blob: 3492a2797092ea4338e62da4d8563941a341cfc8 [file] [log] [blame]
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
"http://www.w3.org/TR/html4/strict.dtd">
<html>
<head>
<title>LLVM's Analysis and Transform Passes</title>
<link rel="stylesheet" href="llvm.css" type="text/css">
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
</head>
<body>
<!--
If Passes.html is up to date, the following "one-liner" should print
an empty diff.
egrep -e '^<tr><td><a href="#.*">-.*</a></td><td>.*</td></tr>$' \
-e '^ <a name=".*">.*</a>$' < Passes.html >html; \
perl >help <<'EOT' && diff -u help html; rm -f help html
open HTML, "<Passes.html" or die "open: Passes.html: $!\n";
while (<HTML>) {
m:^<tr><td><a href="#(.*)">-.*</a></td><td>.*</td></tr>$: or next;
$order{$1} = sprintf("%03d", 1 + int %order);
}
open HELP, "../Release/bin/opt -help|" or die "open: opt -help: $!\n";
while (<HELP>) {
m:^ -([^ ]+) +- (.*)$: or next;
my $o = $order{$1};
$o = "000" unless defined $o;
push @x, "$o<tr><td><a href=\"#$1\">-$1</a></td><td>$2</td></tr>\n";
push @y, "$o <a name=\"$1\">$2</a>\n";
}
@x = map { s/^\d\d\d//; $_ } sort @x;
@y = map { s/^\d\d\d//; $_ } sort @y;
print @x, @y;
EOT
This (real) one-liner can also be helpful when converting comments to HTML:
perl -e '$/ = undef; for (split(/\n/, <>)) { s:^ *///? ?::; print " <p>\n" if !$on && $_ =~ /\S/; print " </p>\n" if $on && $_ =~ /^\s*$/; print " $_\n"; $on = ($_ =~ /\S/); } print " </p>\n" if $on'
-->
<div class="doc_title">LLVM's Analysis and Transform Passes</div>
<ol>
<li><a href="#intro">Introduction</a></li>
<li><a href="#analyses">Analysis Passes</a>
<li><a href="#transforms">Transform Passes</a></li>
<li><a href="#utilities">Utility Passes</a></li>
</ol>
<div class="doc_author">
<p>Written by <a href="mailto:rspencer@x10sys.com">Reid Spencer</a>
and Gordon Henriksen</p>
</div>
<!-- ======================================================================= -->
<div class="doc_section"> <a name="intro">Introduction</a> </div>
<div class="doc_text">
<p>This document serves as a high level summary of the optimization features
that LLVM provides. Optimizations are implemented as Passes that traverse some
portion of a program to either collect information or transform the program.
The table below divides the passes that LLVM provides into three categories.
Analysis passes compute information that other passes can use or for debugging
or program visualization purposes. Transform passes can use (or invalidate)
the analysis passes. Transform passes all mutate the program in some way.
Utility passes provides some utility but don't otherwise fit categorization.
For example passes to extract functions to bitcode or write a module to
bitcode are neither analysis nor transform passes.
<p>The table below provides a quick summary of each pass and links to the more
complete pass description later in the document.</p>
</div>
<div class="doc_text" >
<table>
<tr><th colspan="2"><b>ANALYSIS PASSES</b></th></tr>
<tr><th>Option</th><th>Name</th></tr>
<tr><td><a href="#aa-eval">-aa-eval</a></td><td>Exhaustive Alias Analysis Precision Evaluator</td></tr>
<tr><td><a href="#anders-aa">-anders-aa</a></td><td>Andersen's Interprocedural Alias Analysis</td></tr>
<tr><td><a href="#basicaa">-basicaa</a></td><td>Basic Alias Analysis (default AA impl)</td></tr>
<tr><td><a href="#basiccg">-basiccg</a></td><td>Basic CallGraph Construction</td></tr>
<tr><td><a href="#basicvn">-basicvn</a></td><td>Basic Value Numbering (default GVN impl)</td></tr>
<tr><td><a href="#callgraph">-callgraph</a></td><td>Print a call graph</td></tr>
<tr><td><a href="#callscc">-callscc</a></td><td>Print SCCs of the Call Graph</td></tr>
<tr><td><a href="#cfgscc">-cfgscc</a></td><td>Print SCCs of each function CFG</td></tr>
<tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Optimize for code generation</td></tr>
<tr><td><a href="#count-aa">-count-aa</a></td><td>Count Alias Analysis Query Responses</td></tr>
<tr><td><a href="#debug-aa">-debug-aa</a></td><td>AA use debugger</td></tr>
<tr><td><a href="#domfrontier">-domfrontier</a></td><td>Dominance Frontier Construction</td></tr>
<tr><td><a href="#domtree">-domtree</a></td><td>Dominator Tree Construction</td></tr>
<tr><td><a href="#externalfnconstants">-externalfnconstants</a></td><td>Print external fn callsites passed constants</td></tr>
<tr><td><a href="#globalsmodref-aa">-globalsmodref-aa</a></td><td>Simple mod/ref analysis for globals</td></tr>
<tr><td><a href="#instcount">-instcount</a></td><td>Counts the various types of Instructions</td></tr>
<tr><td><a href="#intervals">-intervals</a></td><td>Interval Partition Construction</td></tr>
<tr><td><a href="#load-vn">-load-vn</a></td><td>Load Value Numbering</td></tr>
<tr><td><a href="#loops">-loops</a></td><td>Natural Loop Construction</td></tr>
<tr><td><a href="#memdep">-memdep</a></td><td>Memory Dependence Analysis</td></tr>
<tr><td><a href="#no-aa">-no-aa</a></td><td>No Alias Analysis (always returns 'may' alias)</td></tr>
<tr><td><a href="#no-profile">-no-profile</a></td><td>No Profile Information</td></tr>
<tr><td><a href="#postdomfrontier">-postdomfrontier</a></td><td>Post-Dominance Frontier Construction</td></tr>
<tr><td><a href="#postdomtree">-postdomtree</a></td><td>Post-Dominator Tree Construction</td></tr>
<tr><td><a href="#print">-print</a></td><td>Print function to stderr</td></tr>
<tr><td><a href="#print-alias-sets">-print-alias-sets</a></td><td>Alias Set Printer</td></tr>
<tr><td><a href="#print-callgraph">-print-callgraph</a></td><td>Print Call Graph to 'dot' file</td></tr>
<tr><td><a href="#print-cfg">-print-cfg</a></td><td>Print CFG of function to 'dot' file</td></tr>
<tr><td><a href="#print-cfg-only">-print-cfg-only</a></td><td>Print CFG of function to 'dot' file (with no function bodies)</td></tr>
<tr><td><a href="#printm">-printm</a></td><td>Print module to stderr</td></tr>
<tr><td><a href="#printusedtypes">-printusedtypes</a></td><td>Find Used Types</td></tr>
<tr><td><a href="#profile-loader">-profile-loader</a></td><td>Load profile information from llvmprof.out</td></tr>
<tr><td><a href="#scalar-evolution">-scalar-evolution</a></td><td>Scalar Evolution Analysis</td></tr>
<tr><td><a href="#targetdata">-targetdata</a></td><td>Target Data Layout</td></tr>
<tr><th colspan="2"><b>TRANSFORM PASSES</b></th></tr>
<tr><th>Option</th><th>Name</th></tr>
<tr><td><a href="#adce">-adce</a></td><td>Aggressive Dead Code Elimination</td></tr>
<tr><td><a href="#argpromotion">-argpromotion</a></td><td>Promote 'by reference' arguments to scalars</td></tr>
<tr><td><a href="#block-placement">-block-placement</a></td><td>Profile Guided Basic Block Placement</td></tr>
<tr><td><a href="#break-crit-edges">-break-crit-edges</a></td><td>Break critical edges in CFG</td></tr>
<tr><td><a href="#cee">-cee</a></td><td>Correlated Expression Elimination</td></tr>
<tr><td><a href="#condprop">-condprop</a></td><td>Conditional Propagation</td></tr>
<tr><td><a href="#constmerge">-constmerge</a></td><td>Merge Duplicate Global Constants</td></tr>
<tr><td><a href="#constprop">-constprop</a></td><td>Simple constant propagation</td></tr>
<tr><td><a href="#dce">-dce</a></td><td>Dead Code Elimination</td></tr>
<tr><td><a href="#deadargelim">-deadargelim</a></td><td>Dead Argument Elimination</td></tr>
<tr><td><a href="#deadtypeelim">-deadtypeelim</a></td><td>Dead Type Elimination</td></tr>
<tr><td><a href="#die">-die</a></td><td>Dead Instruction Elimination</td></tr>
<tr><td><a href="#dse">-dse</a></td><td>Dead Store Elimination</td></tr>
<tr><td><a href="#gcse">-gcse</a></td><td>Global Common Subexpression Elimination</td></tr>
<tr><td><a href="#globaldce">-globaldce</a></td><td>Dead Global Elimination</td></tr>
<tr><td><a href="#globalopt">-globalopt</a></td><td>Global Variable Optimizer</td></tr>
<tr><td><a href="#gvn">-gvn</a></td><td>Global Value Numbering</td></tr>
<tr><td><a href="#gvnpre">-gvnpre</a></td><td>Global Value Numbering/Partial Redundancy Elimination</td></tr>
<tr><td><a href="#indmemrem">-indmemrem</a></td><td>Indirect Malloc and Free Removal</td></tr>
<tr><td><a href="#indvars">-indvars</a></td><td>Canonicalize Induction Variables</td></tr>
<tr><td><a href="#inline">-inline</a></td><td>Function Integration/Inlining</td></tr>
<tr><td><a href="#insert-block-profiling">-insert-block-profiling</a></td><td>Insert instrumentation for block profiling</td></tr>
<tr><td><a href="#insert-edge-profiling">-insert-edge-profiling</a></td><td>Insert instrumentation for edge profiling</td></tr>
<tr><td><a href="#insert-function-profiling">-insert-function-profiling</a></td><td>Insert instrumentation for function profiling</td></tr>
<tr><td><a href="#insert-null-profiling-rs">-insert-null-profiling-rs</a></td><td>Measure profiling framework overhead</td></tr>
<tr><td><a href="#insert-rs-profiling-framework">-insert-rs-profiling-framework</a></td><td>Insert random sampling instrumentation framework</td></tr>
<tr><td><a href="#instcombine">-instcombine</a></td><td>Combine redundant instructions</td></tr>
<tr><td><a href="#internalize">-internalize</a></td><td>Internalize Global Symbols</td></tr>
<tr><td><a href="#ipconstprop">-ipconstprop</a></td><td>Interprocedural constant propagation</td></tr>
<tr><td><a href="#ipsccp">-ipsccp</a></td><td>Interprocedural Sparse Conditional Constant Propagation</td></tr>
<tr><td><a href="#lcssa">-lcssa</a></td><td>Loop-Closed SSA Form Pass</td></tr>
<tr><td><a href="#licm">-licm</a></td><td>Loop Invariant Code Motion</td></tr>
<tr><td><a href="#loop-extract">-loop-extract</a></td><td>Extract loops into new functions</td></tr>
<tr><td><a href="#loop-extract-single">-loop-extract-single</a></td><td>Extract at most one loop into a new function</td></tr>
<tr><td><a href="#loop-index-split">-loop-index-split</a></td><td>Index Split Loops</td></tr>
<tr><td><a href="#loop-reduce">-loop-reduce</a></td><td>Loop Strength Reduction</td></tr>
<tr><td><a href="#loop-rotate">-loop-rotate</a></td><td>Rotate Loops</td></tr>
<tr><td><a href="#loop-unroll">-loop-unroll</a></td><td>Unroll loops</td></tr>
<tr><td><a href="#loop-unswitch">-loop-unswitch</a></td><td>Unswitch loops</td></tr>
<tr><td><a href="#loopsimplify">-loopsimplify</a></td><td>Canonicalize natural loops</td></tr>
<tr><td><a href="#lower-packed">-lower-packed</a></td><td>lowers packed operations to operations on smaller packed datatypes</td></tr>
<tr><td><a href="#lowerallocs">-lowerallocs</a></td><td>Lower allocations from instructions to calls</td></tr>
<tr><td><a href="#lowergc">-lowergc</a></td><td>Lower GC intrinsics, for GCless code generators</td></tr>
<tr><td><a href="#lowerinvoke">-lowerinvoke</a></td><td>Lower invoke and unwind, for unwindless code generators</td></tr>
<tr><td><a href="#lowerselect">-lowerselect</a></td><td>Lower select instructions to branches</td></tr>
<tr><td><a href="#lowersetjmp">-lowersetjmp</a></td><td>Lower Set Jump</td></tr>
<tr><td><a href="#lowerswitch">-lowerswitch</a></td><td>Lower SwitchInst's to branches</td></tr>
<tr><td><a href="#mem2reg">-mem2reg</a></td><td>Promote Memory to Register</td></tr>
<tr><td><a href="#mergereturn">-mergereturn</a></td><td>Unify function exit nodes</td></tr>
<tr><td><a href="#predsimplify">-predsimplify</a></td><td>Predicate Simplifier</td></tr>
<tr><td><a href="#prune-eh">-prune-eh</a></td><td>Remove unused exception handling info</td></tr>
<tr><td><a href="#raiseallocs">-raiseallocs</a></td><td>Raise allocations from calls to instructions</td></tr>
<tr><td><a href="#reassociate">-reassociate</a></td><td>Reassociate expressions</td></tr>
<tr><td><a href="#reg2mem">-reg2mem</a></td><td>Demote all values to stack slots</td></tr>
<tr><td><a href="#scalarrepl">-scalarrepl</a></td><td>Scalar Replacement of Aggregates</td></tr>
<tr><td><a href="#sccp">-sccp</a></td><td>Sparse Conditional Constant Propagation</td></tr>
<tr><td><a href="#simplify-libcalls">-simplify-libcalls</a></td><td>Simplify well-known library calls</td></tr>
<tr><td><a href="#simplifycfg">-simplifycfg</a></td><td>Simplify the CFG</td></tr>
<tr><td><a href="#strip">-strip</a></td><td>Strip all symbols from a module</td></tr>
<tr><td><a href="#tailcallelim">-tailcallelim</a></td><td>Tail Call Elimination</td></tr>
<tr><td><a href="#tailduplicate">-tailduplicate</a></td><td>Tail Duplication</td></tr>
<tr><th colspan="2"><b>UTILITY PASSES</b></th></tr>
<tr><th>Option</th><th>Name</th></tr>
<tr><td><a href="#deadarghaX0r">-deadarghaX0r</a></td><td>Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</td></tr>
<tr><td><a href="#extract-blocks">-extract-blocks</a></td><td>Extract Basic Blocks From Module (for bugpoint use)</td></tr>
<tr><td><a href="#preverify">-preverify</a></td><td>Preliminary module verification</td></tr>
<tr><td><a href="#verify">-verify</a></td><td>Module Verifier</td></tr>
<tr><td><a href="#view-cfg">-view-cfg</a></td><td>View CFG of function</td></tr>
<tr><td><a href="#view-cfg-only">-view-cfg-only</a></td><td>View CFG of function (with no function bodies)</td></tr>
</table>
</div>
<!-- ======================================================================= -->
<div class="doc_section"> <a name="example">Analysis Passes</a></div>
<div class="doc_text">
<p>This section describes the LLVM Analysis Passes.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="aa-eval">Exhaustive Alias Analysis Precision Evaluator</a>
</div>
<div class="doc_text">
<p>This is a simple N^2 alias analysis accuracy evaluator.
Basically, for each function in the program, it simply queries to see how the
alias analysis implementation answers alias queries between each pair of
pointers in the function.</p>
<p>This is inspired and adapted from code by: Naveen Neelakantam, Francesco
Spadini, and Wojciech Stryjewski.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="anders-aa">Andersen's Interprocedural Alias Analysis</a>
</div>
<div class="doc_text">
<p>
This is an implementation of Andersen's interprocedural alias
analysis
</p>
<p>
In pointer analysis terms, this is a subset-based, flow-insensitive,
field-sensitive, and context-insensitive algorithm pointer algorithm.
</p>
<p>
This algorithm is implemented as three stages:
</p>
<ol>
<li>Object identification.</li>
<li>Inclusion constraint identification.</li>
<li>Offline constraint graph optimization.</li>
<li>Inclusion constraint solving.</li>
</ol>
<p>
The object identification stage identifies all of the memory objects in the
program, which includes globals, heap allocated objects, and stack allocated
objects.
</p>
<p>
The inclusion constraint identification stage finds all inclusion constraints
in the program by scanning the program, looking for pointer assignments and
other statements that effect the points-to graph. For a statement like
<code><var>A</var> = <var>B</var></code>, this statement is processed to
indicate that <var>A</var> can point to anything that <var>B</var> can point
to. Constraints can handle copies, loads, and stores, and address taking.
</p>
<p>
The offline constraint graph optimization portion includes offline variable
substitution algorithms intended to computer pointer and location
equivalences. Pointer equivalences are those pointers that will have the
same points-to sets, and location equivalences are those variables that
always appear together in points-to sets.
</p>
<p>
The inclusion constraint solving phase iteratively propagates the inclusion
constraints until a fixed point is reached. This is an O(<var>n</var>³)
algorithm.
</p>
<p>
Function constraints are handled as if they were structs with <var>X</var>
fields. Thus, an access to argument <var>X</var> of function <var>Y</var> is
an access to node index <code>getNode(<var>Y</var>) + <var>X</var></code>.
This representation allows handling of indirect calls without any issues. To
wit, an indirect call <code><var>Y</var>(<var>a</var>,<var>b</var>)</code> is
equivalent to <code>*(<var>Y</var> + 1) = <var>a</var>, *(<var>Y</var> + 2) =
<var>b</var></code>. The return node for a function <var>F</var> is always
located at <code>getNode(<var>F</var>) + CallReturnPos</code>. The arguments
start at <code>getNode(<var>F</var>) + CallArgPos</code>.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="basicaa">Basic Alias Analysis (default AA impl)</a>
</div>
<div class="doc_text">
<p>
This is the default implementation of the Alias Analysis interface
that simply implements a few identities (two different globals cannot alias,
etc), but otherwise does no analysis.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="basiccg">Basic CallGraph Construction</a>
</div>
<div class="doc_text">
<p>Yet to be written.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="basicvn">Basic Value Numbering (default GVN impl)</a>
</div>
<div class="doc_text">
<p>
This is the default implementation of the <code>ValueNumbering</code>
interface. It walks the SSA def-use chains to trivially identify
lexically identical expressions. This does not require any ahead of time
analysis, so it is a very fast default implementation.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="callgraph">Print a call graph</a>
</div>
<div class="doc_text">
<p>
This pass, only available in <code>opt</code>, prints the call graph to
standard output in a human-readable form.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="callscc">Print SCCs of the Call Graph</a>
</div>
<div class="doc_text">
<p>
This pass, only available in <code>opt</code>, prints the SCCs of the call
graph to standard output in a human-readable form.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="cfgscc">Print SCCs of each function CFG</a>
</div>
<div class="doc_text">
<p>
This pass, only available in <code>opt</code>, prints the SCCs of each
function CFG to standard output in a human-readable form.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="codegenprepare">Optimize for code generation</a>
</div>
<div class="doc_text">
<p>
This pass munges the code in the input function to better prepare it for
SelectionDAG-based code generation. This works around limitations in it's
basic-block-at-a-time approach. It should eventually be removed.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="count-aa">Count Alias Analysis Query Responses</a>
</div>
<div class="doc_text">
<p>
A pass which can be used to count how many alias queries
are being made and how the alias analysis implementation being used responds.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="debug-aa">AA use debugger</a>
</div>
<div class="doc_text">
<p>
This simple pass checks alias analysis users to ensure that if they
create a new value, they do not query AA without informing it of the value.
It acts as a shim over any other AA pass you want.
</p>
<p>
Yes keeping track of every value in the program is expensive, but this is
a debugging pass.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="domfrontier">Dominance Frontier Construction</a>
</div>
<div class="doc_text">
<p>
This pass is a simple dominator construction algorithm for finding forward
dominator frontiers.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="domtree">Dominator Tree Construction</a>
</div>
<div class="doc_text">
<p>
This pass is a simple dominator construction algorithm for finding forward
dominators.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="externalfnconstants">Print external fn callsites passed constants</a>
</div>
<div class="doc_text">
<p>
This pass, only available in <code>opt</code>, prints out call sites to
external functions that are called with constant arguments. This can be
useful when looking for standard library functions we should constant fold
or handle in alias analyses.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="globalsmodref-aa">Simple mod/ref analysis for globals</a>
</div>
<div class="doc_text">
<p>
This simple pass provides alias and mod/ref information for global values
that do not have their address taken, and keeps track of whether functions
read or write memory (are "pure"). For this simple (but very common) case,
we can provide pretty accurate and useful information.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="instcount">Counts the various types of Instructions</a>
</div>
<div class="doc_text">
<p>
This pass collects the count of all instructions and reports them
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="intervals">Interval Partition Construction</a>
</div>
<div class="doc_text">
<p>
This analysis calculates and represents the interval partition of a function,
or a preexisting interval partition.
</p>
<p>
In this way, the interval partition may be used to reduce a flow graph down
to its degenerate single node interval partition (unless it is irreducible).
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="load-vn">Load Value Numbering</a>
</div>
<div class="doc_text">
<p>
This pass value numbers load and call instructions. To do this, it finds
lexically identical load instructions, and uses alias analysis to determine
which loads are guaranteed to produce the same value. To value number call
instructions, it looks for calls to functions that do not write to memory
which do not have intervening instructions that clobber the memory that is
read from.
</p>
<p>
This pass builds off of another value numbering pass to implement value
numbering for non-load and non-call instructions. It uses Alias Analysis so
that it can disambiguate the load instructions. The more powerful these base
analyses are, the more powerful the resultant value numbering will be.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="loops">Natural Loop Construction</a>
</div>
<div class="doc_text">
<p>
This analysis is used to identify natural loops and determine the loop depth
of various nodes of the CFG. Note that the loops identified may actually be
several natural loops that share the same header node... not just a single
natural loop.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="memdep">Memory Dependence Analysis</a>
</div>
<div class="doc_text">
<p>
An analysis that determines, for a given memory operation, what preceding
memory operations it depends on. It builds on alias analysis information, and
tries to provide a lazy, caching interface to a common kind of alias
information query.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="no-aa">No Alias Analysis (always returns 'may' alias)</a>
</div>
<div class="doc_text">
<p>
Always returns "I don't know" for alias queries. NoAA is unlike other alias
analysis implementations, in that it does not chain to a previous analysis. As
such it doesn't follow many of the rules that other alias analyses must.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="no-profile">No Profile Information</a>
</div>
<div class="doc_text">
<p>
The default "no profile" implementation of the abstract
<code>ProfileInfo</code> interface.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="postdomfrontier">Post-Dominance Frontier Construction</a>
</div>
<div class="doc_text">
<p>
This pass is a simple post-dominator construction algorithm for finding
post-dominator frontiers.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="postdomtree">Post-Dominator Tree Construction</a>
</div>
<div class="doc_text">
<p>
This pass is a simple post-dominator construction algorithm for finding
post-dominators.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="print">Print function to stderr</a>
</div>
<div class="doc_text">
<p>
The <code>PrintFunctionPass</code> class is designed to be pipelined with
other <code>FunctionPass</code>es, and prints out the functions of the module
as they are processed.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="print-alias-sets">Alias Set Printer</a>
</div>
<div class="doc_text">
<p>Yet to be written.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="print-callgraph">Print Call Graph to 'dot' file</a>
</div>
<div class="doc_text">
<p>
This pass, only available in <code>opt</code>, prints the call graph into a
<code>.dot</code> graph. This graph can then be processed with the "dot" tool
to convert it to postscript or some other suitable format.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="print-cfg">Print CFG of function to 'dot' file</a>
</div>
<div class="doc_text">
<p>
This pass, only available in <code>opt</code>, prints the control flow graph
into a <code>.dot</code> graph. This graph can then be processed with the
"dot" tool to convert it to postscript or some other suitable format.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="print-cfg-only">Print CFG of function to 'dot' file (with no function bodies)</a>
</div>
<div class="doc_text">
<p>
This pass, only available in <code>opt</code>, prints the control flow graph
into a <code>.dot</code> graph, omitting the function bodies. This graph can
then be processed with the "dot" tool to convert it to postscript or some
other suitable format.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="printm">Print module to stderr</a>
</div>
<div class="doc_text">
<p>
This pass simply prints out the entire module when it is executed.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="printusedtypes">Find Used Types</a>
</div>
<div class="doc_text">
<p>
This pass is used to seek out all of the types in use by the program. Note
that this analysis explicitly does not include types only used by the symbol
table.
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="profile-loader">Load profile information from llvmprof.out</a>
</div>
<div class="doc_text">
<p>
A concrete implementation of profiling information that loads the information
from a profile dump file.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="scalar-evolution">Scalar Evolution Analysis</a>
</div>
<div class="doc_text">
<p>
The <code>ScalarEvolution</code> analysis can be used to analyze and
catagorize scalar expressions in loops. It specializes in recognizing general
induction variables, representing them with the abstract and opaque
<code>SCEV</code> class. Given this analysis, trip counts of loops and other
important properties can be obtained.
</p>
<p>
This analysis is primarily useful for induction variable substitution and
strength reduction.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="targetdata">Target Data Layout</a>
</div>
<div class="doc_text">
<p>Provides other passes access to information on how the size and alignment
required by the the target ABI for various data types.</p>
</div>
<!-- ======================================================================= -->
<div class="doc_section"> <a name="transform">Transform Passes</a></div>
<div class="doc_text">
<p>This section describes the LLVM Transform Passes.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="adce">Aggressive Dead Code Elimination</a>
</div>
<div class="doc_text">
<p>ADCE aggressively tries to eliminate code. This pass is similar to
<a href="#dce">DCE</a> but it assumes that values are dead until proven
otherwise. This is similar to <a href="#sccp">SCCP</a>, except applied to
the liveness of values.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="argpromotion">Promote 'by reference' arguments to scalars</a>
</div>
<div class="doc_text">
<p>
This pass promotes "by reference" arguments to be "by value" arguments. In
practice, this means looking for internal functions that have pointer
arguments. If it can prove, through the use of alias analysis, that an
argument is *only* loaded, then it can pass the value into the function
instead of the address of the value. This can cause recursive simplification
of code and lead to the elimination of allocas (especially in C++ template
code like the STL).
</p>
<p>
This pass also handles aggregate arguments that are passed into a function,
scalarizing them if the elements of the aggregate are only loaded. Note that
it refuses to scalarize aggregates which would require passing in more than
three operands to the function, because passing thousands of operands for a
large array or structure is unprofitable!
</p>
<p>
Note that this transformation could also be done for arguments that are only
stored to (returning the value instead), but does not currently. This case
would be best handled when and if LLVM starts supporting multiple return
values from functions.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="block-placement">Profile Guided Basic Block Placement</a>
</div>
<div class="doc_text">
<p>This pass is a very simple profile guided basic block placement algorithm.
The idea is to put frequently executed blocks together at the start of the
function and hopefully increase the number of fall-through conditional
branches. If there is no profile information for a particular function, this
pass basically orders blocks in depth-first order.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="break-crit-edges">Break critical edges in CFG</a>
</div>
<div class="doc_text">
<p>
Break all of the critical edges in the CFG by inserting a dummy basic block.
It may be "required" by passes that cannot deal with critical edges. This
transformation obviously invalidates the CFG, but can update forward dominator
(set, immediate dominators, tree, and frontier) information.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="cee">Correlated Expression Elimination</a>
</div>
<div class="doc_text">
<p>Correlated Expression Elimination propagates information from conditional
branches to blocks dominated by destinations of the branch. It propagates
information from the condition check itself into the body of the branch,
allowing transformations like these for example:</p>
<blockquote><pre>
if (i == 7)
... 4*i; // constant propagation
M = i+1; N = j+1;
if (i == j)
X = M-N; // = M-M == 0;
</pre></blockquote>
<p>This is called Correlated Expression Elimination because we eliminate or
simplify expressions that are correlated with the direction of a branch. In
this way we use static information to give us some information about the
dynamic value of a variable.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="condprop">Conditional Propagation</a>
</div>
<div class="doc_text">
<p>This pass propagates information about conditional expressions through the
program, allowing it to eliminate conditional branches in some cases.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="constmerge">Merge Duplicate Global Constants</a>
</div>
<div class="doc_text">
<p>
Merges duplicate global constants together into a single constant that is
shared. This is useful because some passes (ie TraceValues) insert a lot of
string constants into the program, regardless of whether or not an existing
string is available.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="constprop">Simple constant propagation</a>
</div>
<div class="doc_text">
<p>This file implements constant propagation and merging. It looks for
instructions involving only constant operands and replaces them with a
constant value instead of an instruction. For example:</p>
<blockquote><pre>add i32 1, 2</pre></blockquote>
<p>becomes</p>
<blockquote><pre>i32 3</pre></blockquote>
<p>NOTE: this pass has a habit of making definitions be dead. It is a good
idea to to run a <a href="#die">DIE</a> (Dead Instruction Elimination) pass
sometime after running this pass.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="dce">Dead Code Elimination</a>
</div>
<div class="doc_text">
<p>
Dead code elimination is similar to <a href="#die">dead instruction
elimination</a>, but it rechecks instructions that were used by removed
instructions to see if they are newly dead.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="deadargelim">Dead Argument Elimination</a>
</div>
<div class="doc_text">
<p>
This pass deletes dead arguments from internal functions. Dead argument
elimination removes arguments which are directly dead, as well as arguments
only passed into function calls as dead arguments of other functions. This
pass also deletes dead arguments in a similar way.
</p>
<p>
This pass is often useful as a cleanup pass to run after aggressive
interprocedural passes, which add possibly-dead arguments.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="deadtypeelim">Dead Type Elimination</a>
</div>
<div class="doc_text">
<p>
This pass is used to cleanup the output of GCC. It eliminate names for types
that are unused in the entire translation unit, using the <a
href="#findusedtypes">find used types</a> pass.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="die">Dead Instruction Elimination</a>
</div>
<div class="doc_text">
<p>
Dead instruction elimination performs a single pass over the function,
removing instructions that are obviously dead.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="dse">Dead Store Elimination</a>
</div>
<div class="doc_text">
<p>
A trivial dead store elimination that only considers basic-block local
redundant stores.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="gcse">Global Common Subexpression Elimination</a>
</div>
<div class="doc_text">
<p>
This pass is designed to be a very quick global transformation that
eliminates global common subexpressions from a function. It does this by
using an existing value numbering implementation to identify the common
subexpressions, eliminating them when possible.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="globaldce">Dead Global Elimination</a>
</div>
<div class="doc_text">
<p>
This transform is designed to eliminate unreachable internal globals from the
program. It uses an aggressive algorithm, searching out globals that are
known to be alive. After it finds all of the globals which are needed, it
deletes whatever is left over. This allows it to delete recursive chunks of
the program which are unreachable.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="globalopt">Global Variable Optimizer</a>
</div>
<div class="doc_text">
<p>
This pass transforms simple global variables that never have their address
taken. If obviously true, it marks read/write globals as constant, deletes
variables only stored to, etc.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="gvn">Global Value Numbering</a>
</div>
<div class="doc_text">
<p>
This pass performs global value numbering to eliminate fully redundant
instructions. It also performs simple dead load elimination.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="gvnpre">Global Value Numbering/Partial Redundancy Elimination</a>
</div>
<div class="doc_text">
<p>
This pass performs a hybrid of global value numbering and partial redundancy
elimination, known as GVN-PRE. It performs partial redundancy elimination on
values, rather than lexical expressions, allowing a more comprehensive view
the optimization. It replaces redundant values with uses of earlier
occurences of the same value. While this is beneficial in that it eliminates
unneeded computation, it also increases register pressure by creating large
live ranges, and should be used with caution on platforms that are very
sensitive to register pressure.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="indmemrem">Indirect Malloc and Free Removal</a>
</div>
<div class="doc_text">
<p>
This pass finds places where memory allocation functions may escape into
indirect land. Some transforms are much easier (aka possible) only if free
or malloc are not called indirectly.
</p>
<p>
Thus find places where the address of memory functions are taken and construct
bounce functions with direct calls of those functions.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="indvars">Canonicalize Induction Variables</a>
</div>
<div class="doc_text">
<p>
This transformation analyzes and transforms the induction variables (and
computations derived from them) into simpler forms suitable for subsequent
analysis and transformation.
</p>
<p>
This transformation makes the following changes to each loop with an
identifiable induction variable:
</p>
<ol>
<li>All loops are transformed to have a <em>single</em> canonical
induction variable which starts at zero and steps by one.</li>
<li>The canonical induction variable is guaranteed to be the first PHI node
in the loop header block.</li>
<li>Any pointer arithmetic recurrences are raised to use array
subscripts.</li>
</ol>
<p>
If the trip count of a loop is computable, this pass also makes the following
changes:
</p>
<ol>
<li>The exit condition for the loop is canonicalized to compare the
induction value against the exit value. This turns loops like:
<blockquote><pre>for (i = 7; i*i < 1000; ++i)</pre></blockquote>
into
<blockquote><pre>for (i = 0; i != 25; ++i)</pre></blockquote></li>
<li>Any use outside of the loop of an expression derived from the indvar
is changed to compute the derived value outside of the loop, eliminating
the dependence on the exit value of the induction variable. If the only
purpose of the loop is to compute the exit value of some derived
expression, this transformation will make the loop dead.</li>
</ol>
<p>
This transformation should be followed by strength reduction after all of the
desired loop transformations have been performed. Additionally, on targets
where it is profitable, the loop could be transformed to count down to zero
(the "do loop" optimization).
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="inline">Function Integration/Inlining</a>
</div>
<div class="doc_text">
<p>
Bottom-up inlining of functions into callees.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="insert-block-profiling">Insert instrumentation for block profiling</a>
</div>
<div class="doc_text">
<p>
This pass instruments the specified program with counters for basic block
profiling, which counts the number of times each basic block executes. This
is the most basic form of profiling, which can tell which blocks are hot, but
cannot reliably detect hot paths through the CFG.
</p>
<p>
Note that this implementation is very naïve. Control equivalent regions of
the CFG should not require duplicate counters, but it does put duplicate
counters in.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="insert-edge-profiling">Insert instrumentation for edge profiling</a>
</div>
<div class="doc_text">
<p>
This pass instruments the specified program with counters for edge profiling.
Edge profiling can give a reasonable approximation of the hot paths through a
program, and is used for a wide variety of program transformations.
</p>
<p>
Note that this implementation is very naïve. It inserts a counter for
<em>every</em> edge in the program, instead of using control flow information
to prune the number of counters inserted.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="insert-function-profiling">Insert instrumentation for function profiling</a>
</div>
<div class="doc_text">
<p>
This pass instruments the specified program with counters for function
profiling, which counts the number of times each function is called.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="insert-null-profiling-rs">Measure profiling framework overhead</a>
</div>
<div class="doc_text">
<p>
The basic profiler that does nothing. It is the default profiler and thus
terminates <code>RSProfiler</code> chains. It is useful for measuring
framework overhead.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="insert-rs-profiling-framework">Insert random sampling instrumentation framework</a>
</div>
<div class="doc_text">
<p>
The second stage of the random-sampling instrumentation framework, duplicates
all instructions in a function, ignoring the profiling code, then connects the
two versions together at the entry and at backedges. At each connection point
a choice is made as to whether to jump to the profiled code (take a sample) or
execute the unprofiled code.
</p>
<p>
After this pass, it is highly recommended to run<a href="#mem2reg">mem2reg</a>
and <a href="#adce">adce</a>. <a href="#instcombine">instcombine</a>,
<a href="#load-vn">load-vn</a>, <a href="#gdce">gdce</a>, and
<a href="#dse">dse</a> also are good to run afterwards.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="instcombine">Combine redundant instructions</a>
</div>
<div class="doc_text">
<p>
Combine instructions to form fewer, simple
instructions. This pass does not modify the CFG This pass is where algebraic
simplification happens.
</p>
<p>
This pass combines things like:
</p>
<blockquote><pre
>%Y = add i32 %X, 1
%Z = add i32 %Y, 1</pre></blockquote>
<p>
into:
</p>
<blockquote><pre
>%Z = add i32 %X, 2</pre></blockquote>
<p>
This is a simple worklist driven algorithm.
</p>
<p>
This pass guarantees that the following canonicalizations are performed on
the program:
</p>
<ul>
<li>If a binary operator has a constant operand, it is moved to the right-
hand side.</li>
<li>Bitwise operators with constant operands are always grouped so that
shifts are performed first, then <code>or</code>s, then
<code>and</code>s, then <code>xor</code>s.</li>
<li>Compare instructions are converted from <code>&lt;</code>,
<code>&gt;</code>, <code>≤</code>, or <code>≥</code> to
<code>=</code> or <code>≠</code> if possible.</li>
<li>All <code>cmp</code> instructions on boolean values are replaced with
logical operations.</li>
<li><code>add <var>X</var>, <var>X</var></code> is represented as
<code>mul <var>X</var>, 2</code> ⇒ <code>shl <var>X</var>, 1</code></li>
<li>Multiplies with a constant power-of-two argument are transformed into
shifts.</li>
<li>… etc.</li>
</ul>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="internalize">Internalize Global Symbols</a>
</div>
<div class="doc_text">
<p>
This pass loops over all of the functions in the input module, looking for a
main function. If a main function is found, all other functions and all
global variables with initializers are marked as internal.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="ipconstprop">Interprocedural constant propagation</a>
</div>
<div class="doc_text">
<p>
This pass implements an <em>extremely</em> simple interprocedural constant
propagation pass. It could certainly be improved in many different ways,
like using a worklist. This pass makes arguments dead, but does not remove
them. The existing dead argument elimination pass should be run after this
to clean up the mess.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="ipsccp">Interprocedural Sparse Conditional Constant Propagation</a>
</div>
<div class="doc_text">
<p>
An interprocedural variant of <a href="#sccp">Sparse Conditional Constant
Propagation</a>.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="lcssa">Loop-Closed SSA Form Pass</a>
</div>
<div class="doc_text">
<p>
This pass transforms loops by placing phi nodes at the end of the loops for
all values that are live across the loop boundary. For example, it turns
the left into the right code:
</p>
<pre
>for (...) for (...)
if (c) if (c)
X1 = ... X1 = ...
else else
X2 = ... X2 = ...
X3 = phi(X1, X2) X3 = phi(X1, X2)
... = X3 + 4 X4 = phi(X3)
... = X4 + 4</pre>
<p>
This is still valid LLVM; the extra phi nodes are purely redundant, and will
be trivially eliminated by <code>InstCombine</code>. The major benefit of
this transformation is that it makes many other loop optimizations, such as
LoopUnswitching, simpler.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="licm">Loop Invariant Code Motion</a>
</div>
<div class="doc_text">
<p>
This pass performs loop invariant code motion, attempting to remove as much
code from the body of a loop as possible. It does this by either hoisting
code into the preheader block, or by sinking code to the exit blocks if it is
safe. This pass also promotes must-aliased memory locations in the loop to
live in registers, thus hoisting and sinking "invariant" loads and stores.
</p>
<p>
This pass uses alias analysis for two purposes:
</p>
<ul>
<li>Moving loop invariant loads and calls out of loops. If we can determine
that a load or call inside of a loop never aliases anything stored to,
we can hoist it or sink it like any other instruction.</li>
<li>Scalar Promotion of Memory - If there is a store instruction inside of
the loop, we try to move the store to happen AFTER the loop instead of
inside of the loop. This can only happen if a few conditions are true:
<ul>
<li>The pointer stored through is loop invariant.</li>
<li>There are no stores or loads in the loop which <em>may</em> alias
the pointer. There are no calls in the loop which mod/ref the
pointer.</li>
</ul>
If these conditions are true, we can promote the loads and stores in the
loop of the pointer to use a temporary alloca'd variable. We then use
the mem2reg functionality to construct the appropriate SSA form for the
variable.</li>
</ul>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="loop-extract">Extract loops into new functions</a>
</div>
<div class="doc_text">
<p>
A pass wrapper around the <code>ExtractLoop()</code> scalar transformation to
extract each top-level loop into its own new function. If the loop is the
<em>only</em> loop in a given function, it is not touched. This is a pass most
useful for debugging via bugpoint.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="loop-extract-single">Extract at most one loop into a new function</a>
</div>
<div class="doc_text">
<p>
Similar to <a href="#loop-extract">Extract loops into new functions</a>,
this pass extracts one natural loop from the program into a function if it
can. This is used by bugpoint.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="loop-index-split">Index Split Loops</a>
</div>
<div class="doc_text">
<p>
This pass divides loop's iteration range by spliting loop such that each
individual loop is executed efficiently.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="loop-reduce">Loop Strength Reduction</a>
</div>
<div class="doc_text">
<p>
This pass performs a strength reduction on array references inside loops that
have as one or more of their components the loop induction variable. This is
accomplished by creating a new value to hold the initial value of the array
access for the first iteration, and then creating a new GEP instruction in
the loop to increment the value by the appropriate amount.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="loop-rotate">Rotate Loops</a>
</div>
<div class="doc_text">
<p>A simple loop rotation transformation.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="loop-unroll">Unroll loops</a>
</div>
<div class="doc_text">
<p>
This pass implements a simple loop unroller. It works best when loops have
been canonicalized by the <a href="#indvars"><tt>-indvars</tt></a> pass,
allowing it to determine the trip counts of loops easily.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="loop-unswitch">Unswitch loops</a>
</div>
<div class="doc_text">
<p>
This pass transforms loops that contain branches on loop-invariant conditions
to have multiple loops. For example, it turns the left into the right code:
</p>
<pre
>for (...) if (lic)
A for (...)
if (lic) A; B; C
B else
C for (...)
A; C</pre>
<p>
This can increase the size of the code exponentially (doubling it every time
a loop is unswitched) so we only unswitch if the resultant code will be
smaller than a threshold.
</p>
<p>
This pass expects LICM to be run before it to hoist invariant conditions out
of the loop, to make the unswitching opportunity obvious.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="loopsimplify">Canonicalize natural loops</a>
</div>
<div class="doc_text">
<p>
This pass performs several transformations to transform natural loops into a
simpler form, which makes subsequent analyses and transformations simpler and
more effective.
</p>
<p>
Loop pre-header insertion guarantees that there is a single, non-critical
entry edge from outside of the loop to the loop header. This simplifies a
number of analyses and transformations, such as LICM.
</p>
<p>
Loop exit-block insertion guarantees that all exit blocks from the loop
(blocks which are outside of the loop that have predecessors inside of the
loop) only have predecessors from inside of the loop (and are thus dominated
by the loop header). This simplifies transformations such as store-sinking
that are built into LICM.
</p>
<p>
This pass also guarantees that loops will have exactly one backedge.
</p>
<p>
Note that the simplifycfg pass will clean up blocks which are split out but
end up being unnecessary, so usage of this pass should not pessimize
generated code.
</p>
<p>
This pass obviously modifies the CFG, but updates loop information and
dominator information.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="lower-packed">lowers packed operations to operations on smaller packed datatypes</a>
</div>
<div class="doc_text">
<p>
Lowers operations on vector datatypes into operations on more primitive vector
datatypes, and finally to scalar operations.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="lowerallocs">Lower allocations from instructions to calls</a>
</div>
<div class="doc_text">
<p>
Turn <tt>malloc</tt> and <tt>free</tt> instructions into <tt>@malloc</tt> and
<tt>@free</tt> calls.
</p>
<p>
This is a target-dependent tranformation because it depends on the size of
data types and alignment constraints.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="lowergc">Lower GC intrinsics, for GCless code generators</a>
</div>
<div class="doc_text">
<p>
This file implements lowering for the <tt>llvm.gc*</tt> intrinsics for targets
that do not natively support them (which includes the C backend). Note that
the code generated is not as efficient as it would be for targets that
natively support the GC intrinsics, but it is useful for getting new targets
up-and-running quickly.
</p>
<p>
This pass implements the code transformation described in this paper:
</p>
<blockquote><p>
"Accurate Garbage Collection in an Uncooperative Environment"
Fergus Henderson, ISMM, 2002
</p></blockquote>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="lowerinvoke">Lower invoke and unwind, for unwindless code generators</a>
</div>
<div class="doc_text">
<p>
This transformation is designed for use by code generators which do not yet
support stack unwinding. This pass supports two models of exception handling
lowering, the 'cheap' support and the 'expensive' support.
</p>
<p>
'Cheap' exception handling support gives the program the ability to execute
any program which does not "throw an exception", by turning 'invoke'
instructions into calls and by turning 'unwind' instructions into calls to
abort(). If the program does dynamically use the unwind instruction, the
program will print a message then abort.
</p>
<p>
'Expensive' exception handling support gives the full exception handling
support to the program at the cost of making the 'invoke' instruction
really expensive. It basically inserts setjmp/longjmp calls to emulate the
exception handling as necessary.
</p>
<p>
Because the 'expensive' support slows down programs a lot, and EH is only
used for a subset of the programs, it must be specifically enabled by the
<tt>-enable-correct-eh-support</tt> option.
</p>
<p>
Note that after this pass runs the CFG is not entirely accurate (exceptional
control flow edges are not correct anymore) so only very simple things should
be done after the lowerinvoke pass has run (like generation of native code).
This should not be used as a general purpose "my LLVM-to-LLVM pass doesn't
support the invoke instruction yet" lowering pass.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="lowerselect">Lower select instructions to branches</a>
</div>
<div class="doc_text">
<p>
Lowers select instructions into conditional branches for targets that do not
have conditional moves or that have not implemented the select instruction
yet.
</p>
<p>
Note that this pass could be improved. In particular it turns every select
instruction into a new conditional branch, even though some common cases have
select instructions on the same predicate next to each other. It would be
better to use the same branch for the whole group of selects.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="lowersetjmp">Lower Set Jump</a>
</div>
<div class="doc_text">
<p>
Lowers <tt>setjmp</tt> and <tt>longjmp</tt> to use the LLVM invoke and unwind
instructions as necessary.
</p>
<p>
Lowering of <tt>longjmp</tt> is fairly trivial. We replace the call with a
call to the LLVM library function <tt>__llvm_sjljeh_throw_longjmp()</tt>.
This unwinds the stack for us calling all of the destructors for
objects allocated on the stack.
</p>
<p>
At a <tt>setjmp</tt> call, the basic block is split and the <tt>setjmp</tt>
removed. The calls in a function that have a <tt>setjmp</tt> are converted to
invoke where the except part checks to see if it's a <tt>longjmp</tt>
exception and, if so, if it's handled in the function. If it is, then it gets
the value returned by the <tt>longjmp</tt> and goes to where the basic block
was split. <tt>invoke</tt> instructions are handled in a similar fashion with
the original except block being executed if it isn't a <tt>longjmp</tt>
except that is handled by that function.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="lowerswitch">Lower SwitchInst's to branches</a>
</div>
<div class="doc_text">
<p>
Rewrites <tt>switch</tt> instructions with a sequence of branches, which
allows targets to get away with not implementing the switch instruction until
it is convenient.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="mem2reg">Promote Memory to Register</a>
</div>
<div class="doc_text">
<p>
This file promotes memory references to be register references. It promotes
<tt>alloca</tt> instructions which only have <tt>load</tt>s and
<tt>store</tt>s as uses. An <tt>alloca</tt> is transformed by using dominator
frontiers to place <tt>phi</tt> nodes, then traversing the function in
depth-first order to rewrite <tt>load</tt>s and <tt>store</tt>s as
appropriate. This is just the standard SSA construction algorithm to construct
"pruned" SSA form.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="mergereturn">Unify function exit nodes</a>
</div>
<div class="doc_text">
<p>
Ensure that functions have at most one <tt>ret</tt> instruction in them.
Additionally, it keeps track of which node is the new exit node of the CFG.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="predsimplify">Predicate Simplifier</a>
</div>
<div class="doc_text">
<p>
Path-sensitive optimizer. In a branch where <tt>x == y</tt>, replace uses of
<tt>x</tt> with <tt>y</tt>. Permits further optimization, such as the
elimination of the unreachable call:
</p>
<blockquote><pre
>void test(int *p, int *q)
{
if (p != q)
return;
if (*p != *q)
foo(); // unreachable
}</pre></blockquote>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="prune-eh">Remove unused exception handling info</a>
</div>
<div class="doc_text">
<p>
This file implements a simple interprocedural pass which walks the call-graph,
turning <tt>invoke</tt> instructions into <tt>call</tt> instructions if and
only if the callee cannot throw an exception. It implements this as a
bottom-up traversal of the call-graph.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="raiseallocs">Raise allocations from calls to instructions</a>
</div>
<div class="doc_text">
<p>
Converts <tt>@malloc</tt> and <tt>@free</tt> calls to <tt>malloc</tt> and
<tt>free</tt> instructions.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="reassociate">Reassociate expressions</a>
</div>
<div class="doc_text">
<p>
This pass reassociates commutative expressions in an order that is designed
to promote better constant propagation, GCSE, LICM, PRE, etc.
</p>
<p>
For example: 4 + (<var>x</var> + 5) ⇒ <var>x</var> + (4 + 5)
</p>
<p>
In the implementation of this algorithm, constants are assigned rank = 0,
function arguments are rank = 1, and other values are assigned ranks
corresponding to the reverse post order traversal of current function
(starting at 2), which effectively gives values in deep loops higher rank
than values not in loops.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="reg2mem">Demote all values to stack slots</a>
</div>
<div class="doc_text">
<p>
This file demotes all registers to memory references. It is intented to be
the inverse of <a href="#mem2reg"><tt>-mem2reg</tt></a>. By converting to
<tt>load</tt> instructions, the only values live accross basic blocks are
<tt>alloca</tt> instructions and <tt>load</tt> instructions before
<tt>phi</tt> nodes. It is intended that this should make CFG hacking much
easier. To make later hacking easier, the entry block is split into two, such
that all introduced <tt>alloca</tt> instructions (and nothing else) are in the
entry block.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="scalarrepl">Scalar Replacement of Aggregates</a>
</div>
<div class="doc_text">
<p>
The well-known scalar replacement of aggregates transformation. This
transform breaks up <tt>alloca</tt> instructions of aggregate type (structure
or array) into individual <tt>alloca</tt> instructions for each member if
possible. Then, if possible, it transforms the individual <tt>alloca</tt>
instructions into nice clean scalar SSA form.
</p>
<p>
This combines a simple scalar replacement of aggregates algorithm with the <a
href="#mem2reg"><tt>mem2reg</tt></a> algorithm because often interact,
especially for C++ programs. As such, iterating between <tt>scalarrepl</tt>,
then <a href="#mem2reg"><tt>mem2reg</tt></a> until we run out of things to
promote works well.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="sccp">Sparse Conditional Constant Propagation</a>
</div>
<div class="doc_text">
<p>
Sparse conditional constant propagation and merging, which can be summarized
as:
</p>
<ol>
<li>Assumes values are constant unless proven otherwise</li>
<li>Assumes BasicBlocks are dead unless proven otherwise</li>
<li>Proves values to be constant, and replaces them with constants</li>
<li>Proves conditional branches to be unconditional</li>
</ol>
<p>
Note that this pass has a habit of making definitions be dead. It is a good
idea to to run a DCE pass sometime after running this pass.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="simplify-libcalls">Simplify well-known library calls</a>
</div>
<div class="doc_text">
<p>
Applies a variety of small optimizations for calls to specific well-known
function calls (e.g. runtime library functions). For example, a call
<tt>exit(3)</tt> that occurs within the <tt>main()</tt> function can be
transformed into simply <tt>return 3</tt>.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="simplifycfg">Simplify the CFG</a>
</div>
<div class="doc_text">
<p>
Performs dead code elimination and basic block merging. Specifically:
</p>
<ol>
<li>Removes basic blocks with no predecessors.</li>
<li>Merges a basic block into its predecessor if there is only one and the
predecessor only has one successor.</li>
<li>Eliminates PHI nodes for basic blocks with a single predecessor.</li>
<li>Eliminates a basic block that only contains an unconditional
branch.</li>
</ol>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="strip">Strip all symbols from a module</a>
</div>
<div class="doc_text">
<p>
Performs code stripping. This transformation can delete:
</p>
<ol>
<li>names for virtual registers</li>
<li>symbols for internal globals and functions</li>
<li>debug information</li>
</ol>
<p>
Note that this transformation makes code much less readable, so it should
only be used in situations where the <tt>strip</tt> utility would be used,
such as reducing code size or making it harder to reverse engineer code.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="tailcallelim">Tail Call Elimination</a>
</div>
<div class="doc_text">
<p>
This file transforms calls of the current function (self recursion) followed
by a return instruction with a branch to the entry of the function, creating
a loop. This pass also implements the following extensions to the basic
algorithm:
</p>
<ul>
<li>Trivial instructions between the call and return do not prevent the
transformation from taking place, though currently the analysis cannot
support moving any really useful instructions (only dead ones).
<li>This pass transforms functions that are prevented from being tail
recursive by an associative expression to use an accumulator variable,
thus compiling the typical naive factorial or <tt>fib</tt> implementation
into efficient code.
<li>TRE is performed if the function returns void, if the return
returns the result returned by the call, or if the function returns a
run-time constant on all exits from the function. It is possible, though
unlikely, that the return returns something else (like constant 0), and
can still be TRE'd. It can be TRE'd if <em>all other</em> return
instructions in the function return the exact same value.
<li>If it can prove that callees do not access theier caller stack frame,
they are marked as eligible for tail call elimination (by the code
generator).
</ul>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="tailduplicate">Tail Duplication</a>
</div>
<div class="doc_text">
<p>
This pass performs a limited form of tail duplication, intended to simplify
CFGs by removing some unconditional branches. This pass is necessary to
straighten out loops created by the C front-end, but also is capable of
making other code nicer. After this pass is run, the CFG simplify pass
should be run to clean up the mess.
</p>
</div>
<!-- ======================================================================= -->
<div class="doc_section"> <a name="transform">Utility Passes</a></div>
<div class="doc_text">
<p>This section describes the LLVM Utility Passes.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="deadarghaX0r">Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</a>
</div>
<div class="doc_text">
<p>
Same as dead argument elimination, but deletes arguments to functions which
are external. This is only for use by <a
href="Bugpoint.html">bugpoint</a>.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="extract-blocks">Extract Basic Blocks From Module (for bugpoint use)</a>
</div>
<div class="doc_text">
<p>
This pass is used by bugpoint to extract all blocks from the module into their
own functions.</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="preverify">Preliminary module verification</a>
</div>
<div class="doc_text">
<p>
Ensures that the module is in the form required by the <a
href="#verifier">Module Verifier</a> pass.
</p>
<p>
Running the verifier runs this pass automatically, so there should be no need
to use it directly.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="verify">Module Verifier</a>
</div>
<div class="doc_text">
<p>
Verifies an LLVM IR code. This is useful to run after an optimization which is
undergoing testing. Note that <tt>llvm-as</tt> verifies its input before
emitting bitcode, and also that malformed bitcode is likely to make LLVM
crash. All language front-ends are therefore encouraged to verify their output
before performing optimizing transformations.
</p>
<ul>
<li>Both of a binary operator's parameters are of the same type.</li>
<li>Verify that the indices of mem access instructions match other
operands.</li>
<li>Verify that arithmetic and other things are only performed on
first-class types. Verify that shifts and logicals only happen on
integrals f.e.</li>
<li>All of the constants in a switch statement are of the correct type.</li>
<li>The code is in valid SSA form.</li>
<li>It should be illegal to put a label into any other type (like a
structure) or to return one. [except constant arrays!]</li>
<li>Only phi nodes can be self referential: <tt>%x = add int %x, %x</tt> is
invalid.</li>
<li>PHI nodes must have an entry for each predecessor, with no extras.</li>
<li>PHI nodes must be the first thing in a basic block, all grouped
together.</li>
<li>PHI nodes must have at least one entry.</li>
<li>All basic blocks should only end with terminator insts, not contain
them.</li>
<li>The entry node to a function must not have predecessors.</li>
<li>All Instructions must be embedded into a basic block.</li>
<li>Functions cannot take a void-typed parameter.</li>
<li>Verify that a function's argument list agrees with its declared
type.</li>
<li>It is illegal to specify a name for a void value.</li>
<li>It is illegal to have a internal global value with no initializer.</li>
<li>It is illegal to have a ret instruction that returns a value that does
not agree with the function return value type.</li>
<li>Function call argument types match the function prototype.</li>
<li>All other things that are tested by asserts spread about the code.</li>
</ul>
<p>
Note that this does not provide full security verification (like Java), but
instead just tries to ensure that code is well-formed.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="view-cfg">View CFG of function</a>
</div>
<div class="doc_text">
<p>
Displays the control flow graph using the GraphViz tool.
</p>
</div>
<!-------------------------------------------------------------------------- -->
<div class="doc_subsection">
<a name="view-cfg-only">View CFG of function (with no function bodies)</a>
</div>
<div class="doc_text">
<p>
Displays the control flow graph using the GraphViz tool, but omitting function
bodies.
</p>
</div>
<!-- *********************************************************************** -->
<hr>
<address>
<a href="http://jigsaw.w3.org/css-validator/check/referer"><img
src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
<a href="http://validator.w3.org/check/referer"><img
src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
<a href="mailto:rspencer@x10sys.com">Reid Spencer</a><br>
<a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
Last modified: $Date$
</address>
</body>
</html>