mon P Wang28873102008-06-25 08:15:39 +00008023 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008024</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008025
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008026<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008027
Mon P Wang28873102008-06-25 08:15:39 +00008028<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008029<p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
8030 any integer bit width and for different address spaces. Not all targets
8031 support all bit widths however.</p>
8032
Mon P Wang28873102008-06-25 08:15:39 +00008033<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008034 declare i8 @llvm.atomic.load.sub.i8.p0i32(i8* &lt;ptr&gt;, i8 &lt;delta&gt;)
8035 declare i16 @llvm.atomic.load.sub.i16.p0i32(i16* &lt;ptr&gt;, i16 &lt;delta&gt;)
8036 declare i32 @llvm.atomic.load.sub.i32.p0i32(i32* &lt;ptr&gt;, i32 &lt;delta&gt;)
8037 declare i64 @llvm.atomic.load.sub.i64.p0i32(i64* &lt;ptr&gt;, i64 &lt;delta&gt;)
Mon P Wang28873102008-06-25 08:15:39 +00008038</pre>
Mon P Wang28873102008-06-25 08:15:39 +00008039
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008040<h5>Overview:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00008041<p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008042 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
8043
8044<h5>Arguments:</h5>
8045<p>The intrinsic takes two arguments, the first a pointer to an integer value
8046 and the second an integer value. The result is also an integer value. These
8047 integer types can have any bit width, but they must all have the same bit
8048 width. The targets may only lower integer representations they support.</p>
8049
Mon P Wang28873102008-06-25 08:15:39 +00008050<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008051<p>This intrinsic does a series of operations atomically. It first loads the
8052 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
8053 result to <tt>ptr</tt>. It yields the original value stored
8054 at <tt>ptr</tt>.</p>
Mon P Wang28873102008-06-25 08:15:39 +00008055
8056<h5>Examples:</h5>
8057<pre>
Victor Hernandez2fee2942009-10-26 23:44:29 +00008058%mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8059%ptr = bitcast i8* %mallocP to i32*
8060 store i32 8, %ptr
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008061%result1 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 4)
Mon P Wang28873102008-06-25 08:15:39 +00008062 <i>; yields {i32}:result1 = 8</i>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008063%result2 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 2)
Mon P Wang28873102008-06-25 08:15:39 +00008064 <i>; yields {i32}:result2 = 4</i>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008065%result3 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 5)
Mon P Wang28873102008-06-25 08:15:39 +00008066 <i>; yields {i32}:result3 = 2</i>
8067%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
8068</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008069
Mon P Wang28873102008-06-25 08:15:39 +00008070</div>
8071
8072<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008073<h4>
8074 <a name="int_atomic_load_and">
8075 '<tt>llvm.atomic.load.and.*</tt>' Intrinsic
8076 </a>
8077 <br>
8078 <a name="int_atomic_load_nand">
8079 '<tt>llvm.atomic.load.nand.*</tt>' Intrinsic
8080 </a>
8081 <br>
8082 <a name="int_atomic_load_or">
8083 '<tt>llvm.atomic.load.or.*</tt>' Intrinsic
8084 </a>
8085 <br>
8086 <a name="int_atomic_load_xor">
8087 '<tt>llvm.atomic.load.xor.*</tt>' Intrinsic
8088 </a>
8089</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008090
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008091<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008092
Mon P Wang28873102008-06-25 08:15:39 +00008093<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008094<p>These are overloaded intrinsics. You can
8095 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
8096 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
8097 bit width and for different address spaces. Not all targets support all bit
8098 widths however.</p>
Mon P Wang28873102008-06-25 08:15:39 +00008099
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008100<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008101 declare i8 @llvm.atomic.load.and.i8.p0i8(i8* &lt;ptr&gt;, i8 &lt;delta&gt;)
8102 declare i16 @llvm.atomic.load.and.i16.p0i16(i16* &lt;ptr&gt;, i16 &lt;delta&gt;)
8103 declare i32 @llvm.atomic.load.and.i32.p0i32(i32* &lt;ptr&gt;, i32 &lt;delta&gt;)
8104 declare i64 @llvm.atomic.load.and.i64.p0i64(i64* &lt;ptr&gt;, i64 &lt;delta&gt;)
Mon P Wang28873102008-06-25 08:15:39 +00008105</pre>
8106
8107<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008108 declare i8 @llvm.atomic.load.or.i8.p0i8(i8* &lt;ptr&gt;, i8 &lt;delta&gt;)
8109 declare i16 @llvm.atomic.load.or.i16.p0i16(i16* &lt;ptr&gt;, i16 &lt;delta&gt;)
8110 declare i32 @llvm.atomic.load.or.i32.p0i32(i32* &lt;ptr&gt;, i32 &lt;delta&gt;)
8111 declare i64 @llvm.atomic.load.or.i64.p0i64(i64* &lt;ptr&gt;, i64 &lt;delta&gt;)
Mon P Wang28873102008-06-25 08:15:39 +00008112</pre>
8113
8114<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008115 declare i8 @llvm.atomic.load.nand.i8.p0i32(i8* &lt;ptr&gt;, i8 &lt;delta&gt;)
8116 declare i16 @llvm.atomic.load.nand.i16.p0i32(i16* &lt;ptr&gt;, i16 &lt;delta&gt;)
8117 declare i32 @llvm.atomic.load.nand.i32.p0i32(i32* &lt;ptr&gt;, i32 &lt;delta&gt;)
8118 declare i64 @llvm.atomic.load.nand.i64.p0i32(i64* &lt;ptr&gt;, i64 &lt;delta&gt;)
Mon P Wang28873102008-06-25 08:15:39 +00008119</pre>
8120
8121<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008122 declare i8 @llvm.atomic.load.xor.i8.p0i32(i8* &lt;ptr&gt;, i8 &lt;delta&gt;)
8123 declare i16 @llvm.atomic.load.xor.i16.p0i32(i16* &lt;ptr&gt;, i16 &lt;delta&gt;)
8124 declare i32 @llvm.atomic.load.xor.i32.p0i32(i32* &lt;ptr&gt;, i32 &lt;delta&gt;)
8125 declare i64 @llvm.atomic.load.xor.i64.p0i32(i64* &lt;ptr&gt;, i64 &lt;delta&gt;)
Mon P Wang28873102008-06-25 08:15:39 +00008126</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008127
Mon P Wang28873102008-06-25 08:15:39 +00008128<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008129<p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
8130 the value stored in memory at <tt>ptr</tt>. It yields the original value
8131 at <tt>ptr</tt>.</p>
Mon P Wang28873102008-06-25 08:15:39 +00008132
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008133<h5>Arguments:</h5>
8134<p>These intrinsics take two arguments, the first a pointer to an integer value
8135 and the second an integer value. The result is also an integer value. These
8136 integer types can have any bit width, but they must all have the same bit
8137 width. The targets may only lower integer representations they support.</p>
8138
Mon P Wang28873102008-06-25 08:15:39 +00008139<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008140<p>These intrinsics does a series of operations atomically. They first load the
8141 value stored at <tt>ptr</tt>. They then do the bitwise
8142 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
8143 original value stored at <tt>ptr</tt>.</p>
Mon P Wang28873102008-06-25 08:15:39 +00008144
8145<h5>Examples:</h5>
8146<pre>
Victor Hernandez2fee2942009-10-26 23:44:29 +00008147%mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8148%ptr = bitcast i8* %mallocP to i32*
8149 store i32 0x0F0F, %ptr
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008150%result0 = call i32 @llvm.atomic.load.nand.i32.p0i32(i32* %ptr, i32 0xFF)
Mon P Wang28873102008-06-25 08:15:39 +00008151 <i>; yields {i32}:result0 = 0x0F0F</i>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008152%result1 = call i32 @llvm.atomic.load.and.i32.p0i32(i32* %ptr, i32 0xFF)
Mon P Wang28873102008-06-25 08:15:39 +00008153 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008154%result2 = call i32 @llvm.atomic.load.or.i32.p0i32(i32* %ptr, i32 0F)
Mon P Wang28873102008-06-25 08:15:39 +00008155 <i>; yields {i32}:result2 = 0xF0</i>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008156%result3 = call i32 @llvm.atomic.load.xor.i32.p0i32(i32* %ptr, i32 0F)
Mon P Wang28873102008-06-25 08:15:39 +00008157 <i>; yields {i32}:result3 = FF</i>
8158%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
8159</pre>
Mon P Wang28873102008-06-25 08:15:39 +00008160
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008161</div>
Mon P Wang28873102008-06-25 08:15:39 +00008162
8163<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008164<h4>
8165 <a name="int_atomic_load_max">
8166 '<tt>llvm.atomic.load.max.*</tt>' Intrinsic
8167 </a>
8168 <br>
8169 <a name="int_atomic_load_min">
8170 '<tt>llvm.atomic.load.min.*</tt>' Intrinsic
8171 </a>
8172 <br>
8173 <a name="int_atomic_load_umax">
8174 '<tt>llvm.atomic.load.umax.*</tt>' Intrinsic
8175 </a>
8176 <br>
8177 <a name="int_atomic_load_umin">
8178 '<tt>llvm.atomic.load.umin.*</tt>' Intrinsic
8179 </a>
8180</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008181
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008182<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008183
Mon P Wang28873102008-06-25 08:15:39 +00008184<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008185<p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
8186 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
8187 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
8188 address spaces. Not all targets support all bit widths however.</p>
Mon P Wang28873102008-06-25 08:15:39 +00008189
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008190<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008191 declare i8 @llvm.atomic.load.max.i8.p0i8(i8* &lt;ptr&gt;, i8 &lt;delta&gt;)
8192 declare i16 @llvm.atomic.load.max.i16.p0i16(i16* &lt;ptr&gt;, i16 &lt;delta&gt;)
8193 declare i32 @llvm.atomic.load.max.i32.p0i32(i32* &lt;ptr&gt;, i32 &lt;delta&gt;)
8194 declare i64 @llvm.atomic.load.max.i64.p0i64(i64* &lt;ptr&gt;, i64 &lt;delta&gt;)
Mon P Wang28873102008-06-25 08:15:39 +00008195</pre>
8196
8197<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008198 declare i8 @llvm.atomic.load.min.i8.p0i8(i8* &lt;ptr&gt;, i8 &lt;delta&gt;)
8199 declare i16 @llvm.atomic.load.min.i16.p0i16(i16* &lt;ptr&gt;, i16 &lt;delta&gt;)
8200 declare i32 @llvm.atomic.load.min.i32.p0i32(i32* &lt;ptr&gt;, i32 &lt;delta&gt;)
8201 declare i64 @llvm.atomic.load.min.i64.p0i64(i64* &lt;ptr&gt;, i64 &lt;delta&gt;)
Mon P Wang28873102008-06-25 08:15:39 +00008202</pre>
8203
8204<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008205 declare i8 @llvm.atomic.load.umax.i8.p0i8(i8* &lt;ptr&gt;, i8 &lt;delta&gt;)
8206 declare i16 @llvm.atomic.load.umax.i16.p0i16(i16* &lt;ptr&gt;, i16 &lt;delta&gt;)
8207 declare i32 @llvm.atomic.load.umax.i32.p0i32(i32* &lt;ptr&gt;, i32 &lt;delta&gt;)
8208 declare i64 @llvm.atomic.load.umax.i64.p0i64(i64* &lt;ptr&gt;, i64 &lt;delta&gt;)
Mon P Wang28873102008-06-25 08:15:39 +00008209</pre>
8210
8211<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008212 declare i8 @llvm.atomic.load.umin.i8.p0i8(i8* &lt;ptr&gt;, i8 &lt;delta&gt;)
8213 declare i16 @llvm.atomic.load.umin.i16.p0i16(i16* &lt;ptr&gt;, i16 &lt;delta&gt;)
8214 declare i32 @llvm.atomic.load.umin.i32.p0i32(i32* &lt;ptr&gt;, i32 &lt;delta&gt;)
8215 declare i64 @llvm.atomic.load.umin.i64.p0i64(i64* &lt;ptr&gt;, i64 &lt;delta&gt;)
Mon P Wang28873102008-06-25 08:15:39 +00008216</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008217
Mon P Wang28873102008-06-25 08:15:39 +00008218<h5>Overview:</h5>
Eric Christopher6c7e8a02009-12-05 02:46:03 +00008219<p>These intrinsics takes the signed or unsigned minimum or maximum of
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008220 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
8221 original value at <tt>ptr</tt>.</p>
Mon P Wang28873102008-06-25 08:15:39 +00008222
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008223<h5>Arguments:</h5>
8224<p>These intrinsics take two arguments, the first a pointer to an integer value
8225 and the second an integer value. The result is also an integer value. These
8226 integer types can have any bit width, but they must all have the same bit
8227 width. The targets may only lower integer representations they support.</p>
8228
Mon P Wang28873102008-06-25 08:15:39 +00008229<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008230<p>These intrinsics does a series of operations atomically. They first load the
8231 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
8232 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
8233 yield the original value stored at <tt>ptr</tt>.</p>
Mon P Wang28873102008-06-25 08:15:39 +00008234
8235<h5>Examples:</h5>
8236<pre>
Victor Hernandez2fee2942009-10-26 23:44:29 +00008237%mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8238%ptr = bitcast i8* %mallocP to i32*
8239 store i32 7, %ptr
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008240%result0 = call i32 @llvm.atomic.load.min.i32.p0i32(i32* %ptr, i32 -2)
Mon P Wang28873102008-06-25 08:15:39 +00008241 <i>; yields {i32}:result0 = 7</i>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008242%result1 = call i32 @llvm.atomic.load.max.i32.p0i32(i32* %ptr, i32 8)
Mon P Wang28873102008-06-25 08:15:39 +00008243 <i>; yields {i32}:result1 = -2</i>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008244%result2 = call i32 @llvm.atomic.load.umin.i32.p0i32(i32* %ptr, i32 10)
Mon P Wang28873102008-06-25 08:15:39 +00008245 <i>; yields {i32}:result2 = 8</i>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008246%result3 = call i32 @llvm.atomic.load.umax.i32.p0i32(i32* %ptr, i32 30)
Mon P Wang28873102008-06-25 08:15:39 +00008247 <i>; yields {i32}:result3 = 8</i>
8248%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
8249</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008250
Mon P Wang28873102008-06-25 08:15:39 +00008251</div>
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00008252
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008253</div>
Nick Lewyckycc271862009-10-13 07:03:23 +00008254
8255<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008256<h3>
Nick Lewyckycc271862009-10-13 07:03:23 +00008257 <a name="int_memorymarkers">Memory Use Markers</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008258</h3>
Nick Lewyckycc271862009-10-13 07:03:23 +00008259
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008260<div>
Nick Lewyckycc271862009-10-13 07:03:23 +00008261
8262<p>This class of intrinsics exists to information about the lifetime of memory
8263 objects and ranges where variables are immutable.</p>
8264
Nick Lewyckycc271862009-10-13 07:03:23 +00008265<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008266<h4>
Nick Lewyckycc271862009-10-13 07:03:23 +00008267 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008268</h4>
Nick Lewyckycc271862009-10-13 07:03:23 +00008269
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008270<div>
Nick Lewyckycc271862009-10-13 07:03:23 +00008271
8272<h5>Syntax:</h5>
8273<pre>
8274 declare void @llvm.lifetime.start(i64 &lt;size&gt;, i8* nocapture &lt;ptr&gt;)
8275</pre>
8276
8277<h5>Overview:</h5>
8278<p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8279 object's lifetime.</p>
8280
8281<h5>Arguments:</h5>
Nick Lewycky321333e2009-10-13 07:57:33 +00008282<p>The first argument is a constant integer representing the size of the
8283 object, or -1 if it is variable sized. The second argument is a pointer to
8284 the object.</p>
Nick Lewyckycc271862009-10-13 07:03:23 +00008285
8286<h5>Semantics:</h5>
8287<p>This intrinsic indicates that before this point in the code, the value of the
8288 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
Nick Lewycky8d336592009-10-27 16:56:58 +00008289 never be used and has an undefined value. A load from the pointer that
8290 precedes this intrinsic can be replaced with
Nick Lewyckycc271862009-10-13 07:03:23 +00008291 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8292
8293</div>
8294
8295<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008296<h4>
Nick Lewyckycc271862009-10-13 07:03:23 +00008297 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008298</h4>
Nick Lewyckycc271862009-10-13 07:03:23 +00008299
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008300<div>
Nick Lewyckycc271862009-10-13 07:03:23 +00008301
8302<h5>Syntax:</h5>
8303<pre>
8304 declare void @llvm.lifetime.end(i64 &lt;size&gt;, i8* nocapture &lt;ptr&gt;)
8305</pre>
8306
8307<h5>Overview:</h5>
8308<p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8309 object's lifetime.</p>
8310
8311<h5>Arguments:</h5>
Nick Lewycky321333e2009-10-13 07:57:33 +00008312<p>The first argument is a constant integer representing the size of the
8313 object, or -1 if it is variable sized. The second argument is a pointer to
8314 the object.</p>
Nick Lewyckycc271862009-10-13 07:03:23 +00008315
8316<h5>Semantics:</h5>
8317<p>This intrinsic indicates that after this point in the code, the value of the
8318 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8319 never be used and has an undefined value. Any stores into the memory object
8320 following this intrinsic may be removed as dead.
8321
8322</div>
8323
8324<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008325<h4>
Nick Lewyckycc271862009-10-13 07:03:23 +00008326 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008327</h4>
Nick Lewyckycc271862009-10-13 07:03:23 +00008328
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008329<div>
Nick Lewyckycc271862009-10-13 07:03:23 +00008330
8331<h5>Syntax:</h5>
8332<pre>
Nick Lewycky29b6cb42010-11-30 04:13:41 +00008333 declare {}* @llvm.invariant.start(i64 &lt;size&gt;, i8* nocapture &lt;ptr&gt;)
Nick Lewyckycc271862009-10-13 07:03:23 +00008334</pre>
8335
8336<h5>Overview:</h5>
8337<p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8338 a memory object will not change.</p>
8339
8340<h5>Arguments:</h5>
Nick Lewycky321333e2009-10-13 07:57:33 +00008341<p>The first argument is a constant integer representing the size of the
8342 object, or -1 if it is variable sized. The second argument is a pointer to
8343 the object.</p>
Nick Lewyckycc271862009-10-13 07:03:23 +00008344
8345<h5>Semantics:</h5>
8346<p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8347 the return value, the referenced memory location is constant and
8348 unchanging.</p>
8349
8350</div>
8351
8352<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008353<h4>
Nick Lewyckycc271862009-10-13 07:03:23 +00008354 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008355</h4>
Nick Lewyckycc271862009-10-13 07:03:23 +00008356
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008357<div>
Nick Lewyckycc271862009-10-13 07:03:23 +00008358
8359<h5>Syntax:</h5>
8360<pre>
8361 declare void @llvm.invariant.end({}* &lt;start&gt;, i64 &lt;size&gt;, i8* nocapture &lt;ptr&gt;)
8362</pre>
8363
8364<h5>Overview:</h5>
8365<p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8366 a memory object are mutable.</p>
8367
8368<h5>Arguments:</h5>
8369<p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
Nick Lewycky321333e2009-10-13 07:57:33 +00008370 The second argument is a constant integer representing the size of the
8371 object, or -1 if it is variable sized and the third argument is a pointer
8372 to the object.</p>
Nick Lewyckycc271862009-10-13 07:03:23 +00008373
8374<h5>Semantics:</h5>
8375<p>This intrinsic indicates that the memory is mutable again.</p>
8376
8377</div>
8378
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008379</div>
8380
Andrew Lenharth22c5c1b2008-02-16 01:24:58 +00008381<!-- ======================================================================= -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008382<h3>
Tanya Lattner6d806e92007-06-15 20:50:54 +00008383 <a name="int_general">General Intrinsics</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008384</h3>
Tanya Lattner6d806e92007-06-15 20:50:54 +00008385
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008386<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008387
8388<p>This class of intrinsics is designed to be generic and has no specific
8389 purpose.</p>
8390
Tanya Lattner6d806e92007-06-15 20:50:54 +00008391<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008392<h4>
Tanya Lattner6d806e92007-06-15 20:50:54 +00008393 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008394</h4>
Tanya Lattner6d806e92007-06-15 20:50:54 +00008395
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008396<div>
Tanya Lattner6d806e92007-06-15 20:50:54 +00008397
8398<h5>Syntax:</h5>
8399<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008400 declare void @llvm.var.annotation(i8* &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt;)
Tanya Lattner6d806e92007-06-15 20:50:54 +00008401</pre>
8402
8403<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008404<p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
Tanya Lattner6d806e92007-06-15 20:50:54 +00008405
8406<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008407<p>The first argument is a pointer to a value, the second is a pointer to a
8408 global string, the third is a pointer to a global string which is the source
8409 file name, and the last argument is the line number.</p>
Tanya Lattner6d806e92007-06-15 20:50:54 +00008410
8411<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008412<p>This intrinsic allows annotation of local variables with arbitrary strings.
8413 This can be useful for special purpose optimizations that want to look for
John Criswelle865c032011-08-19 16:57:55 +00008414 these annotations. These have no other defined use; they are ignored by code
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008415 generation and optimization.</p>
Tanya Lattner6d806e92007-06-15 20:50:54 +00008416
Tanya Lattner6d806e92007-06-15 20:50:54 +00008417</div>
8418
Tanya Lattnerb6367882007-09-21 22:59:12 +00008419<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008420<h4>
Tanya Lattnere1a8da02007-09-21 23:57:59 +00008421 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008422</h4>
Tanya Lattnerb6367882007-09-21 22:59:12 +00008423
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008424<div>
Tanya Lattnerb6367882007-09-21 22:59:12 +00008425
8426<h5>Syntax:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008427<p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8428 any integer bit width.</p>
8429
Tanya Lattnerb6367882007-09-21 22:59:12 +00008430<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008431 declare i8 @llvm.annotation.i8(i8 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt;)
8432 declare i16 @llvm.annotation.i16(i16 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt;)
8433 declare i32 @llvm.annotation.i32(i32 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt;)
8434 declare i64 @llvm.annotation.i64(i64 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt;)
8435 declare i256 @llvm.annotation.i256(i256 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt;)
Tanya Lattnerb6367882007-09-21 22:59:12 +00008436</pre>
8437
8438<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008439<p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
Tanya Lattnerb6367882007-09-21 22:59:12 +00008440
8441<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008442<p>The first argument is an integer value (result of some expression), the
8443 second is a pointer to a global string, the third is a pointer to a global
8444 string which is the source file name, and the last argument is the line
8445 number. It returns the value of the first argument.</p>
Tanya Lattnerb6367882007-09-21 22:59:12 +00008446
8447<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008448<p>This intrinsic allows annotations to be put on arbitrary expressions with
8449 arbitrary strings. This can be useful for special purpose optimizations that
John Criswelle865c032011-08-19 16:57:55 +00008450 want to look for these annotations. These have no other defined use; they
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008451 are ignored by code generation and optimization.</p>
Tanya Lattnerb6367882007-09-21 22:59:12 +00008452
Tanya Lattnerb6367882007-09-21 22:59:12 +00008453</div>
Jim Laskeydd4ef1b2007-03-14 19:31:19 +00008454
Anton Korobeynikov4cb86182008-01-15 22:31:34 +00008455<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008456<h4>
Anton Korobeynikov4cb86182008-01-15 22:31:34 +00008457 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008458</h4>
Anton Korobeynikov4cb86182008-01-15 22:31:34 +00008459
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008460<div>
Anton Korobeynikov4cb86182008-01-15 22:31:34 +00008461
8462<h5>Syntax:</h5>
8463<pre>
8464 declare void @llvm.trap()
8465</pre>
8466
8467<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008468<p>The '<tt>llvm.trap</tt>' intrinsic.</p>
Anton Korobeynikov4cb86182008-01-15 22:31:34 +00008469
8470<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008471<p>None.</p>
Anton Korobeynikov4cb86182008-01-15 22:31:34 +00008472
8473<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008474<p>This intrinsics is lowered to the target dependent trap instruction. If the
8475 target does not have a trap instruction, this intrinsic will be lowered to
8476 the call of the <tt>abort()</tt> function.</p>
Anton Korobeynikov4cb86182008-01-15 22:31:34 +00008477
Anton Korobeynikov4cb86182008-01-15 22:31:34 +00008478</div>
8479
Bill Wendling69e4adb2008-11-19 05:56:17 +00008480<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008481<h4>
Misha Brukmandccb0252008-11-22 23:55:29 +00008482 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008483</h4>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008484
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008485<div>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008486
Bill Wendling69e4adb2008-11-19 05:56:17 +00008487<h5>Syntax:</h5>
8488<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008489 declare void @llvm.stackprotector(i8* &lt;guard&gt;, i8** &lt;slot&gt;)
Bill Wendling69e4adb2008-11-19 05:56:17 +00008490</pre>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008491
Bill Wendling69e4adb2008-11-19 05:56:17 +00008492<h5>Overview:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008493<p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8494 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8495 ensure that it is placed on the stack before local variables.</p>
8496
Bill Wendling69e4adb2008-11-19 05:56:17 +00008497<h5>Arguments:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008498<p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8499 arguments. The first argument is the value loaded from the stack
8500 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8501 that has enough space to hold the value of the guard.</p>
8502
Bill Wendling69e4adb2008-11-19 05:56:17 +00008503<h5>Semantics:</h5>
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008504<p>This intrinsic causes the prologue/epilogue inserter to force the position of
8505 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8506 stack. This is to ensure that if a local variable on the stack is
8507 overwritten, it will destroy the value of the guard. When the function exits,
Bill Wendling1b383ba2010-10-27 01:07:41 +00008508 the guard on the stack is checked against the original guard. If they are
Bill Wendlinge910b4c2009-07-20 02:29:24 +00008509 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8510 function.</p>
8511
Bill Wendling69e4adb2008-11-19 05:56:17 +00008512</div>
8513
Eric Christopher0e671492009-11-30 08:03:53 +00008514<!-- _______________________________________________________________________ -->
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008515<h4>
Eric Christopher0e671492009-11-30 08:03:53 +00008516 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
NAKAMURA Takumi05d02652011-04-18 23:59:50 +00008517</h4>
Eric Christopher0e671492009-11-30 08:03:53 +00008518
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008519<div>
Eric Christopher0e671492009-11-30 08:03:53 +00008520
8521<h5>Syntax:</h5>
8522<pre>
Dan Gohman3dfb3cf2010-05-28 17:07:41 +00008523 declare i32 @llvm.objectsize.i32(i8* &lt;object&gt;, i1 &lt;type&gt;)
8524 declare i64 @llvm.objectsize.i64(i8* &lt;object&gt;, i1 &lt;type&gt;)
Eric Christopher0e671492009-11-30 08:03:53 +00008525</pre>
8526
8527<h5>Overview:</h5>
Bill Wendling1b383ba2010-10-27 01:07:41 +00008528<p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8529 the optimizers to determine at compile time whether a) an operation (like
8530 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8531 runtime check for overflow isn't necessary. An object in this context means
8532 an allocation of a specific class, structure, array, or other object.</p>
Eric Christopher0e671492009-11-30 08:03:53 +00008533
8534<h5>Arguments:</h5>
Bill Wendling1b383ba2010-10-27 01:07:41 +00008535<p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
Eric Christopher8295a0a2009-12-23 00:29:49 +00008536 argument is a pointer to or into the <tt>object</tt>. The second argument
Bill Wendling1b383ba2010-10-27 01:07:41 +00008537 is a boolean 0 or 1. This argument determines whether you want the
8538 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
Eric Christopher8295a0a2009-12-23 00:29:49 +00008539 1, variables are not allowed.</p>
8540
Eric Christopher0e671492009-11-30 08:03:53 +00008541<h5>Semantics:</h5>
8542<p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
Bill Wendling1b383ba2010-10-27 01:07:41 +00008543 representing the size of the object concerned, or <tt>i32/i64 -1 or 0</tt>,
8544 depending on the <tt>type</tt> argument, if the size cannot be determined at
8545 compile time.</p>
Eric Christopher0e671492009-11-30 08:03:53 +00008546
8547</div>
8548
NAKAMURA Takumif5af6ad2011-04-23 00:30:22 +00008549</div>
8550
8551</div>
8552
Chris Lattner00950542001-06-06 20:29:01 +00008553<!-- *********************************************************************** -->
Chris Lattner00950542001-06-06 20:29:01 +00008554<hr>
Misha Brukmandaa4cb02004-03-01 17:47:27 +00008555<address>
8556 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
Misha Brukman44408702008-12-11 17:34:48 +00008557 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
Misha Brukmandaa4cb02004-03-01 17:47:27 +00008558 <a href="http://validator.w3.org/check/referer"><img
Misha Brukman44408702008-12-11 17:34:48 +00008559 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
Misha Brukmandaa4cb02004-03-01 17:47:27 +00008560
8561 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
NAKAMURA Takumib9a33632011-04-09 02:13:37 +00008562 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
Misha Brukmandaa4cb02004-03-01 17:47:27 +00008563 Last modified: $Date$
8564</address>
Chris Lattnerc7d3ab32008-01-04 04:33:49 +00008565
Misha Brukman9d0919f2003-11-08 01:05:38 +00008566</body>
8567</html>