Gitiles
Code Review Sign In
gerrit-public.fairphone.software / platform / external / llvm / 3ac1ab835caacdeebbd0d7b4d69160f283928d21 / . / docs / Passes.html
blob: 8c086720a8bb731b8530b5e08ea20c1d007e2fbe [file] [log] [blame]
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
3<html>
4<head>
5 <title>LLVM's Analysis and Transform Passes</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
Gordon Henriksenddaa61d2007-10-25 08:58:56 +0000[diff] [blame]7 <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]8</head>
9<body>
10
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]11<!--
12
13If Passes.html is up to date, the following "one-liner" should print
14an empty diff.
15
16egrep -e '^<tr><td><a href="#.*">-.*</a></td><td>.*</td></tr>$' \
17 -e '^ <a name=".*">.*</a>$' < Passes.html >html; \
18perl >help <<'EOT' && diff -u help html; rm -f help html
19open HTML, "<Passes.html" or die "open: Passes.html: $!\n";
20while (<HTML>) {
21 m:^<tr><td><a href="#(.*)">-.*</a></td><td>.*</td></tr>$: or next;
22 $order{$1} = sprintf("%03d", 1 + int %order);
23}
Gordon Henriksenddaa61d2007-10-25 08:58:56 +0000[diff] [blame]24open HELP, "../Release/bin/opt -help|" or die "open: opt -help: $!\n";
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]25while (<HELP>) {
26 m:^ -([^ ]+) +- (.*)$: or next;
27 my $o = $order{$1};
28 $o = "000" unless defined $o;
29 push @x, "$o<tr><td><a href=\"#$1\">-$1</a></td><td>$2</td></tr>\n";
30 push @y, "$o <a name=\"$1\">$2</a>\n";
31}
32@x = map { s/^\d\d\d//; $_ } sort @x;
33@y = map { s/^\d\d\d//; $_ } sort @y;
34print @x, @y;
35EOT
36
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]37This (real) one-liner can also be helpful when converting comments to HTML:
38
39perl -e '$/ = 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'
40
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]41 -->
42
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]43<div class="doc_title">LLVM's Analysis and Transform Passes</div>
44
45<ol>
46 <li><a href="#intro">Introduction</a></li>
47 <li><a href="#analyses">Analysis Passes</a>
48 <li><a href="#transforms">Transform Passes</a></li>
49 <li><a href="#utilities">Utility Passes</a></li>
50</ol>
51
52<div class="doc_author">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]53 <p>Written by <a href="mailto:rspencer@x10sys.com">Reid Spencer</a>
54 and Gordon Henriksen</p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]55</div>
56
57<!-- ======================================================================= -->
58<div class="doc_section"> <a name="intro">Introduction</a> </div>
59<div class="doc_text">
60 <p>This document serves as a high level summary of the optimization features
61 that LLVM provides. Optimizations are implemented as Passes that traverse some
62 portion of a program to either collect information or transform the program.
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]63 The table below divides the passes that LLVM provides into three categories.
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]64 Analysis passes compute information that other passes can use or for debugging
65 or program visualization purposes. Transform passes can use (or invalidate)
66 the analysis passes. Transform passes all mutate the program in some way.
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]67 Utility passes provides some utility but don't otherwise fit categorization.
Gabor Greif04367bf2007-07-06 22:07:22 +0000[diff] [blame]68 For example passes to extract functions to bitcode or write a module to
69 bitcode are neither analysis nor transform passes.
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]70 <p>The table below provides a quick summary of each pass and links to the more
71 complete pass description later in the document.</p>
72</div>
73<div class="doc_text" >
74<table>
Gordon Henriksenddaa61d2007-10-25 08:58:56 +0000[diff] [blame]75<tr><th colspan="2"><b>ANALYSIS PASSES</b></th></tr>
76<tr><th>Option</th><th>Name</th></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]77<tr><td><a href="#aa-eval">-aa-eval</a></td><td>Exhaustive Alias Analysis Precision Evaluator</td></tr>
78<tr><td><a href="#anders-aa">-anders-aa</a></td><td>Andersen's Interprocedural Alias Analysis</td></tr>
79<tr><td><a href="#basicaa">-basicaa</a></td><td>Basic Alias Analysis (default AA impl)</td></tr>
80<tr><td><a href="#basiccg">-basiccg</a></td><td>Basic CallGraph Construction</td></tr>
81<tr><td><a href="#basicvn">-basicvn</a></td><td>Basic Value Numbering (default GVN impl)</td></tr>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]82<tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Optimize for code generation</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]83<tr><td><a href="#count-aa">-count-aa</a></td><td>Count Alias Analysis Query Responses</td></tr>
84<tr><td><a href="#debug-aa">-debug-aa</a></td><td>AA use debugger</td></tr>
85<tr><td><a href="#domfrontier">-domfrontier</a></td><td>Dominance Frontier Construction</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]86<tr><td><a href="#domtree">-domtree</a></td><td>Dominator Tree Construction</td></tr>
Duncan Sands3ee8fc92008-09-23 12:47:39 +0000[diff] [blame]87<tr><td><a href="#dot-callgraph">-dot-callgraph</a></td><td>Print Call Graph to 'dot' file</td></tr>
88<tr><td><a href="#dot-cfg">-dot-cfg</a></td><td>Print CFG of function to 'dot' file</td></tr>
89<tr><td><a href="#dot-cfg-only">-dot-cfg-only</a></td><td>Print CFG of function to 'dot' file (with no function bodies)</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]90<tr><td><a href="#globalsmodref-aa">-globalsmodref-aa</a></td><td>Simple mod/ref analysis for globals</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]91<tr><td><a href="#instcount">-instcount</a></td><td>Counts the various types of Instructions</td></tr>
92<tr><td><a href="#intervals">-intervals</a></td><td>Interval Partition Construction</td></tr>
93<tr><td><a href="#load-vn">-load-vn</a></td><td>Load Value Numbering</td></tr>
94<tr><td><a href="#loops">-loops</a></td><td>Natural Loop Construction</td></tr>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]95<tr><td><a href="#memdep">-memdep</a></td><td>Memory Dependence Analysis</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]96<tr><td><a href="#no-aa">-no-aa</a></td><td>No Alias Analysis (always returns 'may' alias)</td></tr>
97<tr><td><a href="#no-profile">-no-profile</a></td><td>No Profile Information</td></tr>
98<tr><td><a href="#postdomfrontier">-postdomfrontier</a></td><td>Post-Dominance Frontier Construction</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]99<tr><td><a href="#postdomtree">-postdomtree</a></td><td>Post-Dominator Tree Construction</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]100<tr><td><a href="#print-alias-sets">-print-alias-sets</a></td><td>Alias Set Printer</td></tr>
Duncan Sands3ee8fc92008-09-23 12:47:39 +0000[diff] [blame]101<tr><td><a href="#print-callgraph">-print-callgraph</a></td><td>Print a call graph</td></tr>
102<tr><td><a href="#print-callgraph-sccs">-print-callgraph-sccs</a></td><td>Print SCCs of the Call Graph</td></tr>
103<tr><td><a href="#print-cfg-sccs">-print-cfg-sccs</a></td><td>Print SCCs of each function CFG</td></tr>
104<tr><td><a href="#print-externalfnconstants">-print-externalfnconstants</a></td><td>Print external fn callsites passed constants</td></tr>
105<tr><td><a href="#print-function">-print-function</a></td><td>Print function to stderr</td></tr>
106<tr><td><a href="#print-module">-print-module</a></td><td>Print module to stderr</td></tr>
107<tr><td><a href="#print-used-types">-print-used-types</a></td><td>Find Used Types</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]108<tr><td><a href="#profile-loader">-profile-loader</a></td><td>Load profile information from llvmprof.out</td></tr>
109<tr><td><a href="#scalar-evolution">-scalar-evolution</a></td><td>Scalar Evolution Analysis</td></tr>
110<tr><td><a href="#targetdata">-targetdata</a></td><td>Target Data Layout</td></tr>
111
112
Gordon Henriksenddaa61d2007-10-25 08:58:56 +0000[diff] [blame]113<tr><th colspan="2"><b>TRANSFORM PASSES</b></th></tr>
114<tr><th>Option</th><th>Name</th></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]115<tr><td><a href="#adce">-adce</a></td><td>Aggressive Dead Code Elimination</td></tr>
116<tr><td><a href="#argpromotion">-argpromotion</a></td><td>Promote 'by reference' arguments to scalars</td></tr>
117<tr><td><a href="#block-placement">-block-placement</a></td><td>Profile Guided Basic Block Placement</td></tr>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]118<tr><td><a href="#break-crit-edges">-break-crit-edges</a></td><td>Break critical edges in CFG</td></tr>
Gordon Henriksena8a118b2008-05-08 17:46:35 +0000[diff] [blame]119<tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Prepare a function for code generation </td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]120<tr><td><a href="#condprop">-condprop</a></td><td>Conditional Propagation</td></tr>
121<tr><td><a href="#constmerge">-constmerge</a></td><td>Merge Duplicate Global Constants</td></tr>
122<tr><td><a href="#constprop">-constprop</a></td><td>Simple constant propagation</td></tr>
123<tr><td><a href="#dce">-dce</a></td><td>Dead Code Elimination</td></tr>
124<tr><td><a href="#deadargelim">-deadargelim</a></td><td>Dead Argument Elimination</td></tr>
125<tr><td><a href="#deadtypeelim">-deadtypeelim</a></td><td>Dead Type Elimination</td></tr>
126<tr><td><a href="#die">-die</a></td><td>Dead Instruction Elimination</td></tr>
127<tr><td><a href="#dse">-dse</a></td><td>Dead Store Elimination</td></tr>
128<tr><td><a href="#gcse">-gcse</a></td><td>Global Common Subexpression Elimination</td></tr>
129<tr><td><a href="#globaldce">-globaldce</a></td><td>Dead Global Elimination</td></tr>
130<tr><td><a href="#globalopt">-globalopt</a></td><td>Global Variable Optimizer</td></tr>
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]131<tr><td><a href="#gvn">-gvn</a></td><td>Global Value Numbering</td></tr>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]132<tr><td><a href="#gvnpre">-gvnpre</a></td><td>Global Value Numbering/Partial Redundancy Elimination</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]133<tr><td><a href="#indmemrem">-indmemrem</a></td><td>Indirect Malloc and Free Removal</td></tr>
134<tr><td><a href="#indvars">-indvars</a></td><td>Canonicalize Induction Variables</td></tr>
135<tr><td><a href="#inline">-inline</a></td><td>Function Integration/Inlining</td></tr>
136<tr><td><a href="#insert-block-profiling">-insert-block-profiling</a></td><td>Insert instrumentation for block profiling</td></tr>
137<tr><td><a href="#insert-edge-profiling">-insert-edge-profiling</a></td><td>Insert instrumentation for edge profiling</td></tr>
138<tr><td><a href="#insert-function-profiling">-insert-function-profiling</a></td><td>Insert instrumentation for function profiling</td></tr>
139<tr><td><a href="#insert-null-profiling-rs">-insert-null-profiling-rs</a></td><td>Measure profiling framework overhead</td></tr>
140<tr><td><a href="#insert-rs-profiling-framework">-insert-rs-profiling-framework</a></td><td>Insert random sampling instrumentation framework</td></tr>
141<tr><td><a href="#instcombine">-instcombine</a></td><td>Combine redundant instructions</td></tr>
142<tr><td><a href="#internalize">-internalize</a></td><td>Internalize Global Symbols</td></tr>
143<tr><td><a href="#ipconstprop">-ipconstprop</a></td><td>Interprocedural constant propagation</td></tr>
144<tr><td><a href="#ipsccp">-ipsccp</a></td><td>Interprocedural Sparse Conditional Constant Propagation</td></tr>
Gordon Henriksena8a118b2008-05-08 17:46:35 +0000[diff] [blame]145<tr><td><a href="#jump-threading">-jump-threading</a></td><td>Thread control through conditional blocks </td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]146<tr><td><a href="#lcssa">-lcssa</a></td><td>Loop-Closed SSA Form Pass</td></tr>
147<tr><td><a href="#licm">-licm</a></td><td>Loop Invariant Code Motion</td></tr>
Gordon Henriksena8a118b2008-05-08 17:46:35 +0000[diff] [blame]148<tr><td><a href="#loop-deletion">-loop-deletion</a></td><td>Dead Loop Deletion Pass </td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]149<tr><td><a href="#loop-extract">-loop-extract</a></td><td>Extract loops into new functions</td></tr>
150<tr><td><a href="#loop-extract-single">-loop-extract-single</a></td><td>Extract at most one loop into a new function</td></tr>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]151<tr><td><a href="#loop-index-split">-loop-index-split</a></td><td>Index Split Loops</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]152<tr><td><a href="#loop-reduce">-loop-reduce</a></td><td>Loop Strength Reduction</td></tr>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]153<tr><td><a href="#loop-rotate">-loop-rotate</a></td><td>Rotate Loops</td></tr>
154<tr><td><a href="#loop-unroll">-loop-unroll</a></td><td>Unroll loops</td></tr>
155<tr><td><a href="#loop-unswitch">-loop-unswitch</a></td><td>Unswitch loops</td></tr>
156<tr><td><a href="#loopsimplify">-loopsimplify</a></td><td>Canonicalize natural loops</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]157<tr><td><a href="#lowerallocs">-lowerallocs</a></td><td>Lower allocations from instructions to calls</td></tr>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]158<tr><td><a href="#lowerinvoke">-lowerinvoke</a></td><td>Lower invoke and unwind, for unwindless code generators</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]159<tr><td><a href="#lowersetjmp">-lowersetjmp</a></td><td>Lower Set Jump</td></tr>
160<tr><td><a href="#lowerswitch">-lowerswitch</a></td><td>Lower SwitchInst's to branches</td></tr>
161<tr><td><a href="#mem2reg">-mem2reg</a></td><td>Promote Memory to Register</td></tr>
Gordon Henriksena8a118b2008-05-08 17:46:35 +0000[diff] [blame]162<tr><td><a href="#memcpyopt">-memcpyopt</a></td><td>Optimize use of memcpy and friends</td></tr>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]163<tr><td><a href="#mergereturn">-mergereturn</a></td><td>Unify function exit nodes</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]164<tr><td><a href="#predsimplify">-predsimplify</a></td><td>Predicate Simplifier</td></tr>
165<tr><td><a href="#prune-eh">-prune-eh</a></td><td>Remove unused exception handling info</td></tr>
166<tr><td><a href="#raiseallocs">-raiseallocs</a></td><td>Raise allocations from calls to instructions</td></tr>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]167<tr><td><a href="#reassociate">-reassociate</a></td><td>Reassociate expressions</td></tr>
168<tr><td><a href="#reg2mem">-reg2mem</a></td><td>Demote all values to stack slots</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]169<tr><td><a href="#scalarrepl">-scalarrepl</a></td><td>Scalar Replacement of Aggregates</td></tr>
170<tr><td><a href="#sccp">-sccp</a></td><td>Sparse Conditional Constant Propagation</td></tr>
171<tr><td><a href="#simplify-libcalls">-simplify-libcalls</a></td><td>Simplify well-known library calls</td></tr>
172<tr><td><a href="#simplifycfg">-simplifycfg</a></td><td>Simplify the CFG</td></tr>
173<tr><td><a href="#strip">-strip</a></td><td>Strip all symbols from a module</td></tr>
Gordon Henriksena8a118b2008-05-08 17:46:35 +0000[diff] [blame]174<tr><td><a href="#strip-dead-prototypes">-strip-dead-prototypes</a></td><td>Remove unused function declarations</td></tr>
175<tr><td><a href="#sretpromotion">-sretpromotion</a></td><td>Promote sret arguments</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]176<tr><td><a href="#tailcallelim">-tailcallelim</a></td><td>Tail Call Elimination</td></tr>
177<tr><td><a href="#tailduplicate">-tailduplicate</a></td><td>Tail Duplication</td></tr>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]178
179
Gordon Henriksenddaa61d2007-10-25 08:58:56 +0000[diff] [blame]180<tr><th colspan="2"><b>UTILITY PASSES</b></th></tr>
181<tr><th>Option</th><th>Name</th></tr>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]182<tr><td><a href="#deadarghaX0r">-deadarghaX0r</a></td><td>Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</td></tr>
183<tr><td><a href="#extract-blocks">-extract-blocks</a></td><td>Extract Basic Blocks From Module (for bugpoint use)</td></tr>
Gordon Henriksen90a52142007-11-05 02:05:35 +0000[diff] [blame]184<tr><td><a href="#preverify">-preverify</a></td><td>Preliminary module verification</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]185<tr><td><a href="#verify">-verify</a></td><td>Module Verifier</td></tr>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]186<tr><td><a href="#view-cfg">-view-cfg</a></td><td>View CFG of function</td></tr>
187<tr><td><a href="#view-cfg-only">-view-cfg-only</a></td><td>View CFG of function (with no function bodies)</td></tr>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]188</table>
189</div>
190
191<!-- ======================================================================= -->
192<div class="doc_section"> <a name="example">Analysis Passes</a></div>
193<div class="doc_text">
194 <p>This section describes the LLVM Analysis Passes.</p>
195</div>
196
197<!-------------------------------------------------------------------------- -->
198<div class="doc_subsection">
199 <a name="aa-eval">Exhaustive Alias Analysis Precision Evaluator</a>
200</div>
201<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]202 <p>This is a simple N^2 alias analysis accuracy evaluator.
203 Basically, for each function in the program, it simply queries to see how the
204 alias analysis implementation answers alias queries between each pair of
205 pointers in the function.</p>
206
207 <p>This is inspired and adapted from code by: Naveen Neelakantam, Francesco
208 Spadini, and Wojciech Stryjewski.</p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]209</div>
210
211<!-------------------------------------------------------------------------- -->
212<div class="doc_subsection">
213 <a name="anders-aa">Andersen's Interprocedural Alias Analysis</a>
214</div>
215<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]216 <p>
217 This is an implementation of Andersen's interprocedural alias
218 analysis
219 </p>
220
221 <p>
222 In pointer analysis terms, this is a subset-based, flow-insensitive,
223 field-sensitive, and context-insensitive algorithm pointer algorithm.
224 </p>
225
226 <p>
227 This algorithm is implemented as three stages:
228 </p>
229
230 <ol>
231 <li>Object identification.</li>
232 <li>Inclusion constraint identification.</li>
233 <li>Offline constraint graph optimization.</li>
234 <li>Inclusion constraint solving.</li>
235 </ol>
236
237 <p>
238 The object identification stage identifies all of the memory objects in the
239 program, which includes globals, heap allocated objects, and stack allocated
240 objects.
241 </p>
242
243 <p>
244 The inclusion constraint identification stage finds all inclusion constraints
245 in the program by scanning the program, looking for pointer assignments and
246 other statements that effect the points-to graph. For a statement like
247 <code><var>A</var> = <var>B</var></code>, this statement is processed to
248 indicate that <var>A</var> can point to anything that <var>B</var> can point
249 to. Constraints can handle copies, loads, and stores, and address taking.
250 </p>
251
252 <p>
253 The offline constraint graph optimization portion includes offline variable
254 substitution algorithms intended to computer pointer and location
255 equivalences. Pointer equivalences are those pointers that will have the
256 same points-to sets, and location equivalences are those variables that
257 always appear together in points-to sets.
258 </p>
259
260 <p>
261 The inclusion constraint solving phase iteratively propagates the inclusion
262 constraints until a fixed point is reached. This is an O(<var>n</var>³)
263 algorithm.
264 </p>
265
266 <p>
267 Function constraints are handled as if they were structs with <var>X</var>
268 fields. Thus, an access to argument <var>X</var> of function <var>Y</var> is
269 an access to node index <code>getNode(<var>Y</var>) + <var>X</var></code>.
270 This representation allows handling of indirect calls without any issues. To
271 wit, an indirect call <code><var>Y</var>(<var>a</var>,<var>b</var>)</code> is
272 equivalent to <code>*(<var>Y</var> + 1) = <var>a</var>, *(<var>Y</var> + 2) =
273 <var>b</var></code>. The return node for a function <var>F</var> is always
274 located at <code>getNode(<var>F</var>) + CallReturnPos</code>. The arguments
275 start at <code>getNode(<var>F</var>) + CallArgPos</code>.
276 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]277</div>
278
279<!-------------------------------------------------------------------------- -->
280<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]281 <a name="basicaa">Basic Alias Analysis (default AA impl)</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]282</div>
283<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]284 <p>
285 This is the default implementation of the Alias Analysis interface
286 that simply implements a few identities (two different globals cannot alias,
287 etc), but otherwise does no analysis.
288 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]289</div>
290
291<!-------------------------------------------------------------------------- -->
292<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]293 <a name="basiccg">Basic CallGraph Construction</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]294</div>
295<div class="doc_text">
296 <p>Yet to be written.</p>
297</div>
298
299<!-------------------------------------------------------------------------- -->
300<div class="doc_subsection">
Matthijs Kooijman845f5242008-06-05 07:55:49 +0000[diff] [blame]301 <a name="basicvn">Basic Value Numbering (default Value Numbering impl)</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]302</div>
303<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]304 <p>
305 This is the default implementation of the <code>ValueNumbering</code>
306 interface. It walks the SSA def-use chains to trivially identify
307 lexically identical expressions. This does not require any ahead of time
308 analysis, so it is a very fast default implementation.
309 </p>
Matthijs Kooijman845f5242008-06-05 07:55:49 +0000[diff] [blame]310 <p>
311 The ValueNumbering analysis passes are mostly deprecated. They are only used
312 by the <a href="#gcse">Global Common Subexpression Elimination pass</a>, which
313 is deprecated by the <a href="#gvn">Global Value Numbering pass</a> (which
314 does its value numbering on its own).
315 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]316</div>
317
318<!-------------------------------------------------------------------------- -->
319<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]320 <a name="codegenprepare">Optimize for code generation</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]321</div>
322<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]323 <p>
324 This pass munges the code in the input function to better prepare it for
325 SelectionDAG-based code generation. This works around limitations in it's
326 basic-block-at-a-time approach. It should eventually be removed.
327 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]328</div>
329
330<!-------------------------------------------------------------------------- -->
331<div class="doc_subsection">
332 <a name="count-aa">Count Alias Analysis Query Responses</a>
333</div>
334<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]335 <p>
336 A pass which can be used to count how many alias queries
337 are being made and how the alias analysis implementation being used responds.
338 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]339</div>
340
341<!-------------------------------------------------------------------------- -->
342<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]343 <a name="debug-aa">AA use debugger</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]344</div>
345<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]346 <p>
347 This simple pass checks alias analysis users to ensure that if they
348 create a new value, they do not query AA without informing it of the value.
349 It acts as a shim over any other AA pass you want.
350 </p>
351
352 <p>
353 Yes keeping track of every value in the program is expensive, but this is
354 a debugging pass.
355 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]356</div>
357
358<!-------------------------------------------------------------------------- -->
359<div class="doc_subsection">
360 <a name="domfrontier">Dominance Frontier Construction</a>
361</div>
362<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]363 <p>
364 This pass is a simple dominator construction algorithm for finding forward
365 dominator frontiers.
366 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]367</div>
368
369<!-------------------------------------------------------------------------- -->
370<div class="doc_subsection">
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]371 <a name="domtree">Dominator Tree Construction</a>
372</div>
373<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]374 <p>
375 This pass is a simple dominator construction algorithm for finding forward
376 dominators.
377 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]378</div>
379
380<!-------------------------------------------------------------------------- -->
381<div class="doc_subsection">
Duncan Sands3ee8fc92008-09-23 12:47:39 +0000[diff] [blame]382 <a name="dot-callgraph">Print Call Graph to 'dot' file</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]383</div>
384<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]385 <p>
Duncan Sands3ee8fc92008-09-23 12:47:39 +0000[diff] [blame]386 This pass, only available in <code>opt</code>, prints the call graph into a
387 <code>.dot</code> graph. This graph can then be processed with the "dot" tool
388 to convert it to postscript or some other suitable format.
389 </p>
390</div>
391
392<!-------------------------------------------------------------------------- -->
393<div class="doc_subsection">
394 <a name="dot-cfg">Print CFG of function to 'dot' file</a>
395</div>
396<div class="doc_text">
397 <p>
398 This pass, only available in <code>opt</code>, prints the control flow graph
399 into a <code>.dot</code> graph. This graph can then be processed with the
400 "dot" tool to convert it to postscript or some other suitable format.
401 </p>
402</div>
403
404<!-------------------------------------------------------------------------- -->
405<div class="doc_subsection">
406 <a name="dot-cfg-only">Print CFG of function to 'dot' file (with no function bodies)</a>
407</div>
408<div class="doc_text">
409 <p>
410 This pass, only available in <code>opt</code>, prints the control flow graph
411 into a <code>.dot</code> graph, omitting the function bodies. This graph can
412 then be processed with the "dot" tool to convert it to postscript or some
413 other suitable format.
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]414 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]415</div>
416
417<!-------------------------------------------------------------------------- -->
418<div class="doc_subsection">
419 <a name="globalsmodref-aa">Simple mod/ref analysis for globals</a>
420</div>
421<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]422 <p>
423 This simple pass provides alias and mod/ref information for global values
424 that do not have their address taken, and keeps track of whether functions
425 read or write memory (are "pure"). For this simple (but very common) case,
426 we can provide pretty accurate and useful information.
427 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]428</div>
429
430<!-------------------------------------------------------------------------- -->
431<div class="doc_subsection">
432 <a name="instcount">Counts the various types of Instructions</a>
433</div>
434<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]435 <p>
436 This pass collects the count of all instructions and reports them
437 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]438</div>
439
440<!-------------------------------------------------------------------------- -->
441<div class="doc_subsection">
442 <a name="intervals">Interval Partition Construction</a>
443</div>
444<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]445 <p>
446 This analysis calculates and represents the interval partition of a function,
447 or a preexisting interval partition.
448 </p>
449
450 <p>
451 In this way, the interval partition may be used to reduce a flow graph down
452 to its degenerate single node interval partition (unless it is irreducible).
453 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]454</div>
455
456<!-------------------------------------------------------------------------- -->
457<div class="doc_subsection">
458 <a name="load-vn">Load Value Numbering</a>
459</div>
460<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]461 <p>
462 This pass value numbers load and call instructions. To do this, it finds
463 lexically identical load instructions, and uses alias analysis to determine
464 which loads are guaranteed to produce the same value. To value number call
465 instructions, it looks for calls to functions that do not write to memory
466 which do not have intervening instructions that clobber the memory that is
467 read from.
468 </p>
469
470 <p>
471 This pass builds off of another value numbering pass to implement value
472 numbering for non-load and non-call instructions. It uses Alias Analysis so
473 that it can disambiguate the load instructions. The more powerful these base
474 analyses are, the more powerful the resultant value numbering will be.
475 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]476</div>
477
478<!-------------------------------------------------------------------------- -->
479<div class="doc_subsection">
480 <a name="loops">Natural Loop Construction</a>
481</div>
482<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]483 <p>
484 This analysis is used to identify natural loops and determine the loop depth
485 of various nodes of the CFG. Note that the loops identified may actually be
486 several natural loops that share the same header node... not just a single
487 natural loop.
488 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]489</div>
490
491<!-------------------------------------------------------------------------- -->
492<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]493 <a name="memdep">Memory Dependence Analysis</a>
494</div>
495<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]496 <p>
497 An analysis that determines, for a given memory operation, what preceding
498 memory operations it depends on. It builds on alias analysis information, and
499 tries to provide a lazy, caching interface to a common kind of alias
500 information query.
501 </p>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]502</div>
503
504<!-------------------------------------------------------------------------- -->
505<div class="doc_subsection">
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]506 <a name="no-aa">No Alias Analysis (always returns 'may' alias)</a>
507</div>
508<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]509 <p>
510 Always returns "I don't know" for alias queries. NoAA is unlike other alias
511 analysis implementations, in that it does not chain to a previous analysis. As
512 such it doesn't follow many of the rules that other alias analyses must.
513 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]514</div>
515
516<!-------------------------------------------------------------------------- -->
517<div class="doc_subsection">
518 <a name="no-profile">No Profile Information</a>
519</div>
520<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]521 <p>
522 The default "no profile" implementation of the abstract
523 <code>ProfileInfo</code> interface.
524 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]525</div>
526
527<!-------------------------------------------------------------------------- -->
528<div class="doc_subsection">
529 <a name="postdomfrontier">Post-Dominance Frontier Construction</a>
530</div>
531<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]532 <p>
533 This pass is a simple post-dominator construction algorithm for finding
534 post-dominator frontiers.
535 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]536</div>
537
538<!-------------------------------------------------------------------------- -->
539<div class="doc_subsection">
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]540 <a name="postdomtree">Post-Dominator Tree Construction</a>
541</div>
542<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]543 <p>
544 This pass is a simple post-dominator construction algorithm for finding
545 post-dominators.
546 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]547</div>
548
549<!-------------------------------------------------------------------------- -->
550<div class="doc_subsection">
Duncan Sands3ee8fc92008-09-23 12:47:39 +0000[diff] [blame]551 <a name="print-alias-sets">Alias Set Printer</a>
552</div>
553<div class="doc_text">
554 <p>Yet to be written.</p>
555</div>
556
557<!-------------------------------------------------------------------------- -->
558<div class="doc_subsection">
559 <a name="print-callgraph">Print a call graph</a>
560</div>
561<div class="doc_text">
562 <p>
563 This pass, only available in <code>opt</code>, prints the call graph to
564 standard output in a human-readable form.
565 </p>
566</div>
567
568<!-------------------------------------------------------------------------- -->
569<div class="doc_subsection">
570 <a name="print-callgraph-sccs">Print SCCs of the Call Graph</a>
571</div>
572<div class="doc_text">
573 <p>
574 This pass, only available in <code>opt</code>, prints the SCCs of the call
575 graph to standard output in a human-readable form.
576 </p>
577</div>
578
579<!-------------------------------------------------------------------------- -->
580<div class="doc_subsection">
581 <a name="print-cfg-sccs">Print SCCs of each function CFG</a>
582</div>
583<div class="doc_text">
584 <p>
585 This pass, only available in <code>opt</code>, prints the SCCs of each
586 function CFG to standard output in a human-readable form.
587 </p>
588</div>
589
590<!-------------------------------------------------------------------------- -->
591<div class="doc_subsection">
592 <a name="print-externalfnconstants">Print external fn callsites passed constants</a>
593</div>
594<div class="doc_text">
595 <p>
596 This pass, only available in <code>opt</code>, prints out call sites to
597 external functions that are called with constant arguments. This can be
598 useful when looking for standard library functions we should constant fold
599 or handle in alias analyses.
600 </p>
601</div>
602
603<!-------------------------------------------------------------------------- -->
604<div class="doc_subsection">
605 <a name="print-function">Print function to stderr</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]606</div>
607<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]608 <p>
609 The <code>PrintFunctionPass</code> class is designed to be pipelined with
610 other <code>FunctionPass</code>es, and prints out the functions of the module
611 as they are processed.
612 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]613</div>
614
615<!-------------------------------------------------------------------------- -->
616<div class="doc_subsection">
Duncan Sands3ee8fc92008-09-23 12:47:39 +0000[diff] [blame]617 <a name="print-module">Print module to stderr</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]618</div>
619<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]620 <p>
621 This pass simply prints out the entire module when it is executed.
622 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]623</div>
624
625<!-------------------------------------------------------------------------- -->
626<div class="doc_subsection">
Duncan Sands3ee8fc92008-09-23 12:47:39 +0000[diff] [blame]627 <a name="print-used-types">Find Used Types</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]628</div>
629<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]630 <p>
631 This pass is used to seek out all of the types in use by the program. Note
632 that this analysis explicitly does not include types only used by the symbol
633 table.
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]634</div>
635
636<!-------------------------------------------------------------------------- -->
637<div class="doc_subsection">
638 <a name="profile-loader">Load profile information from llvmprof.out</a>
639</div>
640<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]641 <p>
642 A concrete implementation of profiling information that loads the information
643 from a profile dump file.
644 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]645</div>
646
647<!-------------------------------------------------------------------------- -->
648<div class="doc_subsection">
649 <a name="scalar-evolution">Scalar Evolution Analysis</a>
650</div>
651<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]652 <p>
653 The <code>ScalarEvolution</code> analysis can be used to analyze and
654 catagorize scalar expressions in loops. It specializes in recognizing general
655 induction variables, representing them with the abstract and opaque
656 <code>SCEV</code> class. Given this analysis, trip counts of loops and other
657 important properties can be obtained.
658 </p>
659
660 <p>
661 This analysis is primarily useful for induction variable substitution and
662 strength reduction.
663 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]664</div>
665
666<!-------------------------------------------------------------------------- -->
667<div class="doc_subsection">
668 <a name="targetdata">Target Data Layout</a>
669</div>
670<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]671 <p>Provides other passes access to information on how the size and alignment
672 required by the the target ABI for various data types.</p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]673</div>
674
675<!-- ======================================================================= -->
676<div class="doc_section"> <a name="transform">Transform Passes</a></div>
677<div class="doc_text">
678 <p>This section describes the LLVM Transform Passes.</p>
679</div>
680
681<!-------------------------------------------------------------------------- -->
682<div class="doc_subsection">
683 <a name="adce">Aggressive Dead Code Elimination</a>
684</div>
685<div class="doc_text">
Reid Spenceraf4af3a2007-03-27 02:49:31 +0000[diff] [blame]686 <p>ADCE aggressively tries to eliminate code. This pass is similar to
687 <a href="#dce">DCE</a> but it assumes that values are dead until proven
688 otherwise. This is similar to <a href="#sccp">SCCP</a>, except applied to
689 the liveness of values.</p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]690</div>
691
692<!-------------------------------------------------------------------------- -->
693<div class="doc_subsection">
694 <a name="argpromotion">Promote 'by reference' arguments to scalars</a>
695</div>
696<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]697 <p>
698 This pass promotes "by reference" arguments to be "by value" arguments. In
699 practice, this means looking for internal functions that have pointer
700 arguments. If it can prove, through the use of alias analysis, that an
701 argument is *only* loaded, then it can pass the value into the function
702 instead of the address of the value. This can cause recursive simplification
703 of code and lead to the elimination of allocas (especially in C++ template
704 code like the STL).
705 </p>
706
707 <p>
708 This pass also handles aggregate arguments that are passed into a function,
709 scalarizing them if the elements of the aggregate are only loaded. Note that
710 it refuses to scalarize aggregates which would require passing in more than
711 three operands to the function, because passing thousands of operands for a
712 large array or structure is unprofitable!
713 </p>
714
715 <p>
716 Note that this transformation could also be done for arguments that are only
717 stored to (returning the value instead), but does not currently. This case
718 would be best handled when and if LLVM starts supporting multiple return
719 values from functions.
720 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]721</div>
722
723<!-------------------------------------------------------------------------- -->
724<div class="doc_subsection">
725 <a name="block-placement">Profile Guided Basic Block Placement</a>
726</div>
727<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]728 <p>This pass is a very simple profile guided basic block placement algorithm.
729 The idea is to put frequently executed blocks together at the start of the
730 function and hopefully increase the number of fall-through conditional
731 branches. If there is no profile information for a particular function, this
732 pass basically orders blocks in depth-first order.</p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]733</div>
734
735<!-------------------------------------------------------------------------- -->
736<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]737 <a name="break-crit-edges">Break critical edges in CFG</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]738</div>
739<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]740 <p>
741 Break all of the critical edges in the CFG by inserting a dummy basic block.
742 It may be "required" by passes that cannot deal with critical edges. This
743 transformation obviously invalidates the CFG, but can update forward dominator
744 (set, immediate dominators, tree, and frontier) information.
745 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]746</div>
747
748<!-------------------------------------------------------------------------- -->
749<div class="doc_subsection">
Gordon Henriksena8a118b2008-05-08 17:46:35 +0000[diff] [blame]750 <a name="codegenprepare">Prepare a function for code generation</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]751</div>
752<div class="doc_text">
Gordon Henriksena8a118b2008-05-08 17:46:35 +0000[diff] [blame]753 This pass munges the code in the input function to better prepare it for
754 SelectionDAG-based code generation. This works around limitations in it's
755 basic-block-at-a-time approach. It should eventually be removed.
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]756</div>
757
758<!-------------------------------------------------------------------------- -->
759<div class="doc_subsection">
760 <a name="condprop">Conditional Propagation</a>
761</div>
762<div class="doc_text">
Reid Spenceraf4af3a2007-03-27 02:49:31 +0000[diff] [blame]763 <p>This pass propagates information about conditional expressions through the
764 program, allowing it to eliminate conditional branches in some cases.</p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]765</div>
766
767<!-------------------------------------------------------------------------- -->
768<div class="doc_subsection">
769 <a name="constmerge">Merge Duplicate Global Constants</a>
770</div>
771<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]772 <p>
773 Merges duplicate global constants together into a single constant that is
774 shared. This is useful because some passes (ie TraceValues) insert a lot of
775 string constants into the program, regardless of whether or not an existing
776 string is available.
777 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]778</div>
779
780<!-------------------------------------------------------------------------- -->
781<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]782 <a name="constprop">Simple constant propagation</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]783</div>
784<div class="doc_text">
Reid Spenceraf4af3a2007-03-27 02:49:31 +0000[diff] [blame]785 <p>This file implements constant propagation and merging. It looks for
786 instructions involving only constant operands and replaces them with a
Gordon Henriksenddaa61d2007-10-25 08:58:56 +0000[diff] [blame]787 constant value instead of an instruction. For example:</p>
788 <blockquote><pre>add i32 1, 2</pre></blockquote>
789 <p>becomes</p>
790 <blockquote><pre>i32 3</pre></blockquote>
Reid Spenceraf4af3a2007-03-27 02:49:31 +0000[diff] [blame]791 <p>NOTE: this pass has a habit of making definitions be dead. It is a good
792 idea to to run a <a href="#die">DIE</a> (Dead Instruction Elimination) pass
793 sometime after running this pass.</p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]794</div>
795
796<!-------------------------------------------------------------------------- -->
797<div class="doc_subsection">
798 <a name="dce">Dead Code Elimination</a>
799</div>
800<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]801 <p>
802 Dead code elimination is similar to <a href="#die">dead instruction
803 elimination</a>, but it rechecks instructions that were used by removed
804 instructions to see if they are newly dead.
805 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]806</div>
807
808<!-------------------------------------------------------------------------- -->
809<div class="doc_subsection">
810 <a name="deadargelim">Dead Argument Elimination</a>
811</div>
812<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]813 <p>
814 This pass deletes dead arguments from internal functions. Dead argument
815 elimination removes arguments which are directly dead, as well as arguments
816 only passed into function calls as dead arguments of other functions. This
817 pass also deletes dead arguments in a similar way.
818 </p>
819
820 <p>
821 This pass is often useful as a cleanup pass to run after aggressive
822 interprocedural passes, which add possibly-dead arguments.
823 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]824</div>
825
826<!-------------------------------------------------------------------------- -->
827<div class="doc_subsection">
828 <a name="deadtypeelim">Dead Type Elimination</a>
829</div>
830<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]831 <p>
832 This pass is used to cleanup the output of GCC. It eliminate names for types
833 that are unused in the entire translation unit, using the <a
834 href="#findusedtypes">find used types</a> pass.
835 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]836</div>
837
838<!-------------------------------------------------------------------------- -->
839<div class="doc_subsection">
840 <a name="die">Dead Instruction Elimination</a>
841</div>
842<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]843 <p>
844 Dead instruction elimination performs a single pass over the function,
845 removing instructions that are obviously dead.
846 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]847</div>
848
849<!-------------------------------------------------------------------------- -->
850<div class="doc_subsection">
851 <a name="dse">Dead Store Elimination</a>
852</div>
853<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]854 <p>
855 A trivial dead store elimination that only considers basic-block local
856 redundant stores.
857 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]858</div>
859
860<!-------------------------------------------------------------------------- -->
861<div class="doc_subsection">
862 <a name="gcse">Global Common Subexpression Elimination</a>
863</div>
864<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]865 <p>
866 This pass is designed to be a very quick global transformation that
867 eliminates global common subexpressions from a function. It does this by
Matthijs Kooijman845f5242008-06-05 07:55:49 +0000[diff] [blame]868 using an existing value numbering analysis pass to identify the common
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]869 subexpressions, eliminating them when possible.
870 </p>
Matthijs Kooijman845f5242008-06-05 07:55:49 +0000[diff] [blame]871 <p>
872 This pass is deprecated by the <a href="#gvn">Global Value Numbering pass</a>
873 (which does a better job with its own value numbering).
874 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]875</div>
876
877<!-------------------------------------------------------------------------- -->
878<div class="doc_subsection">
879 <a name="globaldce">Dead Global Elimination</a>
880</div>
881<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]882 <p>
883 This transform is designed to eliminate unreachable internal globals from the
884 program. It uses an aggressive algorithm, searching out globals that are
885 known to be alive. After it finds all of the globals which are needed, it
886 deletes whatever is left over. This allows it to delete recursive chunks of
887 the program which are unreachable.
888 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]889</div>
890
891<!-------------------------------------------------------------------------- -->
892<div class="doc_subsection">
893 <a name="globalopt">Global Variable Optimizer</a>
894</div>
895<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]896 <p>
897 This pass transforms simple global variables that never have their address
898 taken. If obviously true, it marks read/write globals as constant, deletes
899 variables only stored to, etc.
900 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]901</div>
902
903<!-------------------------------------------------------------------------- -->
904<div class="doc_subsection">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]905 <a name="gvn">Global Value Numbering</a>
906</div>
907<div class="doc_text">
908 <p>
909 This pass performs global value numbering to eliminate fully redundant
910 instructions. It also performs simple dead load elimination.
911 </p>
Matthijs Kooijman845f5242008-06-05 07:55:49 +0000[diff] [blame]912 <p>
913 Note that this pass does the value numbering itself, it does not use the
914 ValueNumbering analysis passes.
915 </p>
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]916</div>
917
918<!-------------------------------------------------------------------------- -->
919<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]920 <a name="gvnpre">Global Value Numbering/Partial Redundancy Elimination</a>
921</div>
922<div class="doc_text">
Gordon Henriksen0e15dc22007-10-25 10:18:27 +0000[diff] [blame]923 <p>
924 This pass performs a hybrid of global value numbering and partial redundancy
925 elimination, known as GVN-PRE. It performs partial redundancy elimination on
926 values, rather than lexical expressions, allowing a more comprehensive view
927 the optimization. It replaces redundant values with uses of earlier
928 occurences of the same value. While this is beneficial in that it eliminates
929 unneeded computation, it also increases register pressure by creating large
930 live ranges, and should be used with caution on platforms that are very
931 sensitive to register pressure.
932 </p>
Matthijs Kooijman845f5242008-06-05 07:55:49 +0000[diff] [blame]933 <p>
934 Note that this pass does the value numbering itself, it does not use the
935 ValueNumbering analysis passes.
936 </p>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]937</div>
938
939<!-------------------------------------------------------------------------- -->
940<div class="doc_subsection">
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]941 <a name="indmemrem">Indirect Malloc and Free Removal</a>
942</div>
943<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]944 <p>
945 This pass finds places where memory allocation functions may escape into
946 indirect land. Some transforms are much easier (aka possible) only if free
947 or malloc are not called indirectly.
948 </p>
949
950 <p>
951 Thus find places where the address of memory functions are taken and construct
952 bounce functions with direct calls of those functions.
953 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]954</div>
955
956<!-------------------------------------------------------------------------- -->
957<div class="doc_subsection">
958 <a name="indvars">Canonicalize Induction Variables</a>
959</div>
960<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]961 <p>
962 This transformation analyzes and transforms the induction variables (and
963 computations derived from them) into simpler forms suitable for subsequent
964 analysis and transformation.
965 </p>
966
967 <p>
968 This transformation makes the following changes to each loop with an
969 identifiable induction variable:
970 </p>
971
972 <ol>
973 <li>All loops are transformed to have a <em>single</em> canonical
974 induction variable which starts at zero and steps by one.</li>
975 <li>The canonical induction variable is guaranteed to be the first PHI node
976 in the loop header block.</li>
977 <li>Any pointer arithmetic recurrences are raised to use array
978 subscripts.</li>
979 </ol>
980
981 <p>
982 If the trip count of a loop is computable, this pass also makes the following
983 changes:
984 </p>
985
986 <ol>
987 <li>The exit condition for the loop is canonicalized to compare the
988 induction value against the exit value. This turns loops like:
989 <blockquote><pre>for (i = 7; i*i < 1000; ++i)</pre></blockquote>
990 into
991 <blockquote><pre>for (i = 0; i != 25; ++i)</pre></blockquote></li>
992 <li>Any use outside of the loop of an expression derived from the indvar
993 is changed to compute the derived value outside of the loop, eliminating
994 the dependence on the exit value of the induction variable. If the only
995 purpose of the loop is to compute the exit value of some derived
996 expression, this transformation will make the loop dead.</li>
Gordon Henriksene626bbe2007-11-04 16:17:00 +0000[diff] [blame]997 </ol>
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]998
999 <p>
1000 This transformation should be followed by strength reduction after all of the
1001 desired loop transformations have been performed. Additionally, on targets
1002 where it is profitable, the loop could be transformed to count down to zero
1003 (the "do loop" optimization).
1004 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1005</div>
1006
1007<!-------------------------------------------------------------------------- -->
1008<div class="doc_subsection">
1009 <a name="inline">Function Integration/Inlining</a>
1010</div>
1011<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]1012 <p>
1013 Bottom-up inlining of functions into callees.
1014 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1015</div>
1016
1017<!-------------------------------------------------------------------------- -->
1018<div class="doc_subsection">
1019 <a name="insert-block-profiling">Insert instrumentation for block profiling</a>
1020</div>
1021<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]1022 <p>
1023 This pass instruments the specified program with counters for basic block
1024 profiling, which counts the number of times each basic block executes. This
1025 is the most basic form of profiling, which can tell which blocks are hot, but
1026 cannot reliably detect hot paths through the CFG.
1027 </p>
1028
1029 <p>
1030 Note that this implementation is very naïve. Control equivalent regions of
1031 the CFG should not require duplicate counters, but it does put duplicate
1032 counters in.
1033 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1034</div>
1035
1036<!-------------------------------------------------------------------------- -->
1037<div class="doc_subsection">
1038 <a name="insert-edge-profiling">Insert instrumentation for edge profiling</a>
1039</div>
1040<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]1041 <p>
1042 This pass instruments the specified program with counters for edge profiling.
1043 Edge profiling can give a reasonable approximation of the hot paths through a
1044 program, and is used for a wide variety of program transformations.
1045 </p>
1046
1047 <p>
1048 Note that this implementation is very naïve. It inserts a counter for
1049 <em>every</em> edge in the program, instead of using control flow information
1050 to prune the number of counters inserted.
1051 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1052</div>
1053
1054<!-------------------------------------------------------------------------- -->
1055<div class="doc_subsection">
1056 <a name="insert-function-profiling">Insert instrumentation for function profiling</a>
1057</div>
1058<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]1059 <p>
1060 This pass instruments the specified program with counters for function
1061 profiling, which counts the number of times each function is called.
1062 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1063</div>
1064
1065<!-------------------------------------------------------------------------- -->
1066<div class="doc_subsection">
1067 <a name="insert-null-profiling-rs">Measure profiling framework overhead</a>
1068</div>
1069<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]1070 <p>
1071 The basic profiler that does nothing. It is the default profiler and thus
1072 terminates <code>RSProfiler</code> chains. It is useful for measuring
1073 framework overhead.
1074 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1075</div>
1076
1077<!-------------------------------------------------------------------------- -->
1078<div class="doc_subsection">
1079 <a name="insert-rs-profiling-framework">Insert random sampling instrumentation framework</a>
1080</div>
1081<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]1082 <p>
1083 The second stage of the random-sampling instrumentation framework, duplicates
1084 all instructions in a function, ignoring the profiling code, then connects the
1085 two versions together at the entry and at backedges. At each connection point
1086 a choice is made as to whether to jump to the profiled code (take a sample) or
1087 execute the unprofiled code.
1088 </p>
1089
1090 <p>
1091 After this pass, it is highly recommended to run<a href="#mem2reg">mem2reg</a>
1092 and <a href="#adce">adce</a>. <a href="#instcombine">instcombine</a>,
1093 <a href="#load-vn">load-vn</a>, <a href="#gdce">gdce</a>, and
1094 <a href="#dse">dse</a> also are good to run afterwards.
1095 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1096</div>
1097
1098<!-------------------------------------------------------------------------- -->
1099<div class="doc_subsection">
1100 <a name="instcombine">Combine redundant instructions</a>
1101</div>
1102<div class="doc_text">
Gordon Henriksen55cbec32007-10-26 03:03:51 +0000[diff] [blame]1103 <p>
1104 Combine instructions to form fewer, simple
1105 instructions. This pass does not modify the CFG This pass is where algebraic
1106 simplification happens.
1107 </p>
1108
1109 <p>
1110 This pass combines things like:
1111 </p>
1112
1113<blockquote><pre
1114>%Y = add i32 %X, 1
1115%Z = add i32 %Y, 1</pre></blockquote>
1116
1117 <p>
1118 into:
1119 </p>
1120
1121<blockquote><pre
1122>%Z = add i32 %X, 2</pre></blockquote>
1123
1124 <p>
1125 This is a simple worklist driven algorithm.
1126 </p>
1127
1128 <p>
1129 This pass guarantees that the following canonicalizations are performed on
1130 the program:
1131 </p>
1132
1133 <ul>
1134 <li>If a binary operator has a constant operand, it is moved to the right-
1135 hand side.</li>
1136 <li>Bitwise operators with constant operands are always grouped so that
1137 shifts are performed first, then <code>or</code>s, then
1138 <code>and</code>s, then <code>xor</code>s.</li>
1139 <li>Compare instructions are converted from <code>&lt;</code>,
1140 <code>&gt;</code>, <code>≤</code>, or <code>≥</code> to
1141 <code>=</code> or <code>≠</code> if possible.</li>
1142 <li>All <code>cmp</code> instructions on boolean values are replaced with
1143 logical operations.</li>
1144 <li><code>add <var>X</var>, <var>X</var></code> is represented as
1145 <code>mul <var>X</var>, 2</code> ⇒ <code>shl <var>X</var>, 1</code></li>
1146 <li>Multiplies with a constant power-of-two argument are transformed into
1147 shifts.</li>
1148 <li>… etc.</li>
1149 </ul>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1150</div>
1151
1152<!-------------------------------------------------------------------------- -->
1153<div class="doc_subsection">
1154 <a name="internalize">Internalize Global Symbols</a>
1155</div>
1156<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1157 <p>
1158 This pass loops over all of the functions in the input module, looking for a
1159 main function. If a main function is found, all other functions and all
1160 global variables with initializers are marked as internal.
1161 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1162</div>
1163
1164<!-------------------------------------------------------------------------- -->
1165<div class="doc_subsection">
1166 <a name="ipconstprop">Interprocedural constant propagation</a>
1167</div>
1168<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1169 <p>
1170 This pass implements an <em>extremely</em> simple interprocedural constant
1171 propagation pass. It could certainly be improved in many different ways,
1172 like using a worklist. This pass makes arguments dead, but does not remove
1173 them. The existing dead argument elimination pass should be run after this
1174 to clean up the mess.
1175 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1176</div>
1177
1178<!-------------------------------------------------------------------------- -->
1179<div class="doc_subsection">
1180 <a name="ipsccp">Interprocedural Sparse Conditional Constant Propagation</a>
1181</div>
1182<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1183 <p>
1184 An interprocedural variant of <a href="#sccp">Sparse Conditional Constant
1185 Propagation</a>.
1186 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1187</div>
1188
1189<!-------------------------------------------------------------------------- -->
1190<div class="doc_subsection">
Gordon Henriksena8a118b2008-05-08 17:46:35 +0000[diff] [blame]1191 <a name="jump-threading">Thread control through conditional blocks</a>
1192</div>
1193<div class="doc_text">
1194 <p>
1195 Jump threading tries to find distinct threads of control flow running through
1196 a basic block. This pass looks at blocks that have multiple predecessors and
1197 multiple successors. If one or more of the predecessors of the block can be
1198 proven to always cause a jump to one of the successors, we forward the edge
1199 from the predecessor to the successor by duplicating the contents of this
1200 block.
1201 </p>
1202 <p>
1203 An example of when this can occur is code like this:
1204 </p>
1205
1206 <pre
1207>if () { ...
1208 X = 4;
1209}
1210if (X &lt; 3) {</pre>
1211
1212 <p>
1213 In this case, the unconditional branch at the end of the first if can be
1214 revectored to the false side of the second if.
1215 </p>
1216</div>
1217
1218<!-------------------------------------------------------------------------- -->
1219<div class="doc_subsection">
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1220 <a name="lcssa">Loop-Closed SSA Form Pass</a>
1221</div>
1222<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1223 <p>
1224 This pass transforms loops by placing phi nodes at the end of the loops for
1225 all values that are live across the loop boundary. For example, it turns
1226 the left into the right code:
1227 </p>
1228
1229 <pre
1230>for (...) for (...)
1231 if (c) if (c)
1232 X1 = ... X1 = ...
1233 else else
1234 X2 = ... X2 = ...
1235 X3 = phi(X1, X2) X3 = phi(X1, X2)
1236... = X3 + 4 X4 = phi(X3)
1237 ... = X4 + 4</pre>
1238
1239 <p>
1240 This is still valid LLVM; the extra phi nodes are purely redundant, and will
1241 be trivially eliminated by <code>InstCombine</code>. The major benefit of
1242 this transformation is that it makes many other loop optimizations, such as
1243 LoopUnswitching, simpler.
1244 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1245</div>
1246
1247<!-------------------------------------------------------------------------- -->
1248<div class="doc_subsection">
1249 <a name="licm">Loop Invariant Code Motion</a>
1250</div>
1251<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1252 <p>
1253 This pass performs loop invariant code motion, attempting to remove as much
1254 code from the body of a loop as possible. It does this by either hoisting
1255 code into the preheader block, or by sinking code to the exit blocks if it is
1256 safe. This pass also promotes must-aliased memory locations in the loop to
1257 live in registers, thus hoisting and sinking "invariant" loads and stores.
1258 </p>
1259
1260 <p>
1261 This pass uses alias analysis for two purposes:
1262 </p>
1263
1264 <ul>
1265 <li>Moving loop invariant loads and calls out of loops. If we can determine
1266 that a load or call inside of a loop never aliases anything stored to,
1267 we can hoist it or sink it like any other instruction.</li>
1268 <li>Scalar Promotion of Memory - If there is a store instruction inside of
1269 the loop, we try to move the store to happen AFTER the loop instead of
1270 inside of the loop. This can only happen if a few conditions are true:
1271 <ul>
1272 <li>The pointer stored through is loop invariant.</li>
1273 <li>There are no stores or loads in the loop which <em>may</em> alias
1274 the pointer. There are no calls in the loop which mod/ref the
1275 pointer.</li>
1276 </ul>
1277 If these conditions are true, we can promote the loads and stores in the
1278 loop of the pointer to use a temporary alloca'd variable. We then use
1279 the mem2reg functionality to construct the appropriate SSA form for the
1280 variable.</li>
1281 </ul>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1282</div>
Gordon Henriksena8a118b2008-05-08 17:46:35 +0000[diff] [blame]1283<!-------------------------------------------------------------------------- -->
1284<div class="doc_subsection">
1285 <a name="loop-deletion">Dead Loop Deletion Pass</a>
1286</div>
1287<div class="doc_text">
1288 <p>
1289 This file implements the Dead Loop Deletion Pass. This pass is responsible
1290 for eliminating loops with non-infinite computable trip counts that have no
1291 side effects or volatile instructions, and do not contribute to the
1292 computation of the function's return value.
1293 </p>
1294</div>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1295
1296<!-------------------------------------------------------------------------- -->
1297<div class="doc_subsection">
1298 <a name="loop-extract">Extract loops into new functions</a>
1299</div>
1300<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1301 <p>
1302 A pass wrapper around the <code>ExtractLoop()</code> scalar transformation to
1303 extract each top-level loop into its own new function. If the loop is the
1304 <em>only</em> loop in a given function, it is not touched. This is a pass most
1305 useful for debugging via bugpoint.
1306 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1307</div>
1308
1309<!-------------------------------------------------------------------------- -->
1310<div class="doc_subsection">
1311 <a name="loop-extract-single">Extract at most one loop into a new function</a>
1312</div>
1313<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1314 <p>
1315 Similar to <a href="#loop-extract">Extract loops into new functions</a>,
1316 this pass extracts one natural loop from the program into a function if it
1317 can. This is used by bugpoint.
1318 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1319</div>
1320
1321<!-------------------------------------------------------------------------- -->
1322<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1323 <a name="loop-index-split">Index Split Loops</a>
1324</div>
1325<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1326 <p>
1327 This pass divides loop's iteration range by spliting loop such that each
1328 individual loop is executed efficiently.
1329 </p>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1330</div>
1331
1332<!-------------------------------------------------------------------------- -->
1333<div class="doc_subsection">
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1334 <a name="loop-reduce">Loop Strength Reduction</a>
1335</div>
1336<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1337 <p>
1338 This pass performs a strength reduction on array references inside loops that
1339 have as one or more of their components the loop induction variable. This is
1340 accomplished by creating a new value to hold the initial value of the array
1341 access for the first iteration, and then creating a new GEP instruction in
1342 the loop to increment the value by the appropriate amount.
1343 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1344</div>
1345
1346<!-------------------------------------------------------------------------- -->
1347<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1348 <a name="loop-rotate">Rotate Loops</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1349</div>
1350<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1351 <p>A simple loop rotation transformation.</p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1352</div>
1353
1354<!-------------------------------------------------------------------------- -->
1355<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1356 <a name="loop-unroll">Unroll loops</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1357</div>
1358<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1359 <p>
1360 This pass implements a simple loop unroller. It works best when loops have
1361 been canonicalized by the <a href="#indvars"><tt>-indvars</tt></a> pass,
1362 allowing it to determine the trip counts of loops easily.
1363 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1364</div>
1365
1366<!-------------------------------------------------------------------------- -->
1367<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1368 <a name="loop-unswitch">Unswitch loops</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1369</div>
1370<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1371 <p>
1372 This pass transforms loops that contain branches on loop-invariant conditions
1373 to have multiple loops. For example, it turns the left into the right code:
1374 </p>
1375
1376 <pre
1377>for (...) if (lic)
1378 A for (...)
1379 if (lic) A; B; C
1380 B else
1381 C for (...)
1382 A; C</pre>
1383
1384 <p>
1385 This can increase the size of the code exponentially (doubling it every time
1386 a loop is unswitched) so we only unswitch if the resultant code will be
1387 smaller than a threshold.
1388 </p>
1389
1390 <p>
1391 This pass expects LICM to be run before it to hoist invariant conditions out
1392 of the loop, to make the unswitching opportunity obvious.
1393 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1394</div>
1395
1396<!-------------------------------------------------------------------------- -->
1397<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1398 <a name="loopsimplify">Canonicalize natural loops</a>
1399</div>
1400<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1401 <p>
1402 This pass performs several transformations to transform natural loops into a
1403 simpler form, which makes subsequent analyses and transformations simpler and
1404 more effective.
1405 </p>
1406
1407 <p>
1408 Loop pre-header insertion guarantees that there is a single, non-critical
1409 entry edge from outside of the loop to the loop header. This simplifies a
1410 number of analyses and transformations, such as LICM.
1411 </p>
1412
1413 <p>
1414 Loop exit-block insertion guarantees that all exit blocks from the loop
1415 (blocks which are outside of the loop that have predecessors inside of the
1416 loop) only have predecessors from inside of the loop (and are thus dominated
1417 by the loop header). This simplifies transformations such as store-sinking
1418 that are built into LICM.
1419 </p>
1420
1421 <p>
1422 This pass also guarantees that loops will have exactly one backedge.
1423 </p>
1424
1425 <p>
1426 Note that the simplifycfg pass will clean up blocks which are split out but
1427 end up being unnecessary, so usage of this pass should not pessimize
1428 generated code.
1429 </p>
1430
1431 <p>
1432 This pass obviously modifies the CFG, but updates loop information and
1433 dominator information.
1434 </p>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1435</div>
1436
1437<!-------------------------------------------------------------------------- -->
1438<div class="doc_subsection">
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1439 <a name="lowerallocs">Lower allocations from instructions to calls</a>
1440</div>
1441<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1442 <p>
1443 Turn <tt>malloc</tt> and <tt>free</tt> instructions into <tt>@malloc</tt> and
1444 <tt>@free</tt> calls.
1445 </p>
1446
1447 <p>
1448 This is a target-dependent tranformation because it depends on the size of
1449 data types and alignment constraints.
1450 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1451</div>
1452
1453<!-------------------------------------------------------------------------- -->
1454<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1455 <a name="lowerinvoke">Lower invoke and unwind, for unwindless code generators</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1456</div>
1457<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1458 <p>
1459 This transformation is designed for use by code generators which do not yet
1460 support stack unwinding. This pass supports two models of exception handling
1461 lowering, the 'cheap' support and the 'expensive' support.
1462 </p>
1463
1464 <p>
1465 'Cheap' exception handling support gives the program the ability to execute
1466 any program which does not "throw an exception", by turning 'invoke'
1467 instructions into calls and by turning 'unwind' instructions into calls to
1468 abort(). If the program does dynamically use the unwind instruction, the
1469 program will print a message then abort.
1470 </p>
1471
1472 <p>
1473 'Expensive' exception handling support gives the full exception handling
1474 support to the program at the cost of making the 'invoke' instruction
1475 really expensive. It basically inserts setjmp/longjmp calls to emulate the
1476 exception handling as necessary.
1477 </p>
1478
1479 <p>
1480 Because the 'expensive' support slows down programs a lot, and EH is only
1481 used for a subset of the programs, it must be specifically enabled by the
1482 <tt>-enable-correct-eh-support</tt> option.
1483 </p>
1484
1485 <p>
1486 Note that after this pass runs the CFG is not entirely accurate (exceptional
1487 control flow edges are not correct anymore) so only very simple things should
1488 be done after the lowerinvoke pass has run (like generation of native code).
1489 This should not be used as a general purpose "my LLVM-to-LLVM pass doesn't
1490 support the invoke instruction yet" lowering pass.
1491 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1492</div>
1493
1494<!-------------------------------------------------------------------------- -->
1495<div class="doc_subsection">
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1496 <a name="lowersetjmp">Lower Set Jump</a>
1497</div>
1498<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1499 <p>
1500 Lowers <tt>setjmp</tt> and <tt>longjmp</tt> to use the LLVM invoke and unwind
1501 instructions as necessary.
1502 </p>
1503
1504 <p>
1505 Lowering of <tt>longjmp</tt> is fairly trivial. We replace the call with a
1506 call to the LLVM library function <tt>__llvm_sjljeh_throw_longjmp()</tt>.
1507 This unwinds the stack for us calling all of the destructors for
1508 objects allocated on the stack.
1509 </p>
1510
1511 <p>
1512 At a <tt>setjmp</tt> call, the basic block is split and the <tt>setjmp</tt>
1513 removed. The calls in a function that have a <tt>setjmp</tt> are converted to
1514 invoke where the except part checks to see if it's a <tt>longjmp</tt>
1515 exception and, if so, if it's handled in the function. If it is, then it gets
1516 the value returned by the <tt>longjmp</tt> and goes to where the basic block
1517 was split. <tt>invoke</tt> instructions are handled in a similar fashion with
1518 the original except block being executed if it isn't a <tt>longjmp</tt>
1519 except that is handled by that function.
1520 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1521</div>
1522
1523<!-------------------------------------------------------------------------- -->
1524<div class="doc_subsection">
1525 <a name="lowerswitch">Lower SwitchInst's to branches</a>
1526</div>
1527<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1528 <p>
1529 Rewrites <tt>switch</tt> instructions with a sequence of branches, which
1530 allows targets to get away with not implementing the switch instruction until
1531 it is convenient.
1532 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1533</div>
1534
1535<!-------------------------------------------------------------------------- -->
1536<div class="doc_subsection">
1537 <a name="mem2reg">Promote Memory to Register</a>
1538</div>
1539<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1540 <p>
1541 This file promotes memory references to be register references. It promotes
1542 <tt>alloca</tt> instructions which only have <tt>load</tt>s and
1543 <tt>store</tt>s as uses. An <tt>alloca</tt> is transformed by using dominator
1544 frontiers to place <tt>phi</tt> nodes, then traversing the function in
1545 depth-first order to rewrite <tt>load</tt>s and <tt>store</tt>s as
1546 appropriate. This is just the standard SSA construction algorithm to construct
1547 "pruned" SSA form.
1548 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1549</div>
1550
1551<!-------------------------------------------------------------------------- -->
1552<div class="doc_subsection">
Gordon Henriksena8a118b2008-05-08 17:46:35 +0000[diff] [blame]1553 <a name="memcpyopt">Optimize use of memcpy and friend</a>
1554</div>
1555<div class="doc_text">
1556 <p>
1557 This pass performs various transformations related to eliminating memcpy
1558 calls, or transforming sets of stores into memset's.
1559 </p>
1560</div>
1561
1562<!-------------------------------------------------------------------------- -->
1563<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1564 <a name="mergereturn">Unify function exit nodes</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1565</div>
1566<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1567 <p>
1568 Ensure that functions have at most one <tt>ret</tt> instruction in them.
1569 Additionally, it keeps track of which node is the new exit node of the CFG.
1570 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1571</div>
1572
1573<!-------------------------------------------------------------------------- -->
1574<div class="doc_subsection">
1575 <a name="predsimplify">Predicate Simplifier</a>
1576</div>
1577<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1578 <p>
1579 Path-sensitive optimizer. In a branch where <tt>x == y</tt>, replace uses of
1580 <tt>x</tt> with <tt>y</tt>. Permits further optimization, such as the
1581 elimination of the unreachable call:
1582 </p>
1583
1584<blockquote><pre
1585>void test(int *p, int *q)
1586{
1587 if (p != q)
1588 return;
1589
1590 if (*p != *q)
1591 foo(); // unreachable
1592}</pre></blockquote>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1593</div>
1594
1595<!-------------------------------------------------------------------------- -->
1596<div class="doc_subsection">
1597 <a name="prune-eh">Remove unused exception handling info</a>
1598</div>
1599<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1600 <p>
1601 This file implements a simple interprocedural pass which walks the call-graph,
1602 turning <tt>invoke</tt> instructions into <tt>call</tt> instructions if and
1603 only if the callee cannot throw an exception. It implements this as a
1604 bottom-up traversal of the call-graph.
1605 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1606</div>
1607
1608<!-------------------------------------------------------------------------- -->
1609<div class="doc_subsection">
1610 <a name="raiseallocs">Raise allocations from calls to instructions</a>
1611</div>
1612<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1613 <p>
1614 Converts <tt>@malloc</tt> and <tt>@free</tt> calls to <tt>malloc</tt> and
1615 <tt>free</tt> instructions.
1616 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1617</div>
1618
1619<!-------------------------------------------------------------------------- -->
1620<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1621 <a name="reassociate">Reassociate expressions</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1622</div>
1623<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1624 <p>
1625 This pass reassociates commutative expressions in an order that is designed
1626 to promote better constant propagation, GCSE, LICM, PRE, etc.
1627 </p>
1628
1629 <p>
1630 For example: 4 + (<var>x</var> + 5) ⇒ <var>x</var> + (4 + 5)
1631 </p>
1632
1633 <p>
1634 In the implementation of this algorithm, constants are assigned rank = 0,
1635 function arguments are rank = 1, and other values are assigned ranks
1636 corresponding to the reverse post order traversal of current function
1637 (starting at 2), which effectively gives values in deep loops higher rank
1638 than values not in loops.
1639 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1640</div>
1641
1642<!-------------------------------------------------------------------------- -->
1643<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1644 <a name="reg2mem">Demote all values to stack slots</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1645</div>
1646<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1647 <p>
1648 This file demotes all registers to memory references. It is intented to be
1649 the inverse of <a href="#mem2reg"><tt>-mem2reg</tt></a>. By converting to
1650 <tt>load</tt> instructions, the only values live accross basic blocks are
1651 <tt>alloca</tt> instructions and <tt>load</tt> instructions before
1652 <tt>phi</tt> nodes. It is intended that this should make CFG hacking much
1653 easier. To make later hacking easier, the entry block is split into two, such
1654 that all introduced <tt>alloca</tt> instructions (and nothing else) are in the
1655 entry block.
1656 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1657</div>
1658
1659<!-------------------------------------------------------------------------- -->
1660<div class="doc_subsection">
1661 <a name="scalarrepl">Scalar Replacement of Aggregates</a>
1662</div>
1663<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1664 <p>
1665 The well-known scalar replacement of aggregates transformation. This
1666 transform breaks up <tt>alloca</tt> instructions of aggregate type (structure
1667 or array) into individual <tt>alloca</tt> instructions for each member if
1668 possible. Then, if possible, it transforms the individual <tt>alloca</tt>
1669 instructions into nice clean scalar SSA form.
1670 </p>
1671
1672 <p>
1673 This combines a simple scalar replacement of aggregates algorithm with the <a
1674 href="#mem2reg"><tt>mem2reg</tt></a> algorithm because often interact,
1675 especially for C++ programs. As such, iterating between <tt>scalarrepl</tt>,
1676 then <a href="#mem2reg"><tt>mem2reg</tt></a> until we run out of things to
1677 promote works well.
1678 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1679</div>
1680
1681<!-------------------------------------------------------------------------- -->
1682<div class="doc_subsection">
1683 <a name="sccp">Sparse Conditional Constant Propagation</a>
1684</div>
1685<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1686 <p>
1687 Sparse conditional constant propagation and merging, which can be summarized
1688 as:
1689 </p>
1690
1691 <ol>
1692 <li>Assumes values are constant unless proven otherwise</li>
1693 <li>Assumes BasicBlocks are dead unless proven otherwise</li>
1694 <li>Proves values to be constant, and replaces them with constants</li>
1695 <li>Proves conditional branches to be unconditional</li>
1696 </ol>
1697
1698 <p>
1699 Note that this pass has a habit of making definitions be dead. It is a good
1700 idea to to run a DCE pass sometime after running this pass.
1701 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1702</div>
1703
1704<!-------------------------------------------------------------------------- -->
1705<div class="doc_subsection">
1706 <a name="simplify-libcalls">Simplify well-known library calls</a>
1707</div>
1708<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1709 <p>
1710 Applies a variety of small optimizations for calls to specific well-known
1711 function calls (e.g. runtime library functions). For example, a call
1712 <tt>exit(3)</tt> that occurs within the <tt>main()</tt> function can be
1713 transformed into simply <tt>return 3</tt>.
1714 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1715</div>
1716
1717<!-------------------------------------------------------------------------- -->
1718<div class="doc_subsection">
1719 <a name="simplifycfg">Simplify the CFG</a>
1720</div>
1721<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1722 <p>
1723 Performs dead code elimination and basic block merging. Specifically:
1724 </p>
1725
1726 <ol>
1727 <li>Removes basic blocks with no predecessors.</li>
1728 <li>Merges a basic block into its predecessor if there is only one and the
1729 predecessor only has one successor.</li>
1730 <li>Eliminates PHI nodes for basic blocks with a single predecessor.</li>
1731 <li>Eliminates a basic block that only contains an unconditional
1732 branch.</li>
1733 </ol>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1734</div>
1735
1736<!-------------------------------------------------------------------------- -->
1737<div class="doc_subsection">
1738 <a name="strip">Strip all symbols from a module</a>
1739</div>
1740<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1741 <p>
1742 Performs code stripping. This transformation can delete:
1743 </p>
1744
1745 <ol>
1746 <li>names for virtual registers</li>
1747 <li>symbols for internal globals and functions</li>
1748 <li>debug information</li>
1749 </ol>
1750
1751 <p>
1752 Note that this transformation makes code much less readable, so it should
1753 only be used in situations where the <tt>strip</tt> utility would be used,
1754 such as reducing code size or making it harder to reverse engineer code.
1755 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1756</div>
1757
1758<!-------------------------------------------------------------------------- -->
1759<div class="doc_subsection">
Gordon Henriksena8a118b2008-05-08 17:46:35 +0000[diff] [blame]1760 <a name="strip-dead-prototypes">Remove unused function declarations</a>
1761</div>
1762<div class="doc_text">
1763 <p>
1764 This pass loops over all of the functions in the input module, looking for
1765 dead declarations and removes them. Dead declarations are declarations of
1766 functions for which no implementation is available (i.e., declarations for
1767 unused library functions).
1768 </p>
1769</div>
1770
1771<!-------------------------------------------------------------------------- -->
1772<div class="doc_subsection">
1773 <a name="sretpromotion">Promote sret arguments</a>
1774</div>
1775<div class="doc_text">
1776 <p>
1777 This pass finds functions that return a struct (using a pointer to the struct
1778 as the first argument of the function, marked with the '<tt>sret</tt>' attribute) and
1779 replaces them with a new function that simply returns each of the elements of
1780 that struct (using multiple return values).
1781 </p>
1782
1783 <p>
1784 This pass works under a number of conditions:
1785 </p>
1786
1787 <ul>
1788 <li>The returned struct must not contain other structs</li>
1789 <li>The returned struct must only be used to load values from</li>
1790 <li>The placeholder struct passed in is the result of an <tt>alloca</tt></li>
1791 </ul>
1792</div>
1793
1794<!-------------------------------------------------------------------------- -->
1795<div class="doc_subsection">
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1796 <a name="tailcallelim">Tail Call Elimination</a>
1797</div>
1798<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1799 <p>
1800 This file transforms calls of the current function (self recursion) followed
1801 by a return instruction with a branch to the entry of the function, creating
1802 a loop. This pass also implements the following extensions to the basic
1803 algorithm:
1804 </p>
1805
1806 <ul>
1807 <li>Trivial instructions between the call and return do not prevent the
1808 transformation from taking place, though currently the analysis cannot
1809 support moving any really useful instructions (only dead ones).
1810 <li>This pass transforms functions that are prevented from being tail
1811 recursive by an associative expression to use an accumulator variable,
1812 thus compiling the typical naive factorial or <tt>fib</tt> implementation
1813 into efficient code.
1814 <li>TRE is performed if the function returns void, if the return
1815 returns the result returned by the call, or if the function returns a
1816 run-time constant on all exits from the function. It is possible, though
1817 unlikely, that the return returns something else (like constant 0), and
1818 can still be TRE'd. It can be TRE'd if <em>all other</em> return
1819 instructions in the function return the exact same value.
1820 <li>If it can prove that callees do not access theier caller stack frame,
1821 they are marked as eligible for tail call elimination (by the code
1822 generator).
1823 </ul>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1824</div>
1825
1826<!-------------------------------------------------------------------------- -->
1827<div class="doc_subsection">
1828 <a name="tailduplicate">Tail Duplication</a>
1829</div>
1830<div class="doc_text">
Gordon Henriksenc86b6772007-11-04 16:15:04 +0000[diff] [blame]1831 <p>
1832 This pass performs a limited form of tail duplication, intended to simplify
1833 CFGs by removing some unconditional branches. This pass is necessary to
1834 straighten out loops created by the C front-end, but also is capable of
1835 making other code nicer. After this pass is run, the CFG simplify pass
1836 should be run to clean up the mess.
1837 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1838</div>
1839
1840<!-- ======================================================================= -->
1841<div class="doc_section"> <a name="transform">Utility Passes</a></div>
1842<div class="doc_text">
1843 <p>This section describes the LLVM Utility Passes.</p>
1844</div>
1845
1846<!-------------------------------------------------------------------------- -->
1847<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1848 <a name="deadarghaX0r">Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1849</div>
1850<div class="doc_text">
Gordon Henriksen75ff18e2007-11-04 18:10:18 +0000[diff] [blame]1851 <p>
1852 Same as dead argument elimination, but deletes arguments to functions which
1853 are external. This is only for use by <a
1854 href="Bugpoint.html">bugpoint</a>.</p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1855</div>
1856
1857<!-------------------------------------------------------------------------- -->
1858<div class="doc_subsection">
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1859 <a name="extract-blocks">Extract Basic Blocks From Module (for bugpoint use)</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1860</div>
1861<div class="doc_text">
Gordon Henriksen75ff18e2007-11-04 18:10:18 +0000[diff] [blame]1862 <p>
1863 This pass is used by bugpoint to extract all blocks from the module into their
1864 own functions.</p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1865</div>
1866
1867<!-------------------------------------------------------------------------- -->
1868<div class="doc_subsection">
Gordon Henriksen90a52142007-11-05 02:05:35 +0000[diff] [blame]1869 <a name="preverify">Preliminary module verification</a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1870</div>
1871<div class="doc_text">
Gordon Henriksen90a52142007-11-05 02:05:35 +0000[diff] [blame]1872 <p>
1873 Ensures that the module is in the form required by the <a
1874 href="#verifier">Module Verifier</a> pass.
1875 </p>
1876
1877 <p>
1878 Running the verifier runs this pass automatically, so there should be no need
1879 to use it directly.
1880 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1881</div>
1882
1883<!-------------------------------------------------------------------------- -->
1884<div class="doc_subsection">
1885 <a name="verify">Module Verifier</a>
1886</div>
1887<div class="doc_text">
Gordon Henriksen75ff18e2007-11-04 18:10:18 +0000[diff] [blame]1888 <p>
1889 Verifies an LLVM IR code. This is useful to run after an optimization which is
1890 undergoing testing. Note that <tt>llvm-as</tt> verifies its input before
1891 emitting bitcode, and also that malformed bitcode is likely to make LLVM
1892 crash. All language front-ends are therefore encouraged to verify their output
1893 before performing optimizing transformations.
1894 </p>
1895
Gordon Henriksen23a8ce52007-11-04 18:14:08 +0000[diff] [blame]1896 <ul>
Gordon Henriksen75ff18e2007-11-04 18:10:18 +0000[diff] [blame]1897 <li>Both of a binary operator's parameters are of the same type.</li>
1898 <li>Verify that the indices of mem access instructions match other
1899 operands.</li>
1900 <li>Verify that arithmetic and other things are only performed on
1901 first-class types. Verify that shifts and logicals only happen on
1902 integrals f.e.</li>
1903 <li>All of the constants in a switch statement are of the correct type.</li>
1904 <li>The code is in valid SSA form.</li>
1905 <li>It should be illegal to put a label into any other type (like a
1906 structure) or to return one. [except constant arrays!]</li>
Nick Lewycky0c78ac12008-03-28 06:46:51 +0000[diff] [blame]1907 <li>Only phi nodes can be self referential: <tt>%x = add i32 %x, %x</tt> is
Gordon Henriksen873390e2007-11-04 18:17:58 +0000[diff] [blame]1908 invalid.</li>
Gordon Henriksen75ff18e2007-11-04 18:10:18 +0000[diff] [blame]1909 <li>PHI nodes must have an entry for each predecessor, with no extras.</li>
1910 <li>PHI nodes must be the first thing in a basic block, all grouped
1911 together.</li>
1912 <li>PHI nodes must have at least one entry.</li>
1913 <li>All basic blocks should only end with terminator insts, not contain
1914 them.</li>
1915 <li>The entry node to a function must not have predecessors.</li>
1916 <li>All Instructions must be embedded into a basic block.</li>
1917 <li>Functions cannot take a void-typed parameter.</li>
1918 <li>Verify that a function's argument list agrees with its declared
1919 type.</li>
1920 <li>It is illegal to specify a name for a void value.</li>
1921 <li>It is illegal to have a internal global value with no initializer.</li>
1922 <li>It is illegal to have a ret instruction that returns a value that does
1923 not agree with the function return value type.</li>
1924 <li>Function call argument types match the function prototype.</li>
1925 <li>All other things that are tested by asserts spread about the code.</li>
Gordon Henriksen23a8ce52007-11-04 18:14:08 +0000[diff] [blame]1926 </ul>
Gordon Henriksen75ff18e2007-11-04 18:10:18 +0000[diff] [blame]1927
1928 <p>
1929 Note that this does not provide full security verification (like Java), but
1930 instead just tries to ensure that code is well-formed.
1931 </p>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1932</div>
1933
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1934<!-------------------------------------------------------------------------- -->
1935<div class="doc_subsection">
1936 <a name="view-cfg">View CFG of function</a>
1937</div>
1938<div class="doc_text">
Gordon Henriksen75ff18e2007-11-04 18:10:18 +0000[diff] [blame]1939 <p>
1940 Displays the control flow graph using the GraphViz tool.
1941 </p>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1942</div>
1943
1944<!-------------------------------------------------------------------------- -->
1945<div class="doc_subsection">
1946 <a name="view-cfg-only">View CFG of function (with no function bodies)</a>
1947</div>
1948<div class="doc_text">
Gordon Henriksen75ff18e2007-11-04 18:10:18 +0000[diff] [blame]1949 <p>
1950 Displays the control flow graph using the GraphViz tool, but omitting function
1951 bodies.
1952 </p>
Gordon Henriksen1f5cce02007-10-25 08:46:12 +0000[diff] [blame]1953</div>
1954
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1955<!-- *********************************************************************** -->
1956
1957<hr>
1958<address>
1959 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
Misha Brukman44408702008-12-11 17:34:48 +0000[diff] [blame]1960 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1961 <a href="http://validator.w3.org/check/referer"><img
Misha Brukman44408702008-12-11 17:34:48 +0000[diff] [blame]1962 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
Reid Spencerd9aac122007-03-26 09:32:31 +0000[diff] [blame]1963
1964 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a><br>
1965 <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
1966 Last modified: $Date$
1967</address>
1968
1969</body>
1970</html>