blob: ed47f1f00ed5582dd9220dc1aa21db61b5d64238 [file] [log] [blame] [edit]
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
"http://www.w3.org/TR/html4/strict.dtd">
<html>
<head>
<title>LLVM Assembly Language Reference Manual</title>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<meta name="author" content="Chris Lattner">
<meta name="description"
content="LLVM Assembly Language Reference Manual.">
<link rel="stylesheet" href="_static/llvm.css" type="text/css">
</head>
<body>
<h1>LLVM Language Reference Manual</h1>
<ol>
<li><a href="#abstract">Abstract</a></li>
<li><a href="#introduction">Introduction</a></li>
<li><a href="#identifiers">Identifiers</a></li>
<li><a href="#highlevel">High Level Structure</a>
<ol>
<li><a href="#modulestructure">Module Structure</a></li>
<li><a href="#linkage">Linkage Types</a>
<ol>
<li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
<li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
<li><a href="#linkage_linker_private_weak">'<tt>linker_private_weak</tt>' Linkage</a></li>
<li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
<li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
<li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
<li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
<li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
<li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
<li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
<li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
<li><a href="#linkage_linkonce_odr_auto_hide">'<tt>linkonce_odr_auto_hide</tt>' Linkage</a></li>
<li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
<li><a href="#linkage_external">'<tt>external</tt>' Linkage</a></li>
<li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
<li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
</ol>
</li>
<li><a href="#callingconv">Calling Conventions</a></li>
<li><a href="#namedtypes">Named Types</a></li>
<li><a href="#globalvars">Global Variables</a></li>
<li><a href="#functionstructure">Functions</a></li>
<li><a href="#aliasstructure">Aliases</a></li>
<li><a href="#namedmetadatastructure">Named Metadata</a></li>
<li><a href="#paramattrs">Parameter Attributes</a></li>
<li><a href="#fnattrs">Function Attributes</a></li>
<li><a href="#gc">Garbage Collector Names</a></li>
<li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
<li><a href="#datalayout">Data Layout</a></li>
<li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
<li><a href="#volatile">Volatile Memory Accesses</a></li>
<li><a href="#memmodel">Memory Model for Concurrent Operations</a></li>
<li><a href="#ordering">Atomic Memory Ordering Constraints</a></li>
</ol>
</li>
<li><a href="#typesystem">Type System</a>
<ol>
<li><a href="#t_classifications">Type Classifications</a></li>
<li><a href="#t_primitive">Primitive Types</a>
<ol>
<li><a href="#t_integer">Integer Type</a></li>
<li><a href="#t_floating">Floating Point Types</a></li>
<li><a href="#t_x86mmx">X86mmx Type</a></li>
<li><a href="#t_void">Void Type</a></li>
<li><a href="#t_label">Label Type</a></li>
<li><a href="#t_metadata">Metadata Type</a></li>
</ol>
</li>
<li><a href="#t_derived">Derived Types</a>
<ol>
<li><a href="#t_aggregate">Aggregate Types</a>
<ol>
<li><a href="#t_array">Array Type</a></li>
<li><a href="#t_struct">Structure Type</a></li>
<li><a href="#t_opaque">Opaque Structure Types</a></li>
<li><a href="#t_vector">Vector Type</a></li>
</ol>
</li>
<li><a href="#t_function">Function Type</a></li>
<li><a href="#t_pointer">Pointer Type</a></li>
</ol>
</li>
</ol>
</li>
<li><a href="#constants">Constants</a>
<ol>
<li><a href="#simpleconstants">Simple Constants</a></li>
<li><a href="#complexconstants">Complex Constants</a></li>
<li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
<li><a href="#undefvalues">Undefined Values</a></li>
<li><a href="#poisonvalues">Poison Values</a></li>
<li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
<li><a href="#constantexprs">Constant Expressions</a></li>
</ol>
</li>
<li><a href="#othervalues">Other Values</a>
<ol>
<li><a href="#inlineasm">Inline Assembler Expressions</a></li>
<li><a href="#metadata">Metadata Nodes and Metadata Strings</a>
<ol>
<li><a href="#tbaa">'<tt>tbaa</tt>' Metadata</a></li>
<li><a href="#tbaa.struct">'<tt>tbaa.struct</tt>' Metadata</a></li>
<li><a href="#fpmath">'<tt>fpmath</tt>' Metadata</a></li>
<li><a href="#range">'<tt>range</tt>' Metadata</a></li>
</ol>
</li>
</ol>
</li>
<li><a href="#module_flags">Module Flags Metadata</a>
<ol>
<li><a href="#objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a></li>
</ol>
</li>
<li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
<ol>
<li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
<li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
Global Variable</a></li>
<li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
Global Variable</a></li>
<li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
Global Variable</a></li>
</ol>
</li>
<li><a href="#instref">Instruction Reference</a>
<ol>
<li><a href="#terminators">Terminator Instructions</a>
<ol>
<li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
<li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
<li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
<li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
<li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
<li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li>
<li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
</ol>
</li>
<li><a href="#binaryops">Binary Operations</a>
<ol>
<li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
<li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
<li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
<li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
<li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
<li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
<li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
<li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
<li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
<li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
<li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
<li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
</ol>
</li>
<li><a href="#bitwiseops">Bitwise Binary Operations</a>
<ol>
<li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
<li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
<li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
<li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
<li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
<li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
</ol>
</li>
<li><a href="#vectorops">Vector Operations</a>
<ol>
<li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
<li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
<li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
</ol>
</li>
<li><a href="#aggregateops">Aggregate Operations</a>
<ol>
<li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
<li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
</ol>
</li>
<li><a href="#memoryops">Memory Access and Addressing Operations</a>
<ol>
<li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
<li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
<li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
<li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
<li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
<li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
<li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
</ol>
</li>
<li><a href="#convertops">Conversion Operations</a>
<ol>
<li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
<li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
<li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
<li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
<li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
<li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
<li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
<li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
<li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
<li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
<li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
<li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
</ol>
</li>
<li><a href="#otherops">Other Operations</a>
<ol>
<li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
<li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
<li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
<li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
<li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
<li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
<li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
</ol>
</li>
</ol>
</li>
<li><a href="#intrinsics">Intrinsic Functions</a>
<ol>
<li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
<ol>
<li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
<li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
<li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
</ol>
</li>
<li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
<ol>
<li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
<li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
<li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
</ol>
</li>
<li><a href="#int_codegen">Code Generator Intrinsics</a>
<ol>
<li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
<li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
<li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
<li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
<li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
<li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
<li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
</ol>
</li>
<li><a href="#int_libc">Standard C Library Intrinsics</a>
<ol>
<li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
<li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
<li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
<li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
<li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
<li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
<li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
<li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
<li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
<li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
<li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
<li><a href="#int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a></li>
<li><a href="#int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a></li>
</ol>
</li>
<li><a href="#int_manip">Bit Manipulation Intrinsics</a>
<ol>
<li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
<li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
<li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
<li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
</ol>
</li>
<li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
<ol>
<li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
<li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
<li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
<li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
<li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
<li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
</ol>
</li>
<li><a href="#spec_arithmetic">Specialised Arithmetic Intrinsics</a>
<ol>
<li><a href="#fmuladd">'<tt>llvm.fmuladd</tt> Intrinsic</a></li>
</ol>
</li>
<li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
<ol>
<li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
<li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
</ol>
</li>
<li><a href="#int_debugger">Debugger intrinsics</a></li>
<li><a href="#int_eh">Exception Handling intrinsics</a></li>
<li><a href="#int_trampoline">Trampoline Intrinsics</a>
<ol>
<li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
<li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
</ol>
</li>
<li><a href="#int_memorymarkers">Memory Use Markers</a>
<ol>
<li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li>
<li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li>
<li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li>
<li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li>
</ol>
</li>
<li><a href="#int_general">General intrinsics</a>
<ol>
<li><a href="#int_var_annotation">
'<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
<li><a href="#int_annotation">
'<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
<li><a href="#int_trap">
'<tt>llvm.trap</tt>' Intrinsic</a></li>
<li><a href="#int_debugtrap">
'<tt>llvm.debugtrap</tt>' Intrinsic</a></li>
<li><a href="#int_stackprotector">
'<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
<li><a href="#int_objectsize">
'<tt>llvm.objectsize</tt>' Intrinsic</a></li>
<li><a href="#int_expect">
'<tt>llvm.expect</tt>' Intrinsic</a></li>
<li><a href="#int_donothing">
'<tt>llvm.donothing</tt>' Intrinsic</a></li>
</ol>
</li>
</ol>
</li>
</ol>
<div class="doc_author">
<p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
</div>
<!-- *********************************************************************** -->
<h2><a name="abstract">Abstract</a></h2>
<!-- *********************************************************************** -->
<div>
<p>This document is a reference manual for the LLVM assembly language. LLVM is
a Static Single Assignment (SSA) based representation that provides type
safety, low-level operations, flexibility, and the capability of representing
'all' high-level languages cleanly. It is the common code representation
used throughout all phases of the LLVM compilation strategy.</p>
</div>
<!-- *********************************************************************** -->
<h2><a name="introduction">Introduction</a></h2>
<!-- *********************************************************************** -->
<div>
<p>The LLVM code representation is designed to be used in three different forms:
as an in-memory compiler IR, as an on-disk bitcode representation (suitable
for fast loading by a Just-In-Time compiler), and as a human readable
assembly language representation. This allows LLVM to provide a powerful
intermediate representation for efficient compiler transformations and
analysis, while providing a natural means to debug and visualize the
transformations. The three different forms of LLVM are all equivalent. This
document describes the human readable representation and notation.</p>
<p>The LLVM representation aims to be light-weight and low-level while being
expressive, typed, and extensible at the same time. It aims to be a
"universal IR" of sorts, by being at a low enough level that high-level ideas
may be cleanly mapped to it (similar to how microprocessors are "universal
IR's", allowing many source languages to be mapped to them). By providing
type information, LLVM can be used as the target of optimizations: for
example, through pointer analysis, it can be proven that a C automatic
variable is never accessed outside of the current function, allowing it to
be promoted to a simple SSA value instead of a memory location.</p>
<!-- _______________________________________________________________________ -->
<h4>
<a name="wellformed">Well-Formedness</a>
</h4>
<div>
<p>It is important to note that this document describes 'well formed' LLVM
assembly language. There is a difference between what the parser accepts and
what is considered 'well formed'. For example, the following instruction is
syntactically okay, but not well formed:</p>
<pre class="doc_code">
%x = <a href="#i_add">add</a> i32 1, %x
</pre>
<p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
LLVM infrastructure provides a verification pass that may be used to verify
that an LLVM module is well formed. This pass is automatically run by the
parser after parsing input assembly and by the optimizer before it outputs
bitcode. The violations pointed out by the verifier pass indicate bugs in
transformation passes or input to the parser.</p>
</div>
</div>
<!-- Describe the typesetting conventions here. -->
<!-- *********************************************************************** -->
<h2><a name="identifiers">Identifiers</a></h2>
<!-- *********************************************************************** -->
<div>
<p>LLVM identifiers come in two basic types: global and local. Global
identifiers (functions, global variables) begin with the <tt>'@'</tt>
character. Local identifiers (register names, types) begin with
the <tt>'%'</tt> character. Additionally, there are three different formats
for identifiers, for different purposes:</p>
<ol>
<li>Named values are represented as a string of characters with their prefix.
For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
<tt>%a.really.long.identifier</tt>. The actual regular expression used is
'<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
other characters in their names can be surrounded with quotes. Special
characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
ASCII code for the character in hexadecimal. In this way, any character
can be used in a name value, even quotes themselves.</li>
<li>Unnamed values are represented as an unsigned numeric value with their
prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
<li>Constants, which are described in a <a href="#constants">section about
constants</a>, below.</li>
</ol>
<p>LLVM requires that values start with a prefix for two reasons: Compilers
don't need to worry about name clashes with reserved words, and the set of
reserved words may be expanded in the future without penalty. Additionally,
unnamed identifiers allow a compiler to quickly come up with a temporary
variable without having to avoid symbol table conflicts.</p>
<p>Reserved words in LLVM are very similar to reserved words in other
languages. There are keywords for different opcodes
('<tt><a href="#i_add">add</a></tt>',
'<tt><a href="#i_bitcast">bitcast</a></tt>',
'<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
('<tt><a href="#t_void">void</a></tt>',
'<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
reserved words cannot conflict with variable names, because none of them
start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
<p>Here is an example of LLVM code to multiply the integer variable
'<tt>%X</tt>' by 8:</p>
<p>The easy way:</p>
<pre class="doc_code">
%result = <a href="#i_mul">mul</a> i32 %X, 8
</pre>
<p>After strength reduction:</p>
<pre class="doc_code">
%result = <a href="#i_shl">shl</a> i32 %X, i8 3
</pre>
<p>And the hard way:</p>
<pre class="doc_code">
%0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
%1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
%result = <a href="#i_add">add</a> i32 %1, %1
</pre>
<p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
lexical features of LLVM:</p>
<ol>
<li>Comments are delimited with a '<tt>;</tt>' and go until the end of
line.</li>
<li>Unnamed temporaries are created when the result of a computation is not
assigned to a named value.</li>
<li>Unnamed temporaries are numbered sequentially</li>
</ol>
<p>It also shows a convention that we follow in this document. When
demonstrating instructions, we will follow an instruction with a comment that
defines the type and name of value produced. Comments are shown in italic
text.</p>
</div>
<!-- *********************************************************************** -->
<h2><a name="highlevel">High Level Structure</a></h2>
<!-- *********************************************************************** -->
<div>
<!-- ======================================================================= -->
<h3>
<a name="modulestructure">Module Structure</a>
</h3>
<div>
<p>LLVM programs are composed of <tt>Module</tt>s, each of which is a
translation unit of the input programs. Each module consists of functions,
global variables, and symbol table entries. Modules may be combined together
with the LLVM linker, which merges function (and global variable)
definitions, resolves forward declarations, and merges symbol table
entries. Here is an example of the "hello world" module:</p>
<pre class="doc_code">
<i>; Declare the string constant as a global constant.</i>&nbsp;
<a href="#identifiers">@.str</a> = <a href="#linkage_private">private</a>&nbsp;<a href="#globalvars">unnamed_addr</a>&nbsp;<a href="#globalvars">constant</a>&nbsp;<a href="#t_array">[13 x i8]</a> c"hello world\0A\00"&nbsp;
<i>; External declaration of the puts function</i>&nbsp;
<a href="#functionstructure">declare</a> i32 @puts(i8* <a href="#nocapture">nocapture</a>) <a href="#fnattrs">nounwind</a>&nbsp;
<i>; Definition of main function</i>
define i32 @main() { <i>; i32()* </i>&nbsp;
<i>; Convert [13 x i8]* to i8 *...</i>&nbsp;
%cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.str, i64 0, i64 0
<i>; Call puts function to write out the string to stdout.</i>&nbsp;
<a href="#i_call">call</a> i32 @puts(i8* %cast210)
<a href="#i_ret">ret</a> i32 0&nbsp;
}
<i>; Named metadata</i>
!1 = metadata !{i32 42}
!foo = !{!1, null}
</pre>
<p>This example is made up of a <a href="#globalvars">global variable</a> named
"<tt>.str</tt>", an external declaration of the "<tt>puts</tt>" function,
a <a href="#functionstructure">function definition</a> for
"<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
"<tt>foo</tt>".</p>
<p>In general, a module is made up of a list of global values (where both
functions and global variables are global values). Global values are
represented by a pointer to a memory location (in this case, a pointer to an
array of char, and a pointer to a function), and have one of the
following <a href="#linkage">linkage types</a>.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="linkage">Linkage Types</a>
</h3>
<div>
<p>All Global Variables and Functions have one of the following types of
linkage:</p>
<dl>
<dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
<dd>Global values with "<tt>private</tt>" linkage are only directly accessible
by objects in the current module. In particular, linking code into a
module with an private global value may cause the private to be renamed as
necessary to avoid collisions. Because the symbol is private to the
module, all references can be updated. This doesn't show up in any symbol
table in the object file.</dd>
<dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
<dd>Similar to <tt>private</tt>, but the symbol is passed through the
assembler and evaluated by the linker. Unlike normal strong symbols, they
are removed by the linker from the final linked image (executable or
dynamic library).</dd>
<dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
<dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
<tt>linker_private_weak</tt> symbols are subject to coalescing by the
linker. The symbols are removed by the linker from the final linked image
(executable or dynamic library).</dd>
<dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
<dd>Similar to private, but the value shows as a local symbol
(<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
<dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
<dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
into the object file corresponding to the LLVM module. They exist to
allow inlining and other optimizations to take place given knowledge of
the definition of the global, which is known to be somewhere outside the
module. Globals with <tt>available_externally</tt> linkage are allowed to
be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
This linkage type is only allowed on definitions, not declarations.</dd>
<dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
<dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
the same name when linkage occurs. This can be used to implement
some forms of inline functions, templates, or other code which must be
generated in each translation unit that uses it, but where the body may
be overridden with a more definitive definition later. Unreferenced
<tt>linkonce</tt> globals are allowed to be discarded. Note that
<tt>linkonce</tt> linkage does not actually allow the optimizer to
inline the body of this function into callers because it doesn't know if
this definition of the function is the definitive definition within the
program or whether it will be overridden by a stronger definition.
To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
linkage.</dd>
<dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
<dd>"<tt>weak</tt>" linkage has the same merging semantics as
<tt>linkonce</tt> linkage, except that unreferenced globals with
<tt>weak</tt> linkage may not be discarded. This is used for globals that
are declared "weak" in C source code.</dd>
<dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
<dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
global scope.
Symbols with "<tt>common</tt>" linkage are merged in the same way as
<tt>weak symbols</tt>, and they may not be deleted if unreferenced.
<tt>common</tt> symbols may not have an explicit section,
must have a zero initializer, and may not be marked '<a
href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
have common linkage.</dd>
<dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
<dd>"<tt>appending</tt>" linkage may only be applied to global variables of
pointer to array type. When two global variables with appending linkage
are linked together, the two global arrays are appended together. This is
the LLVM, typesafe, equivalent of having the system linker append together
"sections" with identical names when .o files are linked.</dd>
<dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
<dd>The semantics of this linkage follow the ELF object file model: the symbol
is weak until linked, if not linked, the symbol becomes null instead of
being an undefined reference.</dd>
<dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
<dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
<dd>Some languages allow differing globals to be merged, such as two functions
with different semantics. Other languages, such as <tt>C++</tt>, ensure
that only equivalent globals are ever merged (the "one definition rule"
&mdash; "ODR"). Such languages can use the <tt>linkonce_odr</tt>
and <tt>weak_odr</tt> linkage types to indicate that the global will only
be merged with equivalent globals. These linkage types are otherwise the
same as their non-<tt>odr</tt> versions.</dd>
<dt><tt><b><a name="linkage_linkonce_odr_auto_hide">linkonce_odr_auto_hide</a></b></tt></dt>
<dd>Similar to "<tt>linkonce_odr</tt>", but nothing in the translation unit
takes the address of this definition. For instance, functions that had an
inline definition, but the compiler decided not to inline it.
<tt>linkonce_odr_auto_hide</tt> may have only <tt>default</tt> visibility.
The symbols are removed by the linker from the final linked image
(executable or dynamic library).</dd>
<dt><tt><b><a name="linkage_external">external</a></b></tt></dt>
<dd>If none of the above identifiers are used, the global is externally
visible, meaning that it participates in linkage and can be used to
resolve external symbol references.</dd>
</dl>
<p>The next two types of linkage are targeted for Microsoft Windows platform
only. They are designed to support importing (exporting) symbols from (to)
DLLs (Dynamic Link Libraries).</p>
<dl>
<dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
<dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
or variable via a global pointer to a pointer that is set up by the DLL
exporting the symbol. On Microsoft Windows targets, the pointer name is
formed by combining <code>__imp_</code> and the function or variable
name.</dd>
<dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
<dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
pointer to a pointer in a DLL, so that it can be referenced with the
<tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
name is formed by combining <code>__imp_</code> and the function or
variable name.</dd>
</dl>
<p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
another module defined a "<tt>.LC0</tt>" variable and was linked with this
one, one of the two would be renamed, preventing a collision. Since
"<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
declarations), they are accessible outside of the current module.</p>
<p>It is illegal for a function <i>declaration</i> to have any linkage type
other than <tt>external</tt>, <tt>dllimport</tt>
or <tt>extern_weak</tt>.</p>
<p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
or <tt>weak_odr</tt> linkages.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="callingconv">Calling Conventions</a>
</h3>
<div>
<p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
and <a href="#i_invoke">invokes</a> can all have an optional calling
convention specified for the call. The calling convention of any pair of
dynamic caller/callee must match, or the behavior of the program is
undefined. The following calling conventions are supported by LLVM, and more
may be added in the future:</p>
<dl>
<dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
<dd>This calling convention (the default if no other calling convention is
specified) matches the target C calling conventions. This calling
convention supports varargs function calls and tolerates some mismatch in
the declared prototype and implemented declaration of the function (as
does normal C).</dd>
<dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
<dd>This calling convention attempts to make calls as fast as possible
(e.g. by passing things in registers). This calling convention allows the
target to use whatever tricks it wants to produce fast code for the
target, without having to conform to an externally specified ABI
(Application Binary Interface).
<a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
when this or the GHC convention is used.</a> This calling convention
does not support varargs and requires the prototype of all callees to
exactly match the prototype of the function definition.</dd>
<dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
<dd>This calling convention attempts to make code in the caller as efficient
as possible under the assumption that the call is not commonly executed.
As such, these calls often preserve all registers so that the call does
not break any live ranges in the caller side. This calling convention
does not support varargs and requires the prototype of all callees to
exactly match the prototype of the function definition.</dd>
<dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
<dd>This calling convention has been implemented specifically for use by the
<a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
It passes everything in registers, going to extremes to achieve this by
disabling callee save registers. This calling convention should not be
used lightly but only for specific situations such as an alternative to
the <em>register pinning</em> performance technique often used when
implementing functional programming languages.At the moment only X86
supports this convention and it has the following limitations:
<ul>
<li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
floating point types are supported.</li>
<li>On <em>X86-64</em> only supports up to 10 bit type parameters and
6 floating point parameters.</li>
</ul>
This calling convention supports
<a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
requires both the caller and callee are using it.
</dd>
<dt><b>"<tt>cc &lt;<em>n</em>&gt;</tt>" - Numbered convention</b>:</dt>
<dd>Any calling convention may be specified by number, allowing
target-specific calling conventions to be used. Target specific calling
conventions start at 64.</dd>
</dl>
<p>More calling conventions can be added/defined on an as-needed basis, to
support Pascal conventions or any other well-known target-independent
convention.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="visibility">Visibility Styles</a>
</h3>
<div>
<p>All Global Variables and Functions have one of the following visibility
styles:</p>
<dl>
<dt><b>"<tt>default</tt>" - Default style</b>:</dt>
<dd>On targets that use the ELF object file format, default visibility means
that the declaration is visible to other modules and, in shared libraries,
means that the declared entity may be overridden. On Darwin, default
visibility means that the declaration is visible to other modules. Default
visibility corresponds to "external linkage" in the language.</dd>
<dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
<dd>Two declarations of an object with hidden visibility refer to the same
object if they are in the same shared object. Usually, hidden visibility
indicates that the symbol will not be placed into the dynamic symbol
table, so no other module (executable or shared library) can reference it
directly.</dd>
<dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
<dd>On ELF, protected visibility indicates that the symbol will be placed in
the dynamic symbol table, but that references within the defining module
will bind to the local symbol. That is, the symbol cannot be overridden by
another module.</dd>
</dl>
</div>
<!-- ======================================================================= -->
<h3>
<a name="namedtypes">Named Types</a>
</h3>
<div>
<p>LLVM IR allows you to specify name aliases for certain types. This can make
it easier to read the IR and make the IR more condensed (particularly when
recursive types are involved). An example of a name specification is:</p>
<pre class="doc_code">
%mytype = type { %mytype*, i32 }
</pre>
<p>You may give a name to any <a href="#typesystem">type</a> except
"<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
is expected with the syntax "%mytype".</p>
<p>Note that type names are aliases for the structural type that they indicate,
and that you can therefore specify multiple names for the same type. This
often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
uses structural typing, the name is not part of the type. When printing out
LLVM IR, the printer will pick <em>one name</em> to render all types of a
particular shape. This means that if you have code where two different
source types end up having the same LLVM type, that the dumper will sometimes
print the "wrong" or unexpected type. This is an important design point and
isn't going to change.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="globalvars">Global Variables</a>
</h3>
<div>
<p>Global variables define regions of memory allocated at compilation time
instead of run-time. Global variables may optionally be initialized, may
have an explicit section to be placed in, and may have an optional explicit
alignment specified.</p>
<p>A variable may be defined as <tt>thread_local</tt>, which
means that it will not be shared by threads (each thread will have a
separated copy of the variable). Not all targets support thread-local
variables. Optionally, a TLS model may be specified:</p>
<dl>
<dt><b><tt>localdynamic</tt></b>:</dt>
<dd>For variables that are only used within the current shared library.</dd>
<dt><b><tt>initialexec</tt></b>:</dt>
<dd>For variables in modules that will not be loaded dynamically.</dd>
<dt><b><tt>localexec</tt></b>:</dt>
<dd>For variables defined in the executable and only used within it.</dd>
</dl>
<p>The models correspond to the ELF TLS models; see
<a href="http://people.redhat.com/drepper/tls.pdf">ELF
Handling For Thread-Local Storage</a> for more information on under which
circumstances the different models may be used. The target may choose a
different TLS model if the specified model is not supported, or if a better
choice of model can be made.</p>
<p>A variable may be defined as a global
"constant," which indicates that the contents of the variable
will <b>never</b> be modified (enabling better optimization, allowing the
global data to be placed in the read-only section of an executable, etc).
Note that variables that need runtime initialization cannot be marked
"constant" as there is a store to the variable.</p>
<p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
constant, even if the final definition of the global is not. This capability
can be used to enable slightly better optimization of the program, but
requires the language definition to guarantee that optimizations based on the
'constantness' are valid for the translation units that do not include the
definition.</p>
<p>As SSA values, global variables define pointer values that are in scope
(i.e. they dominate) all basic blocks in the program. Global variables
always define a pointer to their "content" type because they describe a
region of memory, and all memory objects in LLVM are accessed through
pointers.</p>
<p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
that the address is not significant, only the content. Constants marked
like this can be merged with other constants if they have the same
initializer. Note that a constant with significant address <em>can</em>
be merged with a <tt>unnamed_addr</tt> constant, the result being a
constant whose address is significant.</p>
<p>A global variable may be declared to reside in a target-specific numbered
address space. For targets that support them, address spaces may affect how
optimizations are performed and/or what target instructions are used to
access the variable. The default address space is zero. The address space
qualifier must precede any other attributes.</p>
<p>LLVM allows an explicit section to be specified for globals. If the target
supports it, it will emit globals to the section specified.</p>
<p>An explicit alignment may be specified for a global, which must be a power
of 2. If not present, or if the alignment is set to zero, the alignment of
the global is set by the target to whatever it feels convenient. If an
explicit alignment is specified, the global is forced to have exactly that
alignment. Targets and optimizers are not allowed to over-align the global
if the global has an assigned section. In this case, the extra alignment
could be observable: for example, code could assume that the globals are
densely packed in their section and try to iterate over them as an array,
alignment padding would break this iteration.</p>
<p>For example, the following defines a global in a numbered address space with
an initializer, section, and alignment:</p>
<pre class="doc_code">
@G = addrspace(5) constant float 1.0, section "foo", align 4
</pre>
<p>The following example defines a thread-local global with
the <tt>initialexec</tt> TLS model:</p>
<pre class="doc_code">
@G = thread_local(initialexec) global i32 0, align 4
</pre>
</div>
<!-- ======================================================================= -->
<h3>
<a name="functionstructure">Functions</a>
</h3>
<div>
<p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
optional <a href="#linkage">linkage type</a>, an optional
<a href="#visibility">visibility style</a>, an optional
<a href="#callingconv">calling convention</a>,
an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
<a href="#paramattrs">parameter attribute</a> for the return type, a function
name, a (possibly empty) argument list (each with optional
<a href="#paramattrs">parameter attributes</a>), optional
<a href="#fnattrs">function attributes</a>, an optional section, an optional
alignment, an optional <a href="#gc">garbage collector name</a>, an opening
curly brace, a list of basic blocks, and a closing curly brace.</p>
<p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
optional <a href="#linkage">linkage type</a>, an optional
<a href="#visibility">visibility style</a>, an optional
<a href="#callingconv">calling convention</a>,
an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
<a href="#paramattrs">parameter attribute</a> for the return type, a function
name, a possibly empty list of arguments, an optional alignment, and an
optional <a href="#gc">garbage collector name</a>.</p>
<p>A function definition contains a list of basic blocks, forming the CFG
(Control Flow Graph) for the function. Each basic block may optionally start
with a label (giving the basic block a symbol table entry), contains a list
of instructions, and ends with a <a href="#terminators">terminator</a>
instruction (such as a branch or function return).</p>
<p>The first basic block in a function is special in two ways: it is immediately
executed on entrance to the function, and it is not allowed to have
predecessor basic blocks (i.e. there can not be any branches to the entry
block of a function). Because the block can have no predecessors, it also
cannot have any <a href="#i_phi">PHI nodes</a>.</p>
<p>LLVM allows an explicit section to be specified for functions. If the target
supports it, it will emit functions to the section specified.</p>
<p>An explicit alignment may be specified for a function. If not present, or if
the alignment is set to zero, the alignment of the function is set by the
target to whatever it feels convenient. If an explicit alignment is
specified, the function is forced to have at least that much alignment. All
alignments must be a power of 2.</p>
<p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
be significant and two identical functions can be merged.</p>
<h5>Syntax:</h5>
<pre class="doc_code">
define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
[<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
&lt;ResultType&gt; @&lt;FunctionName&gt; ([argument list])
[<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
[<a href="#gc">gc</a>] { ... }
</pre>
</div>
<!-- ======================================================================= -->
<h3>
<a name="aliasstructure">Aliases</a>
</h3>
<div>
<p>Aliases act as "second name" for the aliasee value (which can be either
function, global variable, another alias or bitcast of global value). Aliases
may have an optional <a href="#linkage">linkage type</a>, and an
optional <a href="#visibility">visibility style</a>.</p>
<h5>Syntax:</h5>
<pre class="doc_code">
@&lt;Name&gt; = alias [Linkage] [Visibility] &lt;AliaseeTy&gt; @&lt;Aliasee&gt;
</pre>
</div>
<!-- ======================================================================= -->
<h3>
<a name="namedmetadatastructure">Named Metadata</a>
</h3>
<div>
<p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
nodes</a> (but not metadata strings) are the only valid operands for
a named metadata.</p>
<h5>Syntax:</h5>
<pre class="doc_code">
; Some unnamed metadata nodes, which are referenced by the named metadata.
!0 = metadata !{metadata !"zero"}
!1 = metadata !{metadata !"one"}
!2 = metadata !{metadata !"two"}
; A named metadata.
!name = !{!0, !1, !2}
</pre>
</div>
<!-- ======================================================================= -->
<h3>
<a name="paramattrs">Parameter Attributes</a>
</h3>
<div>
<p>The return type and each parameter of a function type may have a set of
<i>parameter attributes</i> associated with them. Parameter attributes are
used to communicate additional information about the result or parameters of
a function. Parameter attributes are considered to be part of the function,
not of the function type, so functions with different parameter attributes
can have the same function type.</p>
<p>Parameter attributes are simple keywords that follow the type specified. If
multiple parameter attributes are needed, they are space separated. For
example:</p>
<pre class="doc_code">
declare i32 @printf(i8* noalias nocapture, ...)
declare i32 @atoi(i8 zeroext)
declare signext i8 @returns_signed_char()
</pre>
<p>Note that any attributes for the function result (<tt>nounwind</tt>,
<tt>readonly</tt>) come immediately after the argument list.</p>
<p>Currently, only the following parameter attributes are defined:</p>
<dl>
<dt><tt><b>zeroext</b></tt></dt>
<dd>This indicates to the code generator that the parameter or return value
should be zero-extended to the extent required by the target's ABI (which
is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
parameter) or the callee (for a return value).</dd>
<dt><tt><b>signext</b></tt></dt>
<dd>This indicates to the code generator that the parameter or return value
should be sign-extended to the extent required by the target's ABI (which
is usually 32-bits) by the caller (for a parameter) or the callee (for a
return value).</dd>
<dt><tt><b>inreg</b></tt></dt>
<dd>This indicates that this parameter or return value should be treated in a
special target-dependent fashion during while emitting code for a function
call or return (usually, by putting it in a register as opposed to memory,
though some targets use it to distinguish between two different kinds of
registers). Use of this attribute is target-specific.</dd>
<dt><tt><b><a name="byval">byval</a></b></tt></dt>
<dd><p>This indicates that the pointer parameter should really be passed by
value to the function. The attribute implies that a hidden copy of the
pointee
is made between the caller and the callee, so the callee is unable to
modify the value in the caller. This attribute is only valid on LLVM
pointer arguments. It is generally used to pass structs and arrays by
value, but is also valid on pointers to scalars. The copy is considered
to belong to the caller not the callee (for example,
<tt><a href="#readonly">readonly</a></tt> functions should not write to
<tt>byval</tt> parameters). This is not a valid attribute for return
values.</p>
<p>The byval attribute also supports specifying an alignment with
the align attribute. It indicates the alignment of the stack slot to
form and the known alignment of the pointer specified to the call site. If
the alignment is not specified, then the code generator makes a
target-specific assumption.</p></dd>
<dt><tt><b><a name="sret">sret</a></b></tt></dt>
<dd>This indicates that the pointer parameter specifies the address of a
structure that is the return value of the function in the source program.
This pointer must be guaranteed by the caller to be valid: loads and
stores to the structure may be assumed by the callee to not to trap and
to be properly aligned. This may only be applied to the first parameter.
This is not a valid attribute for return values. </dd>
<dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
<dd>This indicates that pointer values
<a href="#pointeraliasing"><i>based</i></a> on the argument or return
value do not alias pointer values which are not <i>based</i> on it,
ignoring certain "irrelevant" dependencies.
For a call to the parent function, dependencies between memory
references from before or after the call and from those during the call
are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
return value used in that call.
The caller shares the responsibility with the callee for ensuring that
these requirements are met.
For further details, please see the discussion of the NoAlias response in
<a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
<br>
Note that this definition of <tt>noalias</tt> is intentionally
similar to the definition of <tt>restrict</tt> in C99 for function
arguments, though it is slightly weaker.
<br>
For function return values, C99's <tt>restrict</tt> is not meaningful,
while LLVM's <tt>noalias</tt> is.
</dd>
<dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
<dd>This indicates that the callee does not make any copies of the pointer
that outlive the callee itself. This is not a valid attribute for return
values.</dd>
<dt><tt><b><a name="nest">nest</a></b></tt></dt>
<dd>This indicates that the pointer parameter can be excised using the
<a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
attribute for return values.</dd>
</dl>
</div>
<!-- ======================================================================= -->
<h3>
<a name="gc">Garbage Collector Names</a>
</h3>
<div>
<p>Each function may specify a garbage collector name, which is simply a
string:</p>
<pre class="doc_code">
define void @f() gc "name" { ... }
</pre>
<p>The compiler declares the supported values of <i>name</i>. Specifying a
collector which will cause the compiler to alter its output in order to
support the named garbage collection algorithm.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="fnattrs">Function Attributes</a>
</h3>
<div>
<p>Function attributes are set to communicate additional information about a
function. Function attributes are considered to be part of the function, not
of the function type, so functions with different parameter attributes can
have the same function type.</p>
<p>Function attributes are simple keywords that follow the type specified. If
multiple attributes are needed, they are space separated. For example:</p>
<pre class="doc_code">
define void @f() noinline { ... }
define void @f() alwaysinline { ... }
define void @f() alwaysinline optsize { ... }
define void @f() optsize { ... }
</pre>
<dl>
<dt><tt><b>address_safety</b></tt></dt>
<dd>This attribute indicates that the address safety analysis
is enabled for this function. </dd>
<dt><tt><b>alignstack(&lt;<em>n</em>&gt;)</b></tt></dt>
<dd>This attribute indicates that, when emitting the prologue and epilogue,
the backend should forcibly align the stack pointer. Specify the
desired alignment, which must be a power of two, in parentheses.
<dt><tt><b>alwaysinline</b></tt></dt>
<dd>This attribute indicates that the inliner should attempt to inline this
function into callers whenever possible, ignoring any active inlining size
threshold for this caller.</dd>
<dt><tt><b>nonlazybind</b></tt></dt>
<dd>This attribute suppresses lazy symbol binding for the function. This
may make calls to the function faster, at the cost of extra program
startup time if the function is not called during program startup.</dd>
<dt><tt><b>inlinehint</b></tt></dt>
<dd>This attribute indicates that the source code contained a hint that inlining
this function is desirable (such as the "inline" keyword in C/C++). It
is just a hint; it imposes no requirements on the inliner.</dd>
<dt><tt><b>naked</b></tt></dt>
<dd>This attribute disables prologue / epilogue emission for the function.
This can have very system-specific consequences.</dd>
<dt><tt><b>noimplicitfloat</b></tt></dt>
<dd>This attributes disables implicit floating point instructions.</dd>
<dt><tt><b>noinline</b></tt></dt>
<dd>This attribute indicates that the inliner should never inline this
function in any situation. This attribute may not be used together with
the <tt>alwaysinline</tt> attribute.</dd>
<dt><tt><b>noredzone</b></tt></dt>
<dd>This attribute indicates that the code generator should not use a red
zone, even if the target-specific ABI normally permits it.</dd>
<dt><tt><b>noreturn</b></tt></dt>
<dd>This function attribute indicates that the function never returns
normally. This produces undefined behavior at runtime if the function
ever does dynamically return.</dd>
<dt><tt><b>nounwind</b></tt></dt>
<dd>This function attribute indicates that the function never returns with an
unwind or exceptional control flow. If the function does unwind, its
runtime behavior is undefined.</dd>
<dt><tt><b>optsize</b></tt></dt>
<dd>This attribute suggests that optimization passes and code generator passes
make choices that keep the code size of this function low, and otherwise
do optimizations specifically to reduce code size.</dd>
<dt><tt><b>readnone</b></tt></dt>
<dd>This attribute indicates that the function computes its result (or decides
to unwind an exception) based strictly on its arguments, without
dereferencing any pointer arguments or otherwise accessing any mutable
state (e.g. memory, control registers, etc) visible to caller functions.
It does not write through any pointer arguments
(including <tt><a href="#byval">byval</a></tt> arguments) and never
changes any state visible to callers. This means that it cannot unwind
exceptions by calling the <tt>C++</tt> exception throwing methods.</dd>
<dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
<dd>This attribute indicates that the function does not write through any
pointer arguments (including <tt><a href="#byval">byval</a></tt>
arguments) or otherwise modify any state (e.g. memory, control registers,
etc) visible to caller functions. It may dereference pointer arguments
and read state that may be set in the caller. A readonly function always
returns the same value (or unwinds an exception identically) when called
with the same set of arguments and global state. It cannot unwind an
exception by calling the <tt>C++</tt> exception throwing methods.</dd>
<dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
<dd>This attribute indicates that this function can return twice. The
C <code>setjmp</code> is an example of such a function. The compiler
disables some optimizations (like tail calls) in the caller of these
functions.</dd>
<dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
<dd>This attribute indicates that the function should emit a stack smashing
protector. It is in the form of a "canary"&mdash;a random value placed on
the stack before the local variables that's checked upon return from the
function to see if it has been overwritten. A heuristic is used to
determine if a function needs stack protectors or not.<br>
<br>
If a function that has an <tt>ssp</tt> attribute is inlined into a
function that doesn't have an <tt>ssp</tt> attribute, then the resulting
function will have an <tt>ssp</tt> attribute.</dd>
<dt><tt><b>sspreq</b></tt></dt>
<dd>This attribute indicates that the function should <em>always</em> emit a
stack smashing protector. This overrides
the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
<br>
If a function that has an <tt>sspreq</tt> attribute is inlined into a
function that doesn't have an <tt>sspreq</tt> attribute or which has
an <tt>ssp</tt> attribute, then the resulting function will have
an <tt>sspreq</tt> attribute.</dd>
<dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
<dd>This attribute indicates that the ABI being targeted requires that
an unwind table entry be produce for this function even if we can
show that no exceptions passes by it. This is normally the case for
the ELF x86-64 abi, but it can be disabled for some compilation
units.</dd>
</dl>
</div>
<!-- ======================================================================= -->
<h3>
<a name="moduleasm">Module-Level Inline Assembly</a>
</h3>
<div>
<p>Modules may contain "module-level inline asm" blocks, which corresponds to
the GCC "file scope inline asm" blocks. These blocks are internally
concatenated by LLVM and treated as a single unit, but may be separated in
the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
<pre class="doc_code">
module asm "inline asm code goes here"
module asm "more can go here"
</pre>
<p>The strings can contain any character by escaping non-printable characters.
The escape sequence used is simply "\xx" where "xx" is the two digit hex code
for the number.</p>
<p>The inline asm code is simply printed to the machine code .s file when
assembly code is generated.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="datalayout">Data Layout</a>
</h3>
<div>
<p>A module may specify a target specific data layout string that specifies how
data is to be laid out in memory. The syntax for the data layout is
simply:</p>
<pre class="doc_code">
target datalayout = "<i>layout specification</i>"
</pre>
<p>The <i>layout specification</i> consists of a list of specifications
separated by the minus sign character ('-'). Each specification starts with
a letter and may include other information after the letter to define some
aspect of the data layout. The specifications accepted are as follows:</p>
<dl>
<dt><tt>E</tt></dt>
<dd>Specifies that the target lays out data in big-endian form. That is, the
bits with the most significance have the lowest address location.</dd>
<dt><tt>e</tt></dt>
<dd>Specifies that the target lays out data in little-endian form. That is,
the bits with the least significance have the lowest address
location.</dd>
<dt><tt>S<i>size</i></tt></dt>
<dd>Specifies the natural alignment of the stack in bits. Alignment promotion
of stack variables is limited to the natural stack alignment to avoid
dynamic stack realignment. The stack alignment must be a multiple of
8-bits. If omitted, the natural stack alignment defaults to "unspecified",
which does not prevent any alignment promotions.</dd>
<dt><tt>p[n]:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
<dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
<i>preferred</i> alignments for address space <i>n</i>. All sizes are in
bits. Specifying the <i>pref</i> alignment is optional. If omitted, the
preceding <tt>:</tt> should be omitted too. The address space,
<i>n</i> is optional, and if not specified, denotes the default address
space 0. The value of <i>n</i> must be in the range [1,2^23).</dd>
<dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
<dd>This specifies the alignment for an integer type of a given bit
<i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
<dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
<dd>This specifies the alignment for a vector type of a given bit
<i>size</i>.</dd>
<dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
<dd>This specifies the alignment for a floating point type of a given bit
<i>size</i>. Only values of <i>size</i> that are supported by the target
will work. 32 (float) and 64 (double) are supported on all targets;
80 or 128 (different flavors of long double) are also supported on some
targets.
<dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
<dd>This specifies the alignment for an aggregate type of a given bit
<i>size</i>.</dd>
<dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
<dd>This specifies the alignment for a stack object of a given bit
<i>size</i>.</dd>
<dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
<dd>This specifies a set of native integer widths for the target CPU
in bits. For example, it might contain "n32" for 32-bit PowerPC,
"n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
this set are considered to support most general arithmetic
operations efficiently.</dd>
</dl>
<p>When constructing the data layout for a given target, LLVM starts with a
default set of specifications which are then (possibly) overridden by the
specifications in the <tt>datalayout</tt> keyword. The default specifications
are given in this list:</p>
<ul>
<li><tt>E</tt> - big endian</li>
<li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
<li><tt>p1:32:32:32</tt> - 32-bit pointers with 32-bit alignment for
address space 1</li>
<li><tt>p2:16:32:32</tt> - 16-bit pointers with 32-bit alignment for
address space 2</li>
<li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
<li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
<li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
<li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
<li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
alignment of 64-bits</li>
<li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
<li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
<li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
<li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
<li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
<li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
</ul>
<p>When LLVM is determining the alignment for a given type, it uses the
following rules:</p>
<ol>
<li>If the type sought is an exact match for one of the specifications, that
specification is used.</li>
<li>If no match is found, and the type sought is an integer type, then the
smallest integer type that is larger than the bitwidth of the sought type
is used. If none of the specifications are larger than the bitwidth then
the largest integer type is used. For example, given the default
specifications above, the i7 type will use the alignment of i8 (next
largest) while both i65 and i256 will use the alignment of i64 (largest
specified).</li>
<li>If no match is found, and the type sought is a vector type, then the
largest vector type that is smaller than the sought vector type will be
used as a fall back. This happens because &lt;128 x double&gt; can be
implemented in terms of 64 &lt;2 x double&gt;, for example.</li>
</ol>
<p>The function of the data layout string may not be what you expect. Notably,
this is not a specification from the frontend of what alignment the code
generator should use.</p>
<p>Instead, if specified, the target data layout is required to match what the
ultimate <em>code generator</em> expects. This string is used by the
mid-level optimizers to
improve code, and this only works if it matches what the ultimate code
generator uses. If you would like to generate IR that does not embed this
target-specific detail into the IR, then you don't have to specify the
string. This will disable some optimizations that require precise layout
information, but this also prevents those optimizations from introducing
target specificity into the IR.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="pointeraliasing">Pointer Aliasing Rules</a>
</h3>
<div>
<p>Any memory access must be done through a pointer value associated
with an address range of the memory access, otherwise the behavior
is undefined. Pointer values are associated with address ranges
according to the following rules:</p>
<ul>
<li>A pointer value is associated with the addresses associated with
any value it is <i>based</i> on.
<li>An address of a global variable is associated with the address
range of the variable's storage.</li>
<li>The result value of an allocation instruction is associated with
the address range of the allocated storage.</li>
<li>A null pointer in the default address-space is associated with
no address.</li>
<li>An integer constant other than zero or a pointer value returned
from a function not defined within LLVM may be associated with address
ranges allocated through mechanisms other than those provided by
LLVM. Such ranges shall not overlap with any ranges of addresses
allocated by mechanisms provided by LLVM.</li>
</ul>
<p>A pointer value is <i>based</i> on another pointer value according
to the following rules:</p>
<ul>
<li>A pointer value formed from a
<tt><a href="#i_getelementptr">getelementptr</a></tt> operation
is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
<li>The result value of a
<tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
of the <tt>bitcast</tt>.</li>
<li>A pointer value formed by an
<tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
pointer values that contribute (directly or indirectly) to the
computation of the pointer's value.</li>
<li>The "<i>based</i> on" relationship is transitive.</li>
</ul>
<p>Note that this definition of <i>"based"</i> is intentionally
similar to the definition of <i>"based"</i> in C99, though it is
slightly weaker.</p>
<p>LLVM IR does not associate types with memory. The result type of a
<tt><a href="#i_load">load</a></tt> merely indicates the size and
alignment of the memory from which to load, as well as the
interpretation of the value. The first operand type of a
<tt><a href="#i_store">store</a></tt> similarly only indicates the size
and alignment of the store.</p>
<p>Consequently, type-based alias analysis, aka TBAA, aka
<tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
additional information which specialized optimization passes may use
to implement type-based alias analysis.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="volatile">Volatile Memory Accesses</a>
</h3>
<div>
<p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
href="#i_store"><tt>store</tt></a>s, and <a
href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
The optimizers must not change the number of volatile operations or change their
order of execution relative to other volatile operations. The optimizers
<i>may</i> change the order of volatile operations relative to non-volatile
operations. This is not Java's "volatile" and has no cross-thread
synchronization behavior.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="memmodel">Memory Model for Concurrent Operations</a>
</h3>
<div>
<p>The LLVM IR does not define any way to start parallel threads of execution
or to register signal handlers. Nonetheless, there are platform-specific
ways to create them, and we define LLVM IR's behavior in their presence. This
model is inspired by the C++0x memory model.</p>
<p>For a more informal introduction to this model, see the
<a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
<p>We define a <i>happens-before</i> partial order as the least partial order
that</p>
<ul>
<li>Is a superset of single-thread program order, and</li>
<li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
<tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
by platform-specific techniques, like pthread locks, thread
creation, thread joining, etc., and by atomic instructions.
(See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
</li>
</ul>
<p>Note that program order does not introduce <i>happens-before</i> edges
between a thread and signals executing inside that thread.</p>
<p>Every (defined) read operation (load instructions, memcpy, atomic
loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
(defined) write operations (store instructions, atomic
stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
initialized globals are considered to have a write of the initializer which is
atomic and happens before any other read or write of the memory in question.
For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
any write to the same byte, except:</p>
<ul>
<li>If <var>write<sub>1</sub></var> happens before
<var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
does not see <var>write<sub>1</sub></var>.
<li>If <var>R<sub>byte</sub></var> happens before
<var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
see <var>write<sub>3</sub></var>.
</ul>
<p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
<ul>
<li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
is supposed to give guarantees which can support
<code>sig_atomic_t</code> in C/C++, and may be used for accesses to
addresses which do not behave like normal memory. It does not generally
provide cross-thread synchronization.)
<li>Otherwise, if there is no write to the same byte that happens before
<var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
<tt>undef</tt> for that byte.
<li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
<var>R<sub>byte</sub></var> returns the value written by that
write.</li>
<li>Otherwise, if <var>R</var> is atomic, and all the writes
<var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
values written. See the <a href="#ordering">Atomic Memory Ordering
Constraints</a> section for additional constraints on how the choice
is made.
<li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
</ul>
<p><var>R</var> returns the value composed of the series of bytes it read.
This implies that some bytes within the value may be <tt>undef</tt>
<b>without</b> the entire value being <tt>undef</tt>. Note that this only
defines the semantics of the operation; it doesn't mean that targets will
emit more than one instruction to read the series of bytes.</p>
<p>Note that in cases where none of the atomic intrinsics are used, this model
places only one restriction on IR transformations on top of what is required
for single-threaded execution: introducing a store to a byte which might not
otherwise be stored is not allowed in general. (Specifically, in the case
where another thread might write to and read from an address, introducing a
store can change a load that may see exactly one write into a load that may
see multiple writes.)</p>
<!-- FIXME: This model assumes all targets where concurrency is relevant have
a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
none of the backends currently in the tree fall into this category; however,
there might be targets which care. If there are, we want a paragraph
like the following:
Targets may specify that stores narrower than a certain width are not
available; on such a target, for the purposes of this model, treat any
non-atomic write with an alignment or width less than the minimum width
as if it writes to the relevant surrounding bytes.
-->
</div>
<!-- ======================================================================= -->
<h3>
<a name="ordering">Atomic Memory Ordering Constraints</a>
</h3>
<div>
<p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
<a href="#i_atomicrmw"><code>atomicrmw</code></a>,
<a href="#i_fence"><code>fence</code></a>,
<a href="#i_load"><code>atomic load</code></a>, and
<a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
that determines which other atomic instructions on the same address they
<i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
but are somewhat more colloquial. If these descriptions aren't precise enough,
check those specs (see spec references in the
<a href="Atomics.html#introduction">atomics guide</a>).
<a href="#i_fence"><code>fence</code></a> instructions
treat these orderings somewhat differently since they don't take an address.
See that instruction's documentation for details.</p>
<p>For a simpler introduction to the ordering constraints, see the
<a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
<dl>
<dt><code>unordered</code></dt>
<dd>The set of values that can be read is governed by the happens-before
partial order. A value cannot be read unless some operation wrote it.
This is intended to provide a guarantee strong enough to model Java's
non-volatile shared variables. This ordering cannot be specified for
read-modify-write operations; it is not strong enough to make them atomic
in any interesting way.</dd>
<dt><code>monotonic</code></dt>
<dd>In addition to the guarantees of <code>unordered</code>, there is a single
total order for modifications by <code>monotonic</code> operations on each
address. All modification orders must be compatible with the happens-before
order. There is no guarantee that the modification orders can be combined to
a global total order for the whole program (and this often will not be
possible). The read in an atomic read-modify-write operation
(<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
<a href="#i_atomicrmw"><code>atomicrmw</code></a>)
reads the value in the modification order immediately before the value it
writes. If one atomic read happens before another atomic read of the same
address, the later read must see the same value or a later value in the
address's modification order. This disallows reordering of
<code>monotonic</code> (or stronger) operations on the same address. If an
address is written <code>monotonic</code>ally by one thread, and other threads
<code>monotonic</code>ally read that address repeatedly, the other threads must
eventually see the write. This corresponds to the C++0x/C1x
<code>memory_order_relaxed</code>.</dd>
<dt><code>acquire</code></dt>
<dd>In addition to the guarantees of <code>monotonic</code>,
a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
<dt><code>release</code></dt>
<dd>In addition to the guarantees of <code>monotonic</code>, if this operation
writes a value which is subsequently read by an <code>acquire</code> operation,
it <i>synchronizes-with</i> that operation. (This isn't a complete
description; see the C++0x definition of a release sequence.) This corresponds
to the C++0x/C1x <code>memory_order_release</code>.</dd>
<dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
<code>acquire</code> and <code>release</code> operation on its address.
This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
<dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
<dd>In addition to the guarantees of <code>acq_rel</code>
(<code>acquire</code> for an operation which only reads, <code>release</code>
for an operation which only writes), there is a global total order on all
sequentially-consistent operations on all addresses, which is consistent with
the <i>happens-before</i> partial order and with the modification orders of
all the affected addresses. Each sequentially-consistent read sees the last
preceding write to the same address in this global order. This corresponds
to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
</dl>
<p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
it only <i>synchronizes with</i> or participates in modification and seq_cst
total orderings with other operations running in the same thread (for example,
in signal handlers).</p>
</div>
</div>
<!-- *********************************************************************** -->
<h2><a name="typesystem">Type System</a></h2>
<!-- *********************************************************************** -->
<div>
<p>The LLVM type system is one of the most important features of the
intermediate representation. Being typed enables a number of optimizations
to be performed on the intermediate representation directly, without having
to do extra analyses on the side before the transformation. A strong type
system makes it easier to read the generated code and enables novel analyses
and transformations that are not feasible to perform on normal three address
code representations.</p>
<!-- ======================================================================= -->
<h3>
<a name="t_classifications">Type Classifications</a>
</h3>
<div>
<p>The types fall into a few useful classifications:</p>
<table border="1" cellspacing="0" cellpadding="4">
<tbody>
<tr><th>Classification</th><th>Types</th></tr>
<tr>
<td><a href="#t_integer">integer</a></td>
<td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
</tr>
<tr>
<td><a href="#t_floating">floating point</a></td>
<td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
</tr>
<tr>
<td><a name="t_firstclass">first class</a></td>
<td><a href="#t_integer">integer</a>,
<a href="#t_floating">floating point</a>,
<a href="#t_pointer">pointer</a>,
<a href="#t_vector">vector</a>,
<a href="#t_struct">structure</a>,
<a href="#t_array">array</a>,
<a href="#t_label">label</a>,
<a href="#t_metadata">metadata</a>.
</td>
</tr>
<tr>
<td><a href="#t_primitive">primitive</a></td>
<td><a href="#t_label">label</a>,
<a href="#t_void">void</a>,
<a href="#t_integer">integer</a>,
<a href="#t_floating">floating point</a>,
<a href="#t_x86mmx">x86mmx</a>,
<a href="#t_metadata">metadata</a>.</td>
</tr>
<tr>
<td><a href="#t_derived">derived</a></td>
<td><a href="#t_array">array</a>,
<a href="#t_function">function</a>,
<a href="#t_pointer">pointer</a>,
<a href="#t_struct">structure</a>,
<a href="#t_vector">vector</a>,
<a href="#t_opaque">opaque</a>.
</td>
</tr>
</tbody>
</table>
<p>The <a href="#t_firstclass">first class</a> types are perhaps the most
important. Values of these types are the only ones which can be produced by
instructions.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="t_primitive">Primitive Types</a>
</h3>
<div>
<p>The primitive types are the fundamental building blocks of the LLVM
system.</p>
<!-- _______________________________________________________________________ -->
<h4>
<a name="t_integer">Integer Type</a>
</h4>
<div>
<h5>Overview:</h5>
<p>The integer type is a very simple type that simply specifies an arbitrary
bit width for the integer type desired. Any bit width from 1 bit to
2<sup>23</sup>-1 (about 8 million) can be specified.</p>
<h5>Syntax:</h5>
<pre>
iN
</pre>
<p>The number of bits the integer will occupy is specified by the <tt>N</tt>
value.</p>
<h5>Examples:</h5>
<table class="layout">
<tr class="layout">
<td class="left"><tt>i1</tt></td>
<td class="left">a single-bit integer.</td>
</tr>
<tr class="layout">
<td class="left"><tt>i32</tt></td>
<td class="left">a 32-bit integer.</td>
</tr>
<tr class="layout">
<td class="left"><tt>i1942652</tt></td>
<td class="left">a really big integer of over 1 million bits.</td>
</tr>
</table>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="t_floating">Floating Point Types</a>
</h4>
<div>
<table>
<tbody>
<tr><th>Type</th><th>Description</th></tr>
<tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
<tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
<tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
<tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
<tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
<tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
</tbody>
</table>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="t_x86mmx">X86mmx Type</a>
</h4>
<div>
<h5>Overview:</h5>
<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>
<h5>Syntax:</h5>
<pre>
x86mmx
</pre>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="t_void">Void Type</a>
</h4>
<div>
<h5>Overview:</h5>
<p>The void type does not represent any value and has no size.</p>
<h5>Syntax:</h5>
<pre>
void
</pre>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="t_label">Label Type</a>
</h4>
<div>
<h5>Overview:</h5>
<p>The label type represents code labels.</p>
<h5>Syntax:</h5>
<pre>
label
</pre>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="t_metadata">Metadata Type</a>
</h4>
<div>
<h5>Overview:</h5>
<p>The metadata type represents embedded metadata. No derived types may be
created from metadata except for <a href="#t_function">function</a>
arguments.
<h5>Syntax:</h5>
<pre>
metadata
</pre>
</div>
</div>
<!-- ======================================================================= -->
<h3>
<a name="t_derived">Derived Types</a>
</h3>
<div>
<p>The real power in LLVM comes from the derived types in the system. This is
what allows a programmer to represent arrays, functions, pointers, and other
useful types. Each of these types contain one or more element types which
may be a primitive type, or another derived type. For example, it is
possible to have a two dimensional array, using an array as the element type
of another array.</p>
<!-- _______________________________________________________________________ -->
<h4>
<a name="t_aggregate">Aggregate Types</a>
</h4>
<div>
<p>Aggregate Types are a subset of derived types that can contain multiple
member types. <a href="#t_array">Arrays</a> and
<a href="#t_struct">structs</a> are aggregate types.
<a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="t_array">Array Type</a>
</h4>
<div>
<h5>Overview:</h5>
<p>The array type is a very simple derived type that arranges elements
sequentially in memory. The array type requires a size (number of elements)
and an underlying data type.</p>
<h5>Syntax:</h5>
<pre>
[&lt;# elements&gt; x &lt;elementtype&gt;]
</pre>
<p>The number of elements is a constant integer value; <tt>elementtype</tt> may
be any type with a size.</p>
<h5>Examples:</h5>
<table class="layout">
<tr class="layout">
<td class="left"><tt>[40 x i32]</tt></td>
<td class="left">Array of 40 32-bit integer values.</td>
</tr>
<tr class="layout">
<td class="left"><tt>[41 x i32]</tt></td>
<td class="left">Array of 41 32-bit integer values.</td>
</tr>
<tr class="layout">
<td class="left"><tt>[4 x i8]</tt></td>
<td class="left">Array of 4 8-bit integer values.</td>
</tr>
</table>
<p>Here are some examples of multidimensional arrays:</p>
<table class="layout">
<tr class="layout">
<td class="left"><tt>[3 x [4 x i32]]</tt></td>
<td class="left">3x4 array of 32-bit integer values.</td>
</tr>
<tr class="layout">
<td class="left"><tt>[12 x [10 x float]]</tt></td>
<td class="left">12x10 array of single precision floating point values.</td>
</tr>
<tr class="layout">
<td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
<td class="left">2x3x4 array of 16-bit integer values.</td>
</tr>
</table>
<p>There is no restriction on indexing beyond the end of the array implied by
a static type (though there are restrictions on indexing beyond the bounds
of an allocated object in some cases). This means that single-dimension
'variable sized array' addressing can be implemented in LLVM with a zero
length array type. An implementation of 'pascal style arrays' in LLVM could
use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="t_function">Function Type</a>
</h4>
<div>
<h5>Overview:</h5>
<p>The function type can be thought of as a function signature. It consists of
a return type and a list of formal parameter types. The return type of a
function type is a first class type or a void type.</p>
<h5>Syntax:</h5>
<pre>
&lt;returntype&gt; (&lt;parameter list&gt;)
</pre>
<p>...where '<tt>&lt;parameter list&gt;</tt>' is a comma-separated list of type
specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
which indicates that the function takes a variable number of arguments.
Variable argument functions can access their arguments with
the <a href="#int_varargs">variable argument handling intrinsic</a>
functions. '<tt>&lt;returntype&gt;</tt>' is any type except
<a href="#t_label">label</a>.</p>
<h5>Examples:</h5>
<table class="layout">
<tr class="layout">
<td class="left"><tt>i32 (i32)</tt></td>
<td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
</td>
</tr><tr class="layout">
<td class="left"><tt>float&nbsp;(i16,&nbsp;i32&nbsp;*)&nbsp;*
</tt></td>
<td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
returning <tt>float</tt>.
</td>
</tr><tr class="layout">
<td class="left"><tt>i32 (i8*, ...)</tt></td>
<td class="left">A vararg function that takes at least one
<a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
which returns an integer. This is the signature for <tt>printf</tt> in
LLVM.
</td>
</tr><tr class="layout">
<td class="left"><tt>{i32, i32} (i32)</tt></td>
<td class="left">A function taking an <tt>i32</tt>, returning a
<a href="#t_struct">structure</a> containing two <tt>i32</tt> values
</td>
</tr>
</table>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="t_struct">Structure Type</a>
</h4>
<div>
<h5>Overview:</h5>
<p>The structure type is used to represent a collection of data members together
in memory. The elements of a structure may be any type that has a size.</p>
<p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
Structures in registers are accessed using the
'<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
'<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
<p>Structures may optionally be "packed" structures, which indicate that the
alignment of the struct is one byte, and that there is no padding between
the elements. In non-packed structs, padding between field types is inserted
as defined by the DataLayout string in the module, which is required to match
what the underlying code generator expects.</p>
<p>Structures can either be "literal" or "identified". A literal structure is
defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
types are always defined at the top level with a name. Literal types are
uniqued by their contents and can never be recursive or opaque since there is
no way to write one. Identified types can be recursive, can be opaqued, and are
never uniqued.
</p>
<h5>Syntax:</h5>
<pre>
%T1 = type { &lt;type list&gt; } <i>; Identified normal struct type</i>
%T2 = type &lt;{ &lt;type list&gt; }&gt; <i>; Identified packed struct type</i>
</pre>
<h5>Examples:</h5>
<table class="layout">
<tr class="layout">
<td class="left"><tt>{ i32, i32, i32 }</tt></td>
<td class="left">A triple of three <tt>i32</tt> values</td>
</tr>
<tr class="layout">
<td class="left"><tt>{&nbsp;float,&nbsp;i32&nbsp;(i32)&nbsp;*&nbsp;}</tt></td>
<td class="left">A pair, where the first element is a <tt>float</tt> and the
second element is a <a href="#t_pointer">pointer</a> to a
<a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
an <tt>i32</tt>.</td>
</tr>
<tr class="layout">
<td class="left"><tt>&lt;{ i8, i32 }&gt;</tt></td>
<td class="left">A packed struct known to be 5 bytes in size.</td>
</tr>
</table>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="t_opaque">Opaque Structure Types</a>
</h4>
<div>
<h5>Overview:</h5>
<p>Opaque structure types are used to represent named structure types that do
not have a body specified. This corresponds (for example) to the C notion of
a forward declared structure.</p>
<h5>Syntax:</h5>
<pre>
%X = type opaque
%52 = type opaque
</pre>
<h5>Examples:</h5>
<table class="layout">
<tr class="layout">
<td class="left"><tt>opaque</tt></td>
<td class="left">An opaque type.</td>
</tr>
</table>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="t_pointer">Pointer Type</a>
</h4>
<div>
<h5>Overview:</h5>
<p>The pointer type is used to specify memory locations.
Pointers are commonly used to reference objects in memory.</p>
<p>Pointer types may have an optional address space attribute defining the
numbered address space where the pointed-to object resides. The default
address space is number zero. The semantics of non-zero address
spaces are target-specific.</p>
<p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
<h5>Syntax:</h5>
<pre>
&lt;type&gt; *
</pre>
<h5>Examples:</h5>
<table class="layout">
<tr class="layout">
<td class="left"><tt>[4 x i32]*</tt></td>
<td class="left">A <a href="#t_pointer">pointer</a> to <a
href="#t_array">array</a> of four <tt>i32</tt> values.</td>
</tr>
<tr class="layout">
<td class="left"><tt>i32 (i32*) *</tt></td>
<td class="left"> A <a href="#t_pointer">pointer</a> to a <a
href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
<tt>i32</tt>.</td>
</tr>
<tr class="layout">
<td class="left"><tt>i32 addrspace(5)*</tt></td>
<td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
that resides in address space #5.</td>
</tr>
</table>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="t_vector">Vector Type</a>
</h4>
<div>
<h5>Overview:</h5>
<p>A vector type is a simple derived type that represents a vector of elements.
Vector types are used when multiple primitive data are operated in parallel
using a single instruction (SIMD). A vector type requires a size (number of
elements) and an underlying primitive data type. Vector types are considered
<a href="#t_firstclass">first class</a>.</p>
<h5>Syntax:</h5>
<pre>
&lt; &lt;# elements&gt; x &lt;elementtype&gt; &gt;
</pre>
<p>The number of elements is a constant integer value larger than 0; elementtype
may be any integer or floating point type, or a pointer to these types.
Vectors of size zero are not allowed. </p>
<h5>Examples:</h5>
<table class="layout">
<tr class="layout">
<td class="left"><tt>&lt;4 x i32&gt;</tt></td>
<td class="left">Vector of 4 32-bit integer values.</td>
</tr>
<tr class="layout">
<td class="left"><tt>&lt;8 x float&gt;</tt></td>
<td class="left">Vector of 8 32-bit floating-point values.</td>
</tr>
<tr class="layout">
<td class="left"><tt>&lt;2 x i64&gt;</tt></td>
<td class="left">Vector of 2 64-bit integer values.</td>
</tr>
<tr class="layout">
<td class="left"><tt>&lt;4 x i64*&gt;</tt></td>
<td class="left">Vector of 4 pointers to 64-bit integer values.</td>
</tr>
</table>
</div>
</div>
</div>
<!-- *********************************************************************** -->
<h2><a name="constants">Constants</a></h2>
<!-- *********************************************************************** -->
<div>
<p>LLVM has several different basic types of constants. This section describes
them all and their syntax.</p>
<!-- ======================================================================= -->
<h3>
<a name="simpleconstants">Simple Constants</a>
</h3>
<div>
<dl>
<dt><b>Boolean constants</b></dt>
<dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
<dt><b>Integer constants</b></dt>
<dd>Standard integers (such as '4') are constants of
the <a href="#t_integer">integer</a> type. Negative numbers may be used
with integer types.</dd>
<dt><b>Floating point constants</b></dt>
<dd>Floating point constants use standard decimal notation (e.g. 123.421),
exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
notation (see below). The assembler requires the exact decimal value of a
floating-point constant. For example, the assembler accepts 1.25 but
rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
constants must have a <a href="#t_floating">floating point</a> type. </dd>
<dt><b>Null pointer constants</b></dt>
<dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
and must be of <a href="#t_pointer">pointer type</a>.</dd>
</dl>
<p>The one non-intuitive notation for constants is the hexadecimal form of
floating point constants. For example, the form '<tt>double
0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
'<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
constants are required (and the only time that they are generated by the
disassembler) is when a floating point constant must be emitted but it cannot
be represented as a decimal floating point number in a reasonable number of
digits. For example, NaN's, infinities, and other special values are
represented in their IEEE hexadecimal format so that assembly and disassembly
do not cause any bits to change in the constants.</p>
<p>When using the hexadecimal form, constants of types half, float, and double are
represented using the 16-digit form shown above (which matches the IEEE754
representation for double); half and float values must, however, be exactly
representable as IEE754 half and single precision, respectively.
Hexadecimal format is always used
for long double, and there are three forms of long double. The 80-bit format
used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
The 128-bit format used by PowerPC (two adjacent doubles) is represented
by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
currently supported target uses this format. Long doubles will only work if
they match the long double format on your target. The IEEE 16-bit format
(half precision) is represented by <tt>0xH</tt> followed by 4 hexadecimal
digits. All hexadecimal formats are big-endian (sign bit at the left).</p>
<p>There are no constants of type x86mmx.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="aggregateconstants"></a> <!-- old anchor -->
<a name="complexconstants">Complex Constants</a>
</h3>
<div>
<p>Complex constants are a (potentially recursive) combination of simple
constants and smaller complex constants.</p>
<dl>
<dt><b>Structure constants</b></dt>
<dd>Structure constants are represented with notation similar to structure
type definitions (a comma separated list of elements, surrounded by braces
(<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
Structure constants must have <a href="#t_struct">structure type</a>, and
the number and types of elements must match those specified by the
type.</dd>
<dt><b>Array constants</b></dt>
<dd>Array constants are represented with notation similar to array type
definitions (a comma separated list of elements, surrounded by square
brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
]</tt>". Array constants must have <a href="#t_array">array type</a>, and
the number and types of elements must match those specified by the
type.</dd>
<dt><b>Vector constants</b></dt>
<dd>Vector constants are represented with notation similar to vector type
definitions (a comma separated list of elements, surrounded by
less-than/greater-than's (<tt>&lt;&gt;</tt>)). For example: "<tt>&lt; i32
42, i32 11, i32 74, i32 100 &gt;</tt>". Vector constants must
have <a href="#t_vector">vector type</a>, and the number and types of
elements must match those specified by the type.</dd>
<dt><b>Zero initialization</b></dt>
<dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
value to zero of <em>any</em> type, including scalar and
<a href="#t_aggregate">aggregate</a> types.
This is often used to avoid having to print large zero initializers
(e.g. for large arrays) and is always exactly equivalent to using explicit
zero initializers.</dd>
<dt><b>Metadata node</b></dt>
<dd>A metadata node is a structure-like constant with
<a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
be interpreted as part of the instruction stream, metadata is a place to
attach additional information such as debug info.</dd>
</dl>
</div>
<!-- ======================================================================= -->
<h3>
<a name="globalconstants">Global Variable and Function Addresses</a>
</h3>
<div>
<p>The addresses of <a href="#globalvars">global variables</a>
and <a href="#functionstructure">functions</a> are always implicitly valid
(link-time) constants. These constants are explicitly referenced when
the <a href="#identifiers">identifier for the global</a> is used and always
have <a href="#t_pointer">pointer</a> type. For example, the following is a
legal LLVM file:</p>
<pre class="doc_code">
@X = global i32 17
@Y = global i32 42
@Z = global [2 x i32*] [ i32* @X, i32* @Y ]
</pre>
</div>
<!-- ======================================================================= -->
<h3>
<a name="undefvalues">Undefined Values</a>
</h3>
<div>
<p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
indicates that the user of the value may receive an unspecified bit-pattern.
Undefined values may be of any type (other than '<tt>label</tt>'
or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
<p>Undefined values are useful because they indicate to the compiler that the
program is well defined no matter what value is used. This gives the
compiler more freedom to optimize. Here are some examples of (potentially
surprising) transformations that are valid (in pseudo IR):</p>
<pre class="doc_code">
%A = add %X, undef
%B = sub %X, undef
%C = xor %X, undef
Safe:
%A = undef
%B = undef
%C = undef
</pre>
<p>This is safe because all of the output bits are affected by the undef bits.
Any output bit can have a zero or one depending on the input bits.</p>
<pre class="doc_code">
%A = or %X, undef
%B = and %X, undef
Safe:
%A = -1
%B = 0
Unsafe:
%A = undef
%B = undef
</pre>
<p>These logical operations have bits that are not always affected by the input.
For example, if <tt>%X</tt> has a zero bit, then the output of the
'<tt>and</tt>' operation will always be a zero for that bit, no matter what
the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
set, allowing the '<tt>or</tt>' to be folded to -1.</p>
<pre class="doc_code">
%A = select undef, %X, %Y
%B = select undef, 42, %Y
%C = select %X, %Y, undef
Safe:
%A = %X (or %Y)
%B = 42 (or %Y)
%C = %Y
Unsafe:
%A = undef
%B = undef
%C = undef
</pre>
<p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
branch) conditions can go <em>either way</em>, but they have to come from one
of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
<tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
have to have a cleared low bit. However, in the <tt>%C</tt> example, the
optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
eliminated.</p>
<pre class="doc_code">
%A = xor undef, undef
%B = undef
%C = xor %B, %B
%D = undef
%E = icmp lt %D, 4
%F = icmp gte %D, 4
Safe:
%A = undef
%B = undef
%C = undef
%D = undef
%E = undef
%F = undef
</pre>
<p>This example points out that two '<tt>undef</tt>' operands are not
necessarily the same. This can be surprising to people (and also matches C
semantics) where they assume that "<tt>X^X</tt>" is always zero, even
if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
its value over its "live range". This is true because the variable doesn't
actually <em>have a live range</em>. Instead, the value is logically read
from arbitrary registers that happen to be around when needed, so the value
is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
need to have the same semantics or the core LLVM "replace all uses with"
concept would not hold.</p>
<pre class="doc_code">
%A = fdiv undef, %X
%B = fdiv %X, undef
Safe:
%A = undef
b: unreachable
</pre>
<p>These examples show the crucial difference between an <em>undefined
value</em> and <em>undefined behavior</em>. An undefined value (like
'<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
defined on SNaN's. However, in the second example, we can make a more
aggressive assumption: because the <tt>undef</tt> is allowed to be an
arbitrary value, we are allowed to assume that it could be zero. Since a
divide by zero has <em>undefined behavior</em>, we are allowed to assume that
the operation does not execute at all. This allows us to delete the divide and
all code after it. Because the undefined operation "can't happen", the
optimizer can assume that it occurs in dead code.</p>
<pre class="doc_code">
a: store undef -> %X
b: store %X -> undef
Safe:
a: &lt;deleted&gt;
b: unreachable
</pre>
<p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
undefined value can be assumed to not have any effect; we can assume that the
value is overwritten with bits that happen to match what was already there.
However, a store <em>to</em> an undefined location could clobber arbitrary
memory, therefore, it has undefined behavior.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="poisonvalues">Poison Values</a>
</h3>
<div>
<p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
they also represent the fact that an instruction or constant expression which
cannot evoke side effects has nevertheless detected a condition which results
in undefined behavior.</p>
<p>There is currently no way of representing a poison value in the IR; they
only exist when produced by operations such as
<a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
<p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
<ul>
<li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
their operands.</li>
<li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
to their dynamic predecessor basic block.</li>
<li>Function arguments depend on the corresponding actual argument values in
the dynamic callers of their functions.</li>
<li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
<a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
control back to them.</li>
<li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
<a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>,
or exception-throwing call instructions that dynamically transfer control
back to them.</li>
<li>Non-volatile loads and stores depend on the most recent stores to all of the
referenced memory addresses, following the order in the IR
(including loads and stores implied by intrinsics such as
<a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
<!-- TODO: In the case of multiple threads, this only applies if the store
"happens-before" the load or store. -->
<!-- TODO: floating-point exception state -->
<li>An instruction with externally visible side effects depends on the most
recent preceding instruction with externally visible side effects, following
the order in the IR. (This includes
<a href="#volatile">volatile operations</a>.)</li>
<li>An instruction <i>control-depends</i> on a
<a href="#terminators">terminator instruction</a>
if the terminator instruction has multiple successors and the instruction
is always executed when control transfers to one of the successors, and
may not be executed when control is transferred to another.</li>
<li>Additionally, an instruction also <i>control-depends</i> on a terminator
instruction if the set of instructions it otherwise depends on would be
different if the terminator had transferred control to a different
successor.</li>
<li>Dependence is transitive.</li>
</ul>
<p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
with the additional affect that any instruction which has a <i>dependence</i>
on a poison value has undefined behavior.</p>
<p>Here are some examples:</p>
<pre class="doc_code">
entry:
%poison = sub nuw i32 0, 1 ; Results in a poison value.
%still_poison = and i32 %poison, 0 ; 0, but also poison.
%poison_yet_again = getelementptr i32* @h, i32 %still_poison
store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
store i32 %poison, i32* @g ; Poison value stored to memory.
%poison2 = load i32* @g ; Poison value loaded back from memory.
store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
%narrowaddr = bitcast i32* @g to i16*
%wideaddr = bitcast i32* @g to i64*
%poison3 = load i16* %narrowaddr ; Returns a poison value.
%poison4 = load i64* %wideaddr ; Returns a poison value.
%cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
br i1 %cmp, label %true, label %end ; Branch to either destination.
true:
store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
; it has undefined behavior.
br label %end
end:
%p = phi i32 [ 0, %entry ], [ 1, %true ]
; Both edges into this PHI are
; control-dependent on %cmp, so this
; always results in a poison value.
store volatile i32 0, i32* @g ; This would depend on the store in %true
; if %cmp is true, or the store in %entry
; otherwise, so this is undefined behavior.
br i1 %cmp, label %second_true, label %second_end
; The same branch again, but this time the
; true block doesn't have side effects.
second_true:
; No side effects!
ret void
second_end:
store volatile i32 0, i32* @g ; This time, the instruction always depends
; on the store in %end. Also, it is
; control-equivalent to %end, so this is
; well-defined (ignoring earlier undefined
; behavior in this example).
</pre>
</div>
<!-- ======================================================================= -->
<h3>
<a name="blockaddress">Addresses of Basic Blocks</a>
</h3>
<div>
<p><b><tt>blockaddress(@function, %block)</tt></b></p>
<p>The '<tt>blockaddress</tt>' constant computes the address of the specified
basic block in the specified function, and always has an i8* type. Taking
the address of the entry block is illegal.</p>
<p>This value only has defined behavior when used as an operand to the
'<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
comparisons against null. Pointer equality tests between labels addresses
results in undefined behavior &mdash; though, again, comparison against null
is ok, and no label is equal to the null pointer. This may be passed around
as an opaque pointer sized value as long as the bits are not inspected. This
allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
long as the original value is reconstituted before the <tt>indirectbr</tt>
instruction.</p>
<p>Finally, some targets may provide defined semantics when using the value as
the operand to an inline assembly, but that is target specific.</p>
</div>
<!-- ======================================================================= -->
<h3>
<a name="constantexprs">Constant Expressions</a>
</h3>
<div>
<p>Constant expressions are used to allow expressions involving other constants
to be used as constants. Constant expressions may be of
any <a href="#t_firstclass">first class</a> type and may involve any LLVM
operation that does not have side effects (e.g. load and call are not
supported). The following is the syntax for constant expressions:</p>
<dl>
<dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
<dd>Truncate a constant to another type. The bit size of CST must be larger
than the bit size of TYPE. Both types must be integers.</dd>
<dt><b><tt>zext (CST to TYPE)</tt></b></dt>
<dd>Zero extend a constant to another type. The bit size of CST must be
smaller than the bit size of TYPE. Both types must be integers.</dd>
<dt><b><tt>sext (CST to TYPE)</tt></b></dt>
<dd>Sign extend a constant to another type. The bit size of CST must be
smaller than the bit size of TYPE. Both types must be integers.</dd>
<dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
<dd>Truncate a floating point constant to another floating point type. The
size of CST must be larger than the size of TYPE. Both types must be
floating point.</dd>
<dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
<dd>Floating point extend a constant to another type. The size of CST must be
smaller or equal to the size of TYPE. Both types must be floating
point.</dd>
<dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
<dd>Convert a floating point constant to the corresponding unsigned integer
constant. TYPE must be a scalar or vector integer type. CST must be of
scalar or vector floating point type. Both CST and TYPE must be scalars,
or vectors of the same number of elements. If the value won't fit in the
integer type, the results are undefined.</dd>
<dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
<dd>Convert a floating point constant to the corresponding signed integer
constant. TYPE must be a scalar or vector integer type. CST must be of
scalar or vector floating point type. Both CST and TYPE must be scalars,
or vectors of the same number of elements. If the value won't fit in the
integer type, the results are undefined.</dd>
<dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
<dd>Convert an unsigned integer constant to the corresponding floating point
constant. TYPE must be a scalar or vector floating point type. CST must be
of scalar or vector integer type. Both CST and TYPE must be scalars, or
vectors of the same number of elements. If the value won't fit in the
floating point type, the results are undefined.</dd>
<dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
<dd>Convert a signed integer constant to the corresponding floating point
constant. TYPE must be a scalar or vector floating point type. CST must be
of scalar or vector integer type. Both CST and TYPE must be scalars, or
vectors of the same number of elements. If the value won't fit in the
floating point type, the results are undefined.</dd>
<dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
<dd>Convert a pointer typed constant to the corresponding integer constant
<tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
make it fit in <tt>TYPE</tt>.</dd>
<dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
<dd>Convert an integer constant to a pointer constant. TYPE must be a pointer
type. CST must be of integer type. The CST value is zero extended,
truncated, or unchanged to make it fit in a pointer size. This one is
<i>really</i> dangerous!</dd>
<dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
<dd>Convert a constant, CST, to another TYPE. The constraints of the operands
are the same as those for the <a href="#i_bitcast">bitcast
instruction</a>.</dd>
<dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
<dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
<dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
constants. As with the <a href="#i_getelementptr">getelementptr</a>
instruction, the index list may have zero or more indexes, which are
required to make sense for the type of "CSTPTR".</dd>
<dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
<dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
<dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
<dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
<dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
<dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
<dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
<dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
constants.</dd>
<dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
<dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
constants.</dd>
<dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
<dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
constants.</dd>
<dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
<dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
constants. The index list is interpreted in a similar manner as indices in
a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
index value must be specified.</dd>
<dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
<dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
constants. The index list is interpreted in a similar manner as indices in
a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
index value must be specified.</dd>
<dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
<dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
be any of the <a href="#binaryops">binary</a>
or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
on operands are the same as those for the corresponding instruction
(e.g. no bitwise operations on floating point values are allowed).</dd>
</dl>
</div>
</div>
<!-- *********************************************************************** -->
<h2><a name="othervalues">Other Values</a></h2>
<!-- *********************************************************************** -->
<div>
<!-- ======================================================================= -->
<h3>
<a name="inlineasm">Inline Assembler Expressions</a>
</h3>
<div>
<p>LLVM supports inline assembler expressions (as opposed
to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
a special value. This value represents the inline assembler as a string
(containing the instructions to emit), a list of operand constraints (stored
as a string), a flag that indicates whether or not the inline asm
expression has side effects, and a flag indicating whether the function
containing the asm needs to align its stack conservatively. An example
inline assembler expression is:</p>
<pre class="doc_code">
i32 (i32) asm "bswap $0", "=r,r"
</pre>
<p>Inline assembler expressions may <b>only</b> be used as the callee operand of
a <a href="#i_call"><tt>call</tt></a> or an
<a href="#i_invoke"><tt>invoke</tt></a> instruction.
Thus, typically we have:</p>
<pre class="doc_code">
%X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
</pre>
<p>Inline asms with side effects not visible in the constraint list must be
marked as having side effects. This is done through the use of the
'<tt>sideeffect</tt>' keyword, like so:</p>
<pre class="doc_code">
call void asm sideeffect "eieio", ""()
</pre>
<p>In some cases inline asms will contain code that will not work unless the
stack is aligned in some way, such as calls or SSE instructions on x86,
yet will not contain code that does that alignment within the asm.
The compiler should make conservative assumptions about what the asm might
contain and should generate its usual stack alignment code in the prologue
if the '<tt>alignstack</tt>' keyword is present:</p>
<pre class="doc_code">
call void asm alignstack "eieio", ""()
</pre>
<p>Inline asms also support using non-standard assembly dialects. The assumed
dialect is ATT. When the '<tt>inteldialect</tt>' keyword is present, the
inline asm is using the Intel dialect. Currently, ATT and Intel are the
only supported dialects. An example is:</p>
<pre class="doc_code">
call void asm inteldialect "eieio", ""()
</pre>
<p>If multiple keywords appear the '<tt>sideeffect</tt>' keyword must come
first, the '<tt>alignstack</tt>' keyword second and the
'<tt>inteldialect</tt>' keyword last.</p>
<!--
<p>TODO: The format of the asm and constraints string still need to be
documented here. Constraints on what can be done (e.g. duplication, moving,
etc need to be documented). This is probably best done by reference to
another document that covers inline asm from a holistic perspective.</p>
-->
<!-- _______________________________________________________________________ -->
<h4>
<a name="inlineasm_md">Inline Asm Metadata</a>
</h4>
<div>
<p>The call instructions that wrap inline asm nodes may have a
"<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
integers. If present, the code generator will use the integer as the
location cookie value when report errors through the <tt>LLVMContext</tt>
error reporting mechanisms. This allows a front-end to correlate backend
errors that occur with inline asm back to the source code that produced it.
For example:</p>
<pre class="doc_code">
call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
...
!42 = !{ i32 1234567 }
</pre>
<p>It is up to the front-end to make sense of the magic numbers it places in the
IR. If the MDNode contains multiple constants, the code generator will use
the one that corresponds to the line of the asm that the error occurs on.</p>
</div>
</div>
<!-- ======================================================================= -->
<h3>
<a name="metadata">Metadata Nodes and Metadata Strings</a>
</h3>
<div>
<p>LLVM IR allows metadata to be attached to instructions in the program that
can convey extra information about the code to the optimizers and code
generator. One example application of metadata is source-level debug
information. There are two metadata primitives: strings and nodes. All
metadata has the <tt>metadata</tt> type and is identified in syntax by a
preceding exclamation point ('<tt>!</tt>').</p>
<p>A metadata string is a string surrounded by double quotes. It can contain
any character by escaping non-printable characters with "<tt>\xx</tt>" where
"<tt>xx</tt>" is the two digit hex code. For example:
"<tt>!"test\00"</tt>".</p>
<p>Metadata nodes are represented with notation similar to structure constants
(a comma separated list of elements, surrounded by braces and preceded by an
exclamation point). Metadata nodes can have any values as their operand. For
example:</p>
<div class="doc_code">
<pre>
!{ metadata !"test\00", i32 10}
</pre>
</div>
<p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
metadata nodes, which can be looked up in the module symbol table. For
example:</p>
<div class="doc_code">
<pre>
!foo = metadata !{!4, !3}
</pre>
</div>
<p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
function is using two metadata arguments:</p>
<div class="doc_code">
<pre>
call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
</pre>
</div>
<p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
identifier:</p>
<div class="doc_code">
<pre>
%indvar.next = add i64 %indvar, 1, !dbg !21
</pre>
</div>
<p>More information about specific metadata nodes recognized by the optimizers
and code generator is found below.</p>
<!-- _______________________________________________________________________ -->
<h4>
<a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
</h4>
<div>
<p>In LLVM IR, memory does not have types, so LLVM's own type system is not
suitable for doing TBAA. Instead, metadata is added to the IR to describe
a type system of a higher level language. This can be used to implement
typical C/C++ TBAA, but it can also be used to implement custom alias
analysis behavior for other languages.</p>
<p>The current metadata format is very simple. TBAA metadata nodes have up to
three fields, e.g.:</p>
<div class="doc_code">
<pre>
!0 = metadata !{ metadata !"an example type tree" }
!1 = metadata !{ metadata !"int", metadata !0 }
!2 = metadata !{ metadata !"float", metadata !0 }
!3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
</pre>
</div>
<p>The first field is an identity field. It can be any value, usually
a metadata string, which uniquely identifies the type. The most important
name in the tree is the name of the root node. Two trees with
different root node names are entirely disjoint, even if they
have leaves with common names.</p>
<p>The second field identifies the type's parent node in the tree, or
is null or omitted for a root node. A type is considered to alias
all of its descendants and all of its ancestors in the tree. Also,
a type is considered to alias all types in other trees, so that
bitcode produced from multiple front-ends is handled conservatively.</p>
<p>If the third field is present, it's an integer which if equal to 1
indicates that the type is "constant" (meaning
<tt>pointsToConstantMemory</tt> should return true; see
<a href="AliasAnalysis.html#OtherItfs">other useful
<tt>AliasAnalysis</tt> methods</a>).</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="tbaa.struct">'<tt>tbaa.struct</tt>' Metadata</a>
</h4>
<div>
<p>The <a href="#int_memcpy"><tt>llvm.memcpy</tt></a> is often used to implement
aggregate assignment operations in C and similar languages, however it is
defined to copy a contiguous region of memory, which is more than strictly
necessary for aggregate types which contain holes due to padding. Also, it
doesn't contain any TBAA information about the fields of the aggregate.</p>
<p><tt>!tbaa.struct</tt> metadata can describe which memory subregions in a memcpy
are padding and what the TBAA tags of the struct are.</p>
<p>The current metadata format is very simple. <tt>!tbaa.struct</tt> metadata nodes
are a list of operands which are in conceptual groups of three. For each
group of three, the first operand gives the byte offset of a field in bytes,
the second gives its size in bytes, and the third gives its
tbaa tag. e.g.:</p>
<div class="doc_code">
<pre>
!4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
</pre>
</div>
<p>This describes a struct with two fields. The first is at offset 0 bytes
with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
and has size 4 bytes and has tbaa tag !2.</p>
<p>Note that the fields need not be contiguous. In this example, there is a
4 byte gap between the two fields. This gap represents padding which
does not carry useful data and need not be preserved.</p>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="fpmath">'<tt>fpmath</tt>' Metadata</a>
</h4>
<div>
<p><tt>fpmath</tt> metadata may be attached to any instruction of floating point
type. It can be used to express the maximum acceptable error in the result of
that instruction, in ULPs, thus potentially allowing the compiler to use a
more efficient but less accurate method of computing it. ULP is defined as
follows:</p>
<blockquote>
<p>If <tt>x</tt> is a real number that lies between two finite consecutive
floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
distance between the two non-equal finite floating-point numbers nearest
<tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
</blockquote>
<p>The metadata node shall consist of a single positive floating point number
representing the maximum relative error, for example:</p>
<div class="doc_code">
<pre>
!0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
</pre>
</div>
</div>
<!-- _______________________________________________________________________ -->
<h4>
<a name="range">'<tt>range</tt>' Metadata</a>
</h4>
<div>
<p><tt>range</tt> metadata may be attached only to loads of integer types. It
expresses the possible ranges the loaded value is in. The ranges are
represented with a flattened list of integers. The loaded value is known to
be in the union of the ranges defined by each consecutive pair. Each pair
has the following properties:</p>
<ul>
<li>The type must match the type loaded by the instruction.</li>
<li>The pair <tt>a,b</tt> represents the range <tt>[a,b)</tt>.</li>
<li>Both <tt>a</tt> and <tt>b</tt> are constants.</li>
<li>The range is allowed to wrap.</li>
<li>The range should not represent the full or empty set. That is,
<tt>a!=b</tt>. </li>
</ul>
<p> In addition, the pairs must be in signed order of the lower bound and
they must be non-contiguous.</p>
<p>Examples:</p>
<div class="doc_code">
<pre>
%a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
%b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
%c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
%d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
...
!0 = metadata !{ i8 0, i8 2 }
!1 = metadata !{ i8 255, i8 2 }
!2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
!3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
</pre>
</div>
</div>
</div>
</div>
<!-- *********************************************************************** -->
<h2>
<a name="module_flags">Module Flags Metadata</a>
</h2>
<!-- *********************************************************************** -->
<div>
<p>Information about the module as a whole is difficult to convey to LLVM's
subsystems. The LLVM IR isn't sufficient to transmit this
information. The <tt>llvm.module.flags</tt> named metadata exists in order to
facilitate this. These flags are in the form of key / value pairs &mdash;
much like a dictionary &mdash; making it easy for any subsystem who cares
about a flag to look it up.</p>
<p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata
triplets. Each triplet has the following form:</p>
<ul>
<li>The first element is a <i>behavior</i> flag, which specifies the behavior
when two (or more) modules are merged together, and it encounters two (or
more) metadata with the same ID. The supported behaviors are described
below.</li>
<li>The second element is a metadata string that is a unique ID for the
metadata. How each ID is interpreted is documented below.</li>
<li>The third element is the value of the flag.