1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM's Analysis and Transform Passes</title>
6 <link rel="stylesheet" href="_static/llvm.css" type="text/css">
7 <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
13 If Passes.html is up to date, the following "one-liner" should print
16 egrep -e '^<tr><td><a href="#.*">-.*</a></td><td>.*</td></tr>$' \
17 -e '^ <a name=".*">.*</a>$' < Passes.html >html; \
18 perl >help <<'EOT' && diff -u help html; rm -f help html
19 open HTML, "<Passes.html" or die "open: Passes.html: $!\n";
21 m:^<tr><td><a href="#(.*)">-.*</a></td><td>.*</td></tr>$: or next;
22 $order{$1} = sprintf("%03d", 1 + int %order);
24 open HELP, "../Release/bin/opt -help|" or die "open: opt -help: $!\n";
26 m:^ -([^ ]+) +- (.*)$: or next;
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\">-$1: $2</a>\n";
32 @x = map { s/^\d\d\d//; $_ } sort @x;
33 @y = map { s/^\d\d\d//; $_ } sort @y;
37 This (real) one-liner can also be helpful when converting comments to HTML:
39 perl -e '$/ = undef; for (split(/\n/, <>)) { s:^ *///? ?::; print " <p>\n" if !$on && $_ =~ /\S/; print " </p>\n" if $on && $_ =~ /^\s*$/; print " $_\n"; $on = ($_ =~ /\S/); } print " </p>\n" if $on'
43 <h1>LLVM's Analysis and Transform Passes</h1>
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>
52 <div class="doc_author">
53 <p>Written by <a href="mailto:rspencer@x10sys.com">Reid Spencer</a>
54 and Gordon Henriksen</p>
57 <!-- ======================================================================= -->
58 <h2><a name="intro">Introduction</a></h2>
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.
63 The table below divides the passes that LLVM provides into three categories.
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.
67 Utility passes provides some utility but don't otherwise fit categorization.
68 For example passes to extract functions to bitcode or write a module to
69 bitcode are neither analysis nor transform passes.
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>
74 <tr><th colspan="2"><b>ANALYSIS PASSES</b></th></tr>
75 <tr><th>Option</th><th>Name</th></tr>
76 <tr><td><a href="#aa-eval">-aa-eval</a></td><td>Exhaustive Alias Analysis Precision Evaluator</td></tr>
77 <tr><td><a href="#basicaa">-basicaa</a></td><td>Basic Alias Analysis (stateless AA impl)</td></tr>
78 <tr><td><a href="#basiccg">-basiccg</a></td><td>Basic CallGraph Construction</td></tr>
79 <tr><td><a href="#count-aa">-count-aa</a></td><td>Count Alias Analysis Query Responses</td></tr>
80 <tr><td><a href="#debug-aa">-debug-aa</a></td><td>AA use debugger</td></tr>
81 <tr><td><a href="#domfrontier">-domfrontier</a></td><td>Dominance Frontier Construction</td></tr>
82 <tr><td><a href="#domtree">-domtree</a></td><td>Dominator Tree Construction</td></tr>
83 <tr><td><a href="#dot-callgraph">-dot-callgraph</a></td><td>Print Call Graph to 'dot' file</td></tr>
84 <tr><td><a href="#dot-cfg">-dot-cfg</a></td><td>Print CFG of function to 'dot' file</td></tr>
85 <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>
86 <tr><td><a href="#dot-dom">-dot-dom</a></td><td>Print dominance tree of function to 'dot' file</td></tr>
87 <tr><td><a href="#dot-dom-only">-dot-dom-only</a></td><td>Print dominance tree of function to 'dot' file (with no function bodies)</td></tr>
88 <tr><td><a href="#dot-postdom">-dot-postdom</a></td><td>Print postdominance tree of function to 'dot' file</td></tr>
89 <tr><td><a href="#dot-postdom-only">-dot-postdom-only</a></td><td>Print postdominance tree of function to 'dot' file (with no function bodies)</td></tr>
90 <tr><td><a href="#globalsmodref-aa">-globalsmodref-aa</a></td><td>Simple mod/ref analysis for globals</td></tr>
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="#iv-users">-iv-users</a></td><td>Induction Variable Users</td></tr>
94 <tr><td><a href="#lazy-value-info">-lazy-value-info</a></td><td>Lazy Value Information Analysis</td></tr>
95 <tr><td><a href="#lda">-lda</a></td><td>Loop Dependence Analysis</td></tr>
96 <tr><td><a href="#libcall-aa">-libcall-aa</a></td><td>LibCall Alias Analysis</td></tr>
97 <tr><td><a href="#lint">-lint</a></td><td>Statically lint-checks LLVM IR</td></tr>
98 <tr><td><a href="#loops">-loops</a></td><td>Natural Loop Information</td></tr>
99 <tr><td><a href="#memdep">-memdep</a></td><td>Memory Dependence Analysis</td></tr>
100 <tr><td><a href="#module-debuginfo">-module-debuginfo</a></td><td>Decodes module-level debug info</td></tr>
101 <tr><td><a href="#no-aa">-no-aa</a></td><td>No Alias Analysis (always returns 'may' alias)</td></tr>
102 <tr><td><a href="#no-profile">-no-profile</a></td><td>No Profile Information</td></tr>
103 <tr><td><a href="#postdomfrontier">-postdomfrontier</a></td><td>Post-Dominance Frontier Construction</td></tr>
104 <tr><td><a href="#postdomtree">-postdomtree</a></td><td>Post-Dominator Tree Construction</td></tr>
105 <tr><td><a href="#print-alias-sets">-print-alias-sets</a></td><td>Alias Set Printer</td></tr>
106 <tr><td><a href="#print-callgraph">-print-callgraph</a></td><td>Print a call graph</td></tr>
107 <tr><td><a href="#print-callgraph-sccs">-print-callgraph-sccs</a></td><td>Print SCCs of the Call Graph</td></tr>
108 <tr><td><a href="#print-cfg-sccs">-print-cfg-sccs</a></td><td>Print SCCs of each function CFG</td></tr>
109 <tr><td><a href="#print-dbginfo">-print-dbginfo</a></td><td>Print debug info in human readable form</td></tr>
110 <tr><td><a href="#print-dom-info">-print-dom-info</a></td><td>Dominator Info Printer</td></tr>
111 <tr><td><a href="#print-externalfnconstants">-print-externalfnconstants</a></td><td>Print external fn callsites passed constants</td></tr>
112 <tr><td><a href="#print-function">-print-function</a></td><td>Print function to stderr</td></tr>
113 <tr><td><a href="#print-module">-print-module</a></td><td>Print module to stderr</td></tr>
114 <tr><td><a href="#print-used-types">-print-used-types</a></td><td>Find Used Types</td></tr>
115 <tr><td><a href="#profile-estimator">-profile-estimator</a></td><td>Estimate profiling information</td></tr>
116 <tr><td><a href="#profile-loader">-profile-loader</a></td><td>Load profile information from llvmprof.out</td></tr>
117 <tr><td><a href="#profile-verifier">-profile-verifier</a></td><td>Verify profiling information</td></tr>
118 <tr><td><a href="#regions">-regions</a></td><td>Detect single entry single exit regions</td></tr>
119 <tr><td><a href="#scalar-evolution">-scalar-evolution</a></td><td>Scalar Evolution Analysis</td></tr>
120 <tr><td><a href="#scev-aa">-scev-aa</a></td><td>ScalarEvolution-based Alias Analysis</td></tr>
121 <tr><td><a href="#targetdata">-targetdata</a></td><td>Target Data Layout</td></tr>
124 <tr><th colspan="2"><b>TRANSFORM PASSES</b></th></tr>
125 <tr><th>Option</th><th>Name</th></tr>
126 <tr><td><a href="#adce">-adce</a></td><td>Aggressive Dead Code Elimination</td></tr>
127 <tr><td><a href="#always-inline">-always-inline</a></td><td>Inliner for always_inline functions</td></tr>
128 <tr><td><a href="#argpromotion">-argpromotion</a></td><td>Promote 'by reference' arguments to scalars</td></tr>
129 <tr><td><a href="#bb-vectorize">-bb-vectorize</a></td><td>Combine instructions to form vector instructions within basic blocks</td></tr>
130 <tr><td><a href="#block-placement">-block-placement</a></td><td>Profile Guided Basic Block Placement</td></tr>
131 <tr><td><a href="#break-crit-edges">-break-crit-edges</a></td><td>Break critical edges in CFG</td></tr>
132 <tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Optimize for code generation</td></tr>
133 <tr><td><a href="#constmerge">-constmerge</a></td><td>Merge Duplicate Global Constants</td></tr>
134 <tr><td><a href="#constprop">-constprop</a></td><td>Simple constant propagation</td></tr>
135 <tr><td><a href="#dce">-dce</a></td><td>Dead Code Elimination</td></tr>
136 <tr><td><a href="#deadargelim">-deadargelim</a></td><td>Dead Argument Elimination</td></tr>
137 <tr><td><a href="#deadtypeelim">-deadtypeelim</a></td><td>Dead Type Elimination</td></tr>
138 <tr><td><a href="#die">-die</a></td><td>Dead Instruction Elimination</td></tr>
139 <tr><td><a href="#dse">-dse</a></td><td>Dead Store Elimination</td></tr>
140 <tr><td><a href="#functionattrs">-functionattrs</a></td><td>Deduce function attributes</td></tr>
141 <tr><td><a href="#globaldce">-globaldce</a></td><td>Dead Global Elimination</td></tr>
142 <tr><td><a href="#globalopt">-globalopt</a></td><td>Global Variable Optimizer</td></tr>
143 <tr><td><a href="#gvn">-gvn</a></td><td>Global Value Numbering</td></tr>
144 <tr><td><a href="#indvars">-indvars</a></td><td>Canonicalize Induction Variables</td></tr>
145 <tr><td><a href="#inline">-inline</a></td><td>Function Integration/Inlining</td></tr>
146 <tr><td><a href="#insert-edge-profiling">-insert-edge-profiling</a></td><td>Insert instrumentation for edge profiling</td></tr>
147 <tr><td><a href="#insert-optimal-edge-profiling">-insert-optimal-edge-profiling</a></td><td>Insert optimal instrumentation for edge profiling</td></tr>
148 <tr><td><a href="#instcombine">-instcombine</a></td><td>Combine redundant instructions</td></tr>
149 <tr><td><a href="#internalize">-internalize</a></td><td>Internalize Global Symbols</td></tr>
150 <tr><td><a href="#ipconstprop">-ipconstprop</a></td><td>Interprocedural constant propagation</td></tr>
151 <tr><td><a href="#ipsccp">-ipsccp</a></td><td>Interprocedural Sparse Conditional Constant Propagation</td></tr>
152 <tr><td><a href="#jump-threading">-jump-threading</a></td><td>Jump Threading</td></tr>
153 <tr><td><a href="#lcssa">-lcssa</a></td><td>Loop-Closed SSA Form Pass</td></tr>
154 <tr><td><a href="#licm">-licm</a></td><td>Loop Invariant Code Motion</td></tr>
155 <tr><td><a href="#loop-deletion">-loop-deletion</a></td><td>Delete dead loops</td></tr>
156 <tr><td><a href="#loop-extract">-loop-extract</a></td><td>Extract loops into new functions</td></tr>
157 <tr><td><a href="#loop-extract-single">-loop-extract-single</a></td><td>Extract at most one loop into a new function</td></tr>
158 <tr><td><a href="#loop-reduce">-loop-reduce</a></td><td>Loop Strength Reduction</td></tr>
159 <tr><td><a href="#loop-rotate">-loop-rotate</a></td><td>Rotate Loops</td></tr>
160 <tr><td><a href="#loop-simplify">-loop-simplify</a></td><td>Canonicalize natural loops</td></tr>
161 <tr><td><a href="#loop-unroll">-loop-unroll</a></td><td>Unroll loops</td></tr>
162 <tr><td><a href="#loop-unswitch">-loop-unswitch</a></td><td>Unswitch loops</td></tr>
163 <tr><td><a href="#loweratomic">-loweratomic</a></td><td>Lower atomic intrinsics to non-atomic form</td></tr>
164 <tr><td><a href="#lowerinvoke">-lowerinvoke</a></td><td>Lower invoke and unwind, for unwindless code generators</td></tr>
165 <tr><td><a href="#lowerswitch">-lowerswitch</a></td><td>Lower SwitchInst's to branches</td></tr>
166 <tr><td><a href="#mem2reg">-mem2reg</a></td><td>Promote Memory to Register</td></tr>
167 <tr><td><a href="#memcpyopt">-memcpyopt</a></td><td>MemCpy Optimization</td></tr>
168 <tr><td><a href="#mergefunc">-mergefunc</a></td><td>Merge Functions</td></tr>
169 <tr><td><a href="#mergereturn">-mergereturn</a></td><td>Unify function exit nodes</td></tr>
170 <tr><td><a href="#partial-inliner">-partial-inliner</a></td><td>Partial Inliner</td></tr>
171 <tr><td><a href="#prune-eh">-prune-eh</a></td><td>Remove unused exception handling info</td></tr>
172 <tr><td><a href="#reassociate">-reassociate</a></td><td>Reassociate expressions</td></tr>
173 <tr><td><a href="#reg2mem">-reg2mem</a></td><td>Demote all values to stack slots</td></tr>
174 <tr><td><a href="#scalarrepl">-scalarrepl</a></td><td>Scalar Replacement of Aggregates (DT)</td></tr>
175 <tr><td><a href="#sccp">-sccp</a></td><td>Sparse Conditional Constant Propagation</td></tr>
176 <tr><td><a href="#simplify-libcalls">-simplify-libcalls</a></td><td>Simplify well-known library calls</td></tr>
177 <tr><td><a href="#simplifycfg">-simplifycfg</a></td><td>Simplify the CFG</td></tr>
178 <tr><td><a href="#sink">-sink</a></td><td>Code sinking</td></tr>
179 <tr><td><a href="#sretpromotion">-sretpromotion</a></td><td>Promote sret arguments to multiple ret values</td></tr>
180 <tr><td><a href="#strip">-strip</a></td><td>Strip all symbols from a module</td></tr>
181 <tr><td><a href="#strip-dead-debug-info">-strip-dead-debug-info</a></td><td>Strip debug info for unused symbols</td></tr>
182 <tr><td><a href="#strip-dead-prototypes">-strip-dead-prototypes</a></td><td>Strip Unused Function Prototypes</td></tr>
183 <tr><td><a href="#strip-debug-declare">-strip-debug-declare</a></td><td>Strip all llvm.dbg.declare intrinsics</td></tr>
184 <tr><td><a href="#strip-nondebug">-strip-nondebug</a></td><td>Strip all symbols, except dbg symbols, from a module</td></tr>
185 <tr><td><a href="#tailcallelim">-tailcallelim</a></td><td>Tail Call Elimination</td></tr>
186 <tr><td><a href="#tailduplicate">-tailduplicate</a></td><td>Tail Duplication</td></tr>
189 <tr><th colspan="2"><b>UTILITY PASSES</b></th></tr>
190 <tr><th>Option</th><th>Name</th></tr>
191 <tr><td><a href="#deadarghaX0r">-deadarghaX0r</a></td><td>Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</td></tr>
192 <tr><td><a href="#extract-blocks">-extract-blocks</a></td><td>Extract Basic Blocks From Module (for bugpoint use)</td></tr>
193 <tr><td><a href="#instnamer">-instnamer</a></td><td>Assign names to anonymous instructions</td></tr>
194 <tr><td><a href="#preverify">-preverify</a></td><td>Preliminary module verification</td></tr>
195 <tr><td><a href="#verify">-verify</a></td><td>Module Verifier</td></tr>
196 <tr><td><a href="#view-cfg">-view-cfg</a></td><td>View CFG of function</td></tr>
197 <tr><td><a href="#view-cfg-only">-view-cfg-only</a></td><td>View CFG of function (with no function bodies)</td></tr>
198 <tr><td><a href="#view-dom">-view-dom</a></td><td>View dominance tree of function</td></tr>
199 <tr><td><a href="#view-dom-only">-view-dom-only</a></td><td>View dominance tree of function (with no function bodies)</td></tr>
200 <tr><td><a href="#view-postdom">-view-postdom</a></td><td>View postdominance tree of function</td></tr>
201 <tr><td><a href="#view-postdom-only">-view-postdom-only</a></td><td>View postdominance tree of function (with no function bodies)</td></tr>
206 <!-- ======================================================================= -->
207 <h2><a name="analyses">Analysis Passes</a></h2>
209 <p>This section describes the LLVM Analysis Passes.</p>
211 <!-------------------------------------------------------------------------- -->
213 <a name="aa-eval">-aa-eval: Exhaustive Alias Analysis Precision Evaluator</a>
216 <p>This is a simple N^2 alias analysis accuracy evaluator.
217 Basically, for each function in the program, it simply queries to see how the
218 alias analysis implementation answers alias queries between each pair of
219 pointers in the function.</p>
221 <p>This is inspired and adapted from code by: Naveen Neelakantam, Francesco
222 Spadini, and Wojciech Stryjewski.</p>
225 <!-------------------------------------------------------------------------- -->
227 <a name="basicaa">-basicaa: Basic Alias Analysis (stateless AA impl)</a>
230 <p>A basic alias analysis pass that implements identities (two different
231 globals cannot alias, etc), but does no stateful analysis.</p>
234 <!-------------------------------------------------------------------------- -->
236 <a name="basiccg">-basiccg: Basic CallGraph Construction</a>
239 <p>Yet to be written.</p>
242 <!-------------------------------------------------------------------------- -->
244 <a name="count-aa">-count-aa: Count Alias Analysis Query Responses</a>
248 A pass which can be used to count how many alias queries
249 are being made and how the alias analysis implementation being used responds.
253 <!-------------------------------------------------------------------------- -->
255 <a name="debug-aa">-debug-aa: AA use debugger</a>
259 This simple pass checks alias analysis users to ensure that if they
260 create a new value, they do not query AA without informing it of the value.
261 It acts as a shim over any other AA pass you want.
265 Yes keeping track of every value in the program is expensive, but this is
270 <!-------------------------------------------------------------------------- -->
272 <a name="domfrontier">-domfrontier: Dominance Frontier Construction</a>
276 This pass is a simple dominator construction algorithm for finding forward
281 <!-------------------------------------------------------------------------- -->
283 <a name="domtree">-domtree: Dominator Tree Construction</a>
287 This pass is a simple dominator construction algorithm for finding forward
292 <!-------------------------------------------------------------------------- -->
294 <a name="dot-callgraph">-dot-callgraph: Print Call Graph to 'dot' file</a>
298 This pass, only available in <code>opt</code>, prints the call graph into a
299 <code>.dot</code> graph. This graph can then be processed with the "dot" tool
300 to convert it to postscript or some other suitable format.
304 <!-------------------------------------------------------------------------- -->
306 <a name="dot-cfg">-dot-cfg: Print CFG of function to 'dot' file</a>
310 This pass, only available in <code>opt</code>, prints the control flow graph
311 into a <code>.dot</code> graph. This graph can then be processed with the
312 "dot" tool to convert it to postscript or some other suitable format.
316 <!-------------------------------------------------------------------------- -->
318 <a name="dot-cfg-only">-dot-cfg-only: Print CFG of function to 'dot' file (with no function bodies)</a>
322 This pass, only available in <code>opt</code>, prints the control flow graph
323 into a <code>.dot</code> graph, omitting the function bodies. This graph can
324 then be processed with the "dot" tool to convert it to postscript or some
325 other suitable format.
329 <!-------------------------------------------------------------------------- -->
331 <a name="dot-dom">-dot-dom: Print dominance tree of function to 'dot' file</a>
335 This pass, only available in <code>opt</code>, prints the dominator tree
336 into a <code>.dot</code> graph. This graph can then be processed with the
337 "dot" tool to convert it to postscript or some other suitable format.
341 <!-------------------------------------------------------------------------- -->
343 <a name="dot-dom-only">-dot-dom-only: Print dominance tree of function to 'dot' file (with no function bodies)</a>
347 This pass, only available in <code>opt</code>, prints the dominator tree
348 into a <code>.dot</code> graph, omitting the function bodies. This graph can
349 then be processed with the "dot" tool to convert it to postscript or some
350 other suitable format.
354 <!-------------------------------------------------------------------------- -->
356 <a name="dot-postdom">-dot-postdom: Print postdominance tree of function to 'dot' file</a>
360 This pass, only available in <code>opt</code>, prints the post dominator tree
361 into a <code>.dot</code> graph. This graph can then be processed with the
362 "dot" tool to convert it to postscript or some other suitable format.
366 <!-------------------------------------------------------------------------- -->
368 <a name="dot-postdom-only">-dot-postdom-only: Print postdominance tree of function to 'dot' file (with no function bodies)</a>
372 This pass, only available in <code>opt</code>, prints the post dominator tree
373 into a <code>.dot</code> graph, omitting the function bodies. This graph can
374 then be processed with the "dot" tool to convert it to postscript or some
375 other suitable format.
379 <!-------------------------------------------------------------------------- -->
381 <a name="globalsmodref-aa">-globalsmodref-aa: Simple mod/ref analysis for globals</a>
385 This simple pass provides alias and mod/ref information for global values
386 that do not have their address taken, and keeps track of whether functions
387 read or write memory (are "pure"). For this simple (but very common) case,
388 we can provide pretty accurate and useful information.
392 <!-------------------------------------------------------------------------- -->
394 <a name="instcount">-instcount: Counts the various types of Instructions</a>
398 This pass collects the count of all instructions and reports them
402 <!-------------------------------------------------------------------------- -->
404 <a name="intervals">-intervals: Interval Partition Construction</a>
408 This analysis calculates and represents the interval partition of a function,
409 or a preexisting interval partition.
413 In this way, the interval partition may be used to reduce a flow graph down
414 to its degenerate single node interval partition (unless it is irreducible).
418 <!-------------------------------------------------------------------------- -->
420 <a name="iv-users">-iv-users: Induction Variable Users</a>
423 <p>Bookkeeping for "interesting" users of expressions computed from
424 induction variables.</p>
427 <!-------------------------------------------------------------------------- -->
429 <a name="lazy-value-info">-lazy-value-info: Lazy Value Information Analysis</a>
432 <p>Interface for lazy computation of value constraint information.</p>
435 <!-------------------------------------------------------------------------- -->
437 <a name="lda">-lda: Loop Dependence Analysis</a>
440 <p>Loop dependence analysis framework, which is used to detect dependences in
441 memory accesses in loops.</p>
444 <!-------------------------------------------------------------------------- -->
446 <a name="libcall-aa">-libcall-aa: LibCall Alias Analysis</a>
449 <p>LibCall Alias Analysis.</p>
452 <!-------------------------------------------------------------------------- -->
454 <a name="lint">-lint: Statically lint-checks LLVM IR</a>
457 <p>This pass statically checks for common and easily-identified constructs
458 which produce undefined or likely unintended behavior in LLVM IR.</p>
460 <p>It is not a guarantee of correctness, in two ways. First, it isn't
461 comprehensive. There are checks which could be done statically which are
462 not yet implemented. Some of these are indicated by TODO comments, but
463 those aren't comprehensive either. Second, many conditions cannot be
464 checked statically. This pass does no dynamic instrumentation, so it
465 can't check for all possible problems.</p>
467 <p>Another limitation is that it assumes all code will be executed. A store
468 through a null pointer in a basic block which is never reached is harmless,
469 but this pass will warn about it anyway.</p>
471 <p>Optimization passes may make conditions that this pass checks for more or
472 less obvious. If an optimization pass appears to be introducing a warning,
473 it may be that the optimization pass is merely exposing an existing
474 condition in the code.</p>
476 <p>This code may be run before instcombine. In many cases, instcombine checks
477 for the same kinds of things and turns instructions with undefined behavior
478 into unreachable (or equivalent). Because of this, this pass makes some
479 effort to look through bitcasts and so on.
483 <!-------------------------------------------------------------------------- -->
485 <a name="loops">-loops: Natural Loop Information</a>
489 This analysis is used to identify natural loops and determine the loop depth
490 of various nodes of the CFG. Note that the loops identified may actually be
491 several natural loops that share the same header node... not just a single
496 <!-------------------------------------------------------------------------- -->
498 <a name="memdep">-memdep: Memory Dependence Analysis</a>
502 An analysis that determines, for a given memory operation, what preceding
503 memory operations it depends on. It builds on alias analysis information, and
504 tries to provide a lazy, caching interface to a common kind of alias
509 <!-------------------------------------------------------------------------- -->
511 <a name="module-debuginfo">-module-debuginfo: Decodes module-level debug info</a>
514 <p>This pass decodes the debug info metadata in a module and prints in a
515 (sufficiently-prepared-) human-readable form.
517 For example, run this pass from opt along with the -analyze option, and
518 it'll print to standard output.
522 <!-------------------------------------------------------------------------- -->
524 <a name="no-aa">-no-aa: No Alias Analysis (always returns 'may' alias)</a>
528 This is the default implementation of the Alias Analysis interface. It always
529 returns "I don't know" for alias queries. NoAA is unlike other alias analysis
530 implementations, in that it does not chain to a previous analysis. As such it
531 doesn't follow many of the rules that other alias analyses must.
535 <!-------------------------------------------------------------------------- -->
537 <a name="no-profile">-no-profile: No Profile Information</a>
541 The default "no profile" implementation of the abstract
542 <code>ProfileInfo</code> interface.
546 <!-------------------------------------------------------------------------- -->
548 <a name="postdomfrontier">-postdomfrontier: Post-Dominance Frontier Construction</a>
552 This pass is a simple post-dominator construction algorithm for finding
553 post-dominator frontiers.
557 <!-------------------------------------------------------------------------- -->
559 <a name="postdomtree">-postdomtree: Post-Dominator Tree Construction</a>
563 This pass is a simple post-dominator construction algorithm for finding
568 <!-------------------------------------------------------------------------- -->
570 <a name="print-alias-sets">-print-alias-sets: Alias Set Printer</a>
573 <p>Yet to be written.</p>
576 <!-------------------------------------------------------------------------- -->
578 <a name="print-callgraph">-print-callgraph: Print a call graph</a>
582 This pass, only available in <code>opt</code>, prints the call graph to
583 standard error in a human-readable form.
587 <!-------------------------------------------------------------------------- -->
589 <a name="print-callgraph-sccs">-print-callgraph-sccs: Print SCCs of the Call Graph</a>
593 This pass, only available in <code>opt</code>, prints the SCCs of the call
594 graph to standard error in a human-readable form.
598 <!-------------------------------------------------------------------------- -->
600 <a name="print-cfg-sccs">-print-cfg-sccs: Print SCCs of each function CFG</a>
604 This pass, only available in <code>opt</code>, prints the SCCs of each
605 function CFG to standard error in a human-readable form.
609 <!-------------------------------------------------------------------------- -->
611 <a name="print-dbginfo">-print-dbginfo: Print debug info in human readable form</a>
614 <p>Pass that prints instructions, and associated debug info:</p>
617 <li>source/line/col information</li>
618 <li>original variable name</li>
619 <li>original type name</li>
623 <!-------------------------------------------------------------------------- -->
625 <a name="print-dom-info">-print-dom-info: Dominator Info Printer</a>
628 <p>Dominator Info Printer.</p>
631 <!-------------------------------------------------------------------------- -->
633 <a name="print-externalfnconstants">-print-externalfnconstants: Print external fn callsites passed constants</a>
637 This pass, only available in <code>opt</code>, prints out call sites to
638 external functions that are called with constant arguments. This can be
639 useful when looking for standard library functions we should constant fold
640 or handle in alias analyses.
644 <!-------------------------------------------------------------------------- -->
646 <a name="print-function">-print-function: Print function to stderr</a>
650 The <code>PrintFunctionPass</code> class is designed to be pipelined with
651 other <code>FunctionPass</code>es, and prints out the functions of the module
652 as they are processed.
656 <!-------------------------------------------------------------------------- -->
658 <a name="print-module">-print-module: Print module to stderr</a>
662 This pass simply prints out the entire module when it is executed.
666 <!-------------------------------------------------------------------------- -->
668 <a name="print-used-types">-print-used-types: Find Used Types</a>
672 This pass is used to seek out all of the types in use by the program. Note
673 that this analysis explicitly does not include types only used by the symbol
677 <!-------------------------------------------------------------------------- -->
679 <a name="profile-estimator">-profile-estimator: Estimate profiling information</a>
682 <p>Profiling information that estimates the profiling information
683 in a very crude and unimaginative way.
687 <!-------------------------------------------------------------------------- -->
689 <a name="profile-loader">-profile-loader: Load profile information from llvmprof.out</a>
693 A concrete implementation of profiling information that loads the information
694 from a profile dump file.
698 <!-------------------------------------------------------------------------- -->
700 <a name="profile-verifier">-profile-verifier: Verify profiling information</a>
703 <p>Pass that checks profiling information for plausibility.</p>
706 <a name="regions">-regions: Detect single entry single exit regions</a>
710 The <code>RegionInfo</code> pass detects single entry single exit regions in a
711 function, where a region is defined as any subgraph that is connected to the
712 remaining graph at only two spots. Furthermore, an hierarchical region tree is
717 <!-------------------------------------------------------------------------- -->
719 <a name="scalar-evolution">-scalar-evolution: Scalar Evolution Analysis</a>
723 The <code>ScalarEvolution</code> analysis can be used to analyze and
724 catagorize scalar expressions in loops. It specializes in recognizing general
725 induction variables, representing them with the abstract and opaque
726 <code>SCEV</code> class. Given this analysis, trip counts of loops and other
727 important properties can be obtained.
731 This analysis is primarily useful for induction variable substitution and
736 <!-------------------------------------------------------------------------- -->
738 <a name="scev-aa">-scev-aa: ScalarEvolution-based Alias Analysis</a>
741 <p>Simple alias analysis implemented in terms of ScalarEvolution queries.
743 This differs from traditional loop dependence analysis in that it tests
744 for dependencies within a single iteration of a loop, rather than
745 dependencies between different iterations.
747 ScalarEvolution has a more complete understanding of pointer arithmetic
748 than BasicAliasAnalysis' collection of ad-hoc analyses.
752 <!-------------------------------------------------------------------------- -->
754 <a name="targetdata">-targetdata: Target Data Layout</a>
757 <p>Provides other passes access to information on how the size and alignment
758 required by the the target ABI for various data types.</p>
763 <!-- ======================================================================= -->
764 <h2><a name="transforms">Transform Passes</a></h2>
766 <p>This section describes the LLVM Transform Passes.</p>
768 <!-------------------------------------------------------------------------- -->
770 <a name="adce">-adce: Aggressive Dead Code Elimination</a>
773 <p>ADCE aggressively tries to eliminate code. This pass is similar to
774 <a href="#dce">DCE</a> but it assumes that values are dead until proven
775 otherwise. This is similar to <a href="#sccp">SCCP</a>, except applied to
776 the liveness of values.</p>
779 <!-------------------------------------------------------------------------- -->
781 <a name="always-inline">-always-inline: Inliner for always_inline functions</a>
784 <p>A custom inliner that handles only functions that are marked as
788 <!-------------------------------------------------------------------------- -->
790 <a name="argpromotion">-argpromotion: Promote 'by reference' arguments to scalars</a>
794 This pass promotes "by reference" arguments to be "by value" arguments. In
795 practice, this means looking for internal functions that have pointer
796 arguments. If it can prove, through the use of alias analysis, that an
797 argument is *only* loaded, then it can pass the value into the function
798 instead of the address of the value. This can cause recursive simplification
799 of code and lead to the elimination of allocas (especially in C++ template
804 This pass also handles aggregate arguments that are passed into a function,
805 scalarizing them if the elements of the aggregate are only loaded. Note that
806 it refuses to scalarize aggregates which would require passing in more than
807 three operands to the function, because passing thousands of operands for a
808 large array or structure is unprofitable!
812 Note that this transformation could also be done for arguments that are only
813 stored to (returning the value instead), but does not currently. This case
814 would be best handled when and if LLVM starts supporting multiple return
815 values from functions.
819 <!-------------------------------------------------------------------------- -->
821 <a name="bb-vectorize">-bb-vectorize: Basic-Block Vectorization</a>
824 <p>This pass combines instructions inside basic blocks to form vector
825 instructions. It iterates over each basic block, attempting to pair
826 compatible instructions, repeating this process until no additional
827 pairs are selected for vectorization. When the outputs of some pair
828 of compatible instructions are used as inputs by some other pair of
829 compatible instructions, those pairs are part of a potential
830 vectorization chain. Instruction pairs are only fused into vector
831 instructions when they are part of a chain longer than some
832 threshold length. Moreover, the pass attempts to find the best
833 possible chain for each pair of compatible instructions. These
834 heuristics are intended to prevent vectorization in cases where
835 it would not yield a performance increase of the resulting code.
839 <!-------------------------------------------------------------------------- -->
841 <a name="block-placement">-block-placement: Profile Guided Basic Block Placement</a>
844 <p>This pass is a very simple profile guided basic block placement algorithm.
845 The idea is to put frequently executed blocks together at the start of the
846 function and hopefully increase the number of fall-through conditional
847 branches. If there is no profile information for a particular function, this
848 pass basically orders blocks in depth-first order.</p>
851 <!-------------------------------------------------------------------------- -->
853 <a name="break-crit-edges">-break-crit-edges: Break critical edges in CFG</a>
857 Break all of the critical edges in the CFG by inserting a dummy basic block.
858 It may be "required" by passes that cannot deal with critical edges. This
859 transformation obviously invalidates the CFG, but can update forward dominator
860 (set, immediate dominators, tree, and frontier) information.
864 <!-------------------------------------------------------------------------- -->
866 <a name="codegenprepare">-codegenprepare: Optimize for code generation</a>
869 This pass munges the code in the input function to better prepare it for
870 SelectionDAG-based code generation. This works around limitations in it's
871 basic-block-at-a-time approach. It should eventually be removed.
874 <!-------------------------------------------------------------------------- -->
876 <a name="constmerge">-constmerge: Merge Duplicate Global Constants</a>
880 Merges duplicate global constants together into a single constant that is
881 shared. This is useful because some passes (ie TraceValues) insert a lot of
882 string constants into the program, regardless of whether or not an existing
887 <!-------------------------------------------------------------------------- -->
889 <a name="constprop">-constprop: Simple constant propagation</a>
892 <p>This file implements constant propagation and merging. It looks for
893 instructions involving only constant operands and replaces them with a
894 constant value instead of an instruction. For example:</p>
895 <blockquote><pre>add i32 1, 2</pre></blockquote>
897 <blockquote><pre>i32 3</pre></blockquote>
898 <p>NOTE: this pass has a habit of making definitions be dead. It is a good
899 idea to to run a <a href="#die">DIE</a> (Dead Instruction Elimination) pass
900 sometime after running this pass.</p>
903 <!-------------------------------------------------------------------------- -->
905 <a name="dce">-dce: Dead Code Elimination</a>
909 Dead code elimination is similar to <a href="#die">dead instruction
910 elimination</a>, but it rechecks instructions that were used by removed
911 instructions to see if they are newly dead.
915 <!-------------------------------------------------------------------------- -->
917 <a name="deadargelim">-deadargelim: Dead Argument Elimination</a>
921 This pass deletes dead arguments from internal functions. Dead argument
922 elimination removes arguments which are directly dead, as well as arguments
923 only passed into function calls as dead arguments of other functions. This
924 pass also deletes dead arguments in a similar way.
928 This pass is often useful as a cleanup pass to run after aggressive
929 interprocedural passes, which add possibly-dead arguments.
933 <!-------------------------------------------------------------------------- -->
935 <a name="deadtypeelim">-deadtypeelim: Dead Type Elimination</a>
939 This pass is used to cleanup the output of GCC. It eliminate names for types
940 that are unused in the entire translation unit, using the <a
941 href="#findusedtypes">find used types</a> pass.
945 <!-------------------------------------------------------------------------- -->
947 <a name="die">-die: Dead Instruction Elimination</a>
951 Dead instruction elimination performs a single pass over the function,
952 removing instructions that are obviously dead.
956 <!-------------------------------------------------------------------------- -->
958 <a name="dse">-dse: Dead Store Elimination</a>
962 A trivial dead store elimination that only considers basic-block local
967 <!-------------------------------------------------------------------------- -->
969 <a name="functionattrs">-functionattrs: Deduce function attributes</a>
972 <p>A simple interprocedural pass which walks the call-graph, looking for
973 functions which do not access or only read non-local memory, and marking them
974 readnone/readonly. In addition, it marks function arguments (of pointer type)
975 'nocapture' if a call to the function does not create any copies of the pointer
976 value that outlive the call. This more or less means that the pointer is only
977 dereferenced, and not returned from the function or stored in a global.
978 This pass is implemented as a bottom-up traversal of the call-graph.
982 <!-------------------------------------------------------------------------- -->
984 <a name="globaldce">-globaldce: Dead Global Elimination</a>
988 This transform is designed to eliminate unreachable internal globals from the
989 program. It uses an aggressive algorithm, searching out globals that are
990 known to be alive. After it finds all of the globals which are needed, it
991 deletes whatever is left over. This allows it to delete recursive chunks of
992 the program which are unreachable.
996 <!-------------------------------------------------------------------------- -->
998 <a name="globalopt">-globalopt: Global Variable Optimizer</a>
1002 This pass transforms simple global variables that never have their address
1003 taken. If obviously true, it marks read/write globals as constant, deletes
1004 variables only stored to, etc.
1008 <!-------------------------------------------------------------------------- -->
1010 <a name="gvn">-gvn: Global Value Numbering</a>
1014 This pass performs global value numbering to eliminate fully and partially
1015 redundant instructions. It also performs redundant load elimination.
1019 <!-------------------------------------------------------------------------- -->
1021 <a name="indvars">-indvars: Canonicalize Induction Variables</a>
1025 This transformation analyzes and transforms the induction variables (and
1026 computations derived from them) into simpler forms suitable for subsequent
1027 analysis and transformation.
1031 This transformation makes the following changes to each loop with an
1032 identifiable induction variable:
1036 <li>All loops are transformed to have a <em>single</em> canonical
1037 induction variable which starts at zero and steps by one.</li>
1038 <li>The canonical induction variable is guaranteed to be the first PHI node
1039 in the loop header block.</li>
1040 <li>Any pointer arithmetic recurrences are raised to use array
1045 If the trip count of a loop is computable, this pass also makes the following
1050 <li>The exit condition for the loop is canonicalized to compare the
1051 induction value against the exit value. This turns loops like:
1052 <blockquote><pre>for (i = 7; i*i < 1000; ++i)</pre></blockquote>
1054 <blockquote><pre>for (i = 0; i != 25; ++i)</pre></blockquote></li>
1055 <li>Any use outside of the loop of an expression derived from the indvar
1056 is changed to compute the derived value outside of the loop, eliminating
1057 the dependence on the exit value of the induction variable. If the only
1058 purpose of the loop is to compute the exit value of some derived
1059 expression, this transformation will make the loop dead.</li>
1063 This transformation should be followed by strength reduction after all of the
1064 desired loop transformations have been performed. Additionally, on targets
1065 where it is profitable, the loop could be transformed to count down to zero
1066 (the "do loop" optimization).
1070 <!-------------------------------------------------------------------------- -->
1072 <a name="inline">-inline: Function Integration/Inlining</a>
1076 Bottom-up inlining of functions into callees.
1080 <!-------------------------------------------------------------------------- -->
1082 <a name="insert-edge-profiling">-insert-edge-profiling: Insert instrumentation for edge profiling</a>
1086 This pass instruments the specified program with counters for edge profiling.
1087 Edge profiling can give a reasonable approximation of the hot paths through a
1088 program, and is used for a wide variety of program transformations.
1092 Note that this implementation is very naïve. It inserts a counter for
1093 <em>every</em> edge in the program, instead of using control flow information
1094 to prune the number of counters inserted.
1098 <!-------------------------------------------------------------------------- -->
1100 <a name="insert-optimal-edge-profiling">-insert-optimal-edge-profiling: Insert optimal instrumentation for edge profiling</a>
1103 <p>This pass instruments the specified program with counters for edge profiling.
1104 Edge profiling can give a reasonable approximation of the hot paths through a
1105 program, and is used for a wide variety of program transformations.
1109 <!-------------------------------------------------------------------------- -->
1111 <a name="instcombine">-instcombine: Combine redundant instructions</a>
1115 Combine instructions to form fewer, simple
1116 instructions. This pass does not modify the CFG This pass is where algebraic
1117 simplification happens.
1121 This pass combines things like:
1126 %Z = add i32 %Y, 1</pre></blockquote>
1133 >%Z = add i32 %X, 2</pre></blockquote>
1136 This is a simple worklist driven algorithm.
1140 This pass guarantees that the following canonicalizations are performed on
1145 <li>If a binary operator has a constant operand, it is moved to the right-
1147 <li>Bitwise operators with constant operands are always grouped so that
1148 shifts are performed first, then <code>or</code>s, then
1149 <code>and</code>s, then <code>xor</code>s.</li>
1150 <li>Compare instructions are converted from <code><</code>,
1151 <code>></code>, <code>≤</code>, or <code>≥</code> to
1152 <code>=</code> or <code>≠</code> if possible.</li>
1153 <li>All <code>cmp</code> instructions on boolean values are replaced with
1154 logical operations.</li>
1155 <li><code>add <var>X</var>, <var>X</var></code> is represented as
1156 <code>mul <var>X</var>, 2</code> ⇒ <code>shl <var>X</var>, 1</code></li>
1157 <li>Multiplies with a constant power-of-two argument are transformed into
1163 <!-------------------------------------------------------------------------- -->
1165 <a name="internalize">-internalize: Internalize Global Symbols</a>
1169 This pass loops over all of the functions in the input module, looking for a
1170 main function. If a main function is found, all other functions and all
1171 global variables with initializers are marked as internal.
1175 <!-------------------------------------------------------------------------- -->
1177 <a name="ipconstprop">-ipconstprop: Interprocedural constant propagation</a>
1181 This pass implements an <em>extremely</em> simple interprocedural constant
1182 propagation pass. It could certainly be improved in many different ways,
1183 like using a worklist. This pass makes arguments dead, but does not remove
1184 them. The existing dead argument elimination pass should be run after this
1185 to clean up the mess.
1189 <!-------------------------------------------------------------------------- -->
1191 <a name="ipsccp">-ipsccp: Interprocedural Sparse Conditional Constant Propagation</a>
1195 An interprocedural variant of <a href="#sccp">Sparse Conditional Constant
1200 <!-------------------------------------------------------------------------- -->
1202 <a name="jump-threading">-jump-threading: Jump Threading</a>
1206 Jump threading tries to find distinct threads of control flow running through
1207 a basic block. This pass looks at blocks that have multiple predecessors and
1208 multiple successors. If one or more of the predecessors of the block can be
1209 proven to always cause a jump to one of the successors, we forward the edge
1210 from the predecessor to the successor by duplicating the contents of this
1214 An example of when this can occur is code like this:
1221 if (X < 3) {</pre>
1224 In this case, the unconditional branch at the end of the first if can be
1225 revectored to the false side of the second if.
1229 <!-------------------------------------------------------------------------- -->
1231 <a name="lcssa">-lcssa: Loop-Closed SSA Form Pass</a>
1235 This pass transforms loops by placing phi nodes at the end of the loops for
1236 all values that are live across the loop boundary. For example, it turns
1237 the left into the right code:
1241 >for (...) for (...)
1246 X3 = phi(X1, X2) X3 = phi(X1, X2)
1247 ... = X3 + 4 X4 = phi(X3)
1251 This is still valid LLVM; the extra phi nodes are purely redundant, and will
1252 be trivially eliminated by <code>InstCombine</code>. The major benefit of
1253 this transformation is that it makes many other loop optimizations, such as
1254 LoopUnswitching, simpler.
1258 <!-------------------------------------------------------------------------- -->
1260 <a name="licm">-licm: Loop Invariant Code Motion</a>
1264 This pass performs loop invariant code motion, attempting to remove as much
1265 code from the body of a loop as possible. It does this by either hoisting
1266 code into the preheader block, or by sinking code to the exit blocks if it is
1267 safe. This pass also promotes must-aliased memory locations in the loop to
1268 live in registers, thus hoisting and sinking "invariant" loads and stores.
1272 This pass uses alias analysis for two purposes:
1276 <li>Moving loop invariant loads and calls out of loops. If we can determine
1277 that a load or call inside of a loop never aliases anything stored to,
1278 we can hoist it or sink it like any other instruction.</li>
1279 <li>Scalar Promotion of Memory - If there is a store instruction inside of
1280 the loop, we try to move the store to happen AFTER the loop instead of
1281 inside of the loop. This can only happen if a few conditions are true:
1283 <li>The pointer stored through is loop invariant.</li>
1284 <li>There are no stores or loads in the loop which <em>may</em> alias
1285 the pointer. There are no calls in the loop which mod/ref the
1288 If these conditions are true, we can promote the loads and stores in the
1289 loop of the pointer to use a temporary alloca'd variable. We then use
1290 the mem2reg functionality to construct the appropriate SSA form for the
1295 <!-------------------------------------------------------------------------- -->
1297 <a name="loop-deletion">-loop-deletion: Delete dead loops</a>
1301 This file implements the Dead Loop Deletion Pass. This pass is responsible
1302 for eliminating loops with non-infinite computable trip counts that have no
1303 side effects or volatile instructions, and do not contribute to the
1304 computation of the function's return value.
1308 <!-------------------------------------------------------------------------- -->
1310 <a name="loop-extract">-loop-extract: Extract loops into new functions</a>
1314 A pass wrapper around the <code>ExtractLoop()</code> scalar transformation to
1315 extract each top-level loop into its own new function. If the loop is the
1316 <em>only</em> loop in a given function, it is not touched. This is a pass most
1317 useful for debugging via bugpoint.
1321 <!-------------------------------------------------------------------------- -->
1323 <a name="loop-extract-single">-loop-extract-single: Extract at most one loop into a new function</a>
1327 Similar to <a href="#loop-extract">Extract loops into new functions</a>,
1328 this pass extracts one natural loop from the program into a function if it
1329 can. This is used by bugpoint.
1333 <!-------------------------------------------------------------------------- -->
1335 <a name="loop-reduce">-loop-reduce: Loop Strength Reduction</a>
1339 This pass performs a strength reduction on array references inside loops that
1340 have as one or more of their components the loop induction variable. This is
1341 accomplished by creating a new value to hold the initial value of the array
1342 access for the first iteration, and then creating a new GEP instruction in
1343 the loop to increment the value by the appropriate amount.
1347 <!-------------------------------------------------------------------------- -->
1349 <a name="loop-rotate">-loop-rotate: Rotate Loops</a>
1352 <p>A simple loop rotation transformation.</p>
1355 <!-------------------------------------------------------------------------- -->
1357 <a name="loop-simplify">-loop-simplify: Canonicalize natural loops</a>
1361 This pass performs several transformations to transform natural loops into a
1362 simpler form, which makes subsequent analyses and transformations simpler and
1367 Loop pre-header insertion guarantees that there is a single, non-critical
1368 entry edge from outside of the loop to the loop header. This simplifies a
1369 number of analyses and transformations, such as LICM.
1373 Loop exit-block insertion guarantees that all exit blocks from the loop
1374 (blocks which are outside of the loop that have predecessors inside of the
1375 loop) only have predecessors from inside of the loop (and are thus dominated
1376 by the loop header). This simplifies transformations such as store-sinking
1377 that are built into LICM.
1381 This pass also guarantees that loops will have exactly one backedge.
1385 Note that the simplifycfg pass will clean up blocks which are split out but
1386 end up being unnecessary, so usage of this pass should not pessimize
1391 This pass obviously modifies the CFG, but updates loop information and
1392 dominator information.
1396 <!-------------------------------------------------------------------------- -->
1398 <a name="loop-unroll">-loop-unroll: Unroll loops</a>
1402 This pass implements a simple loop unroller. It works best when loops have
1403 been canonicalized by the <a href="#indvars"><tt>-indvars</tt></a> pass,
1404 allowing it to determine the trip counts of loops easily.
1408 <!-------------------------------------------------------------------------- -->
1410 <a name="loop-unswitch">-loop-unswitch: Unswitch loops</a>
1414 This pass transforms loops that contain branches on loop-invariant conditions
1415 to have multiple loops. For example, it turns the left into the right code:
1427 This can increase the size of the code exponentially (doubling it every time
1428 a loop is unswitched) so we only unswitch if the resultant code will be
1429 smaller than a threshold.
1433 This pass expects LICM to be run before it to hoist invariant conditions out
1434 of the loop, to make the unswitching opportunity obvious.
1438 <!-------------------------------------------------------------------------- -->
1440 <a name="loweratomic">-loweratomic: Lower atomic intrinsics to non-atomic form</a>
1444 This pass lowers atomic intrinsics to non-atomic form for use in a known
1445 non-preemptible environment.
1449 The pass does not verify that the environment is non-preemptible (in
1450 general this would require knowledge of the entire call graph of the
1451 program including any libraries which may not be available in bitcode form);
1452 it simply lowers every atomic intrinsic.
1456 <!-------------------------------------------------------------------------- -->
1458 <a name="lowerinvoke">-lowerinvoke: Lower invoke and unwind, for unwindless code generators</a>
1462 This transformation is designed for use by code generators which do not yet
1463 support stack unwinding. This pass supports two models of exception handling
1464 lowering, the 'cheap' support and the 'expensive' support.
1468 'Cheap' exception handling support gives the program the ability to execute
1469 any program which does not "throw an exception", by turning 'invoke'
1470 instructions into calls and by turning 'unwind' instructions into calls to
1471 abort(). If the program does dynamically use the unwind instruction, the
1472 program will print a message then abort.
1476 'Expensive' exception handling support gives the full exception handling
1477 support to the program at the cost of making the 'invoke' instruction
1478 really expensive. It basically inserts setjmp/longjmp calls to emulate the
1479 exception handling as necessary.
1483 Because the 'expensive' support slows down programs a lot, and EH is only
1484 used for a subset of the programs, it must be specifically enabled by the
1485 <tt>-enable-correct-eh-support</tt> option.
1489 Note that after this pass runs the CFG is not entirely accurate (exceptional
1490 control flow edges are not correct anymore) so only very simple things should
1491 be done after the lowerinvoke pass has run (like generation of native code).
1492 This should not be used as a general purpose "my LLVM-to-LLVM pass doesn't
1493 support the invoke instruction yet" lowering pass.
1497 <!-------------------------------------------------------------------------- -->
1499 <a name="lowerswitch">-lowerswitch: Lower SwitchInst's to branches</a>
1503 Rewrites <tt>switch</tt> instructions with a sequence of branches, which
1504 allows targets to get away with not implementing the switch instruction until
1509 <!-------------------------------------------------------------------------- -->
1511 <a name="mem2reg">-mem2reg: Promote Memory to Register</a>
1515 This file promotes memory references to be register references. It promotes
1516 <tt>alloca</tt> instructions which only have <tt>load</tt>s and
1517 <tt>store</tt>s as uses. An <tt>alloca</tt> is transformed by using dominator
1518 frontiers to place <tt>phi</tt> nodes, then traversing the function in
1519 depth-first order to rewrite <tt>load</tt>s and <tt>store</tt>s as
1520 appropriate. This is just the standard SSA construction algorithm to construct
1525 <!-------------------------------------------------------------------------- -->
1527 <a name="memcpyopt">-memcpyopt: MemCpy Optimization</a>
1531 This pass performs various transformations related to eliminating memcpy
1532 calls, or transforming sets of stores into memset's.
1536 <!-------------------------------------------------------------------------- -->
1538 <a name="mergefunc">-mergefunc: Merge Functions</a>
1541 <p>This pass looks for equivalent functions that are mergable and folds them.
1543 A hash is computed from the function, based on its type and number of
1546 Once all hashes are computed, we perform an expensive equality comparison
1547 on each function pair. This takes n^2/2 comparisons per bucket, so it's
1548 important that the hash function be high quality. The equality comparison
1549 iterates through each instruction in each basic block.
1551 When a match is found the functions are folded. If both functions are
1552 overridable, we move the functionality into a new internal function and
1553 leave two overridable thunks to it.
1557 <!-------------------------------------------------------------------------- -->
1559 <a name="mergereturn">-mergereturn: Unify function exit nodes</a>
1563 Ensure that functions have at most one <tt>ret</tt> instruction in them.
1564 Additionally, it keeps track of which node is the new exit node of the CFG.
1568 <!-------------------------------------------------------------------------- -->
1570 <a name="partial-inliner">-partial-inliner: Partial Inliner</a>
1573 <p>This pass performs partial inlining, typically by inlining an if
1574 statement that surrounds the body of the function.
1578 <!-------------------------------------------------------------------------- -->
1580 <a name="prune-eh">-prune-eh: Remove unused exception handling info</a>
1584 This file implements a simple interprocedural pass which walks the call-graph,
1585 turning <tt>invoke</tt> instructions into <tt>call</tt> instructions if and
1586 only if the callee cannot throw an exception. It implements this as a
1587 bottom-up traversal of the call-graph.
1591 <!-------------------------------------------------------------------------- -->
1593 <a name="reassociate">-reassociate: Reassociate expressions</a>
1597 This pass reassociates commutative expressions in an order that is designed
1598 to promote better constant propagation, GCSE, LICM, PRE, etc.
1602 For example: 4 + (<var>x</var> + 5) ⇒ <var>x</var> + (4 + 5)
1606 In the implementation of this algorithm, constants are assigned rank = 0,
1607 function arguments are rank = 1, and other values are assigned ranks
1608 corresponding to the reverse post order traversal of current function
1609 (starting at 2), which effectively gives values in deep loops higher rank
1610 than values not in loops.
1614 <!-------------------------------------------------------------------------- -->
1616 <a name="reg2mem">-reg2mem: Demote all values to stack slots</a>
1620 This file demotes all registers to memory references. It is intented to be
1621 the inverse of <a href="#mem2reg"><tt>-mem2reg</tt></a>. By converting to
1622 <tt>load</tt> instructions, the only values live across basic blocks are
1623 <tt>alloca</tt> instructions and <tt>load</tt> instructions before
1624 <tt>phi</tt> nodes. It is intended that this should make CFG hacking much
1625 easier. To make later hacking easier, the entry block is split into two, such
1626 that all introduced <tt>alloca</tt> instructions (and nothing else) are in the
1631 <!-------------------------------------------------------------------------- -->
1633 <a name="scalarrepl">-scalarrepl: Scalar Replacement of Aggregates (DT)</a>
1637 The well-known scalar replacement of aggregates transformation. This
1638 transform breaks up <tt>alloca</tt> instructions of aggregate type (structure
1639 or array) into individual <tt>alloca</tt> instructions for each member if
1640 possible. Then, if possible, it transforms the individual <tt>alloca</tt>
1641 instructions into nice clean scalar SSA form.
1645 This combines a simple scalar replacement of aggregates algorithm with the <a
1646 href="#mem2reg"><tt>mem2reg</tt></a> algorithm because often interact,
1647 especially for C++ programs. As such, iterating between <tt>scalarrepl</tt>,
1648 then <a href="#mem2reg"><tt>mem2reg</tt></a> until we run out of things to
1653 <!-------------------------------------------------------------------------- -->
1655 <a name="sccp">-sccp: Sparse Conditional Constant Propagation</a>
1659 Sparse conditional constant propagation and merging, which can be summarized
1664 <li>Assumes values are constant unless proven otherwise</li>
1665 <li>Assumes BasicBlocks are dead unless proven otherwise</li>
1666 <li>Proves values to be constant, and replaces them with constants</li>
1667 <li>Proves conditional branches to be unconditional</li>
1671 Note that this pass has a habit of making definitions be dead. It is a good
1672 idea to to run a DCE pass sometime after running this pass.
1676 <!-------------------------------------------------------------------------- -->
1678 <a name="simplify-libcalls">-simplify-libcalls: Simplify well-known library calls</a>
1682 Applies a variety of small optimizations for calls to specific well-known
1683 function calls (e.g. runtime library functions). For example, a call
1684 <tt>exit(3)</tt> that occurs within the <tt>main()</tt> function can be
1685 transformed into simply <tt>return 3</tt>.
1689 <!-------------------------------------------------------------------------- -->
1691 <a name="simplifycfg">-simplifycfg: Simplify the CFG</a>
1695 Performs dead code elimination and basic block merging. Specifically:
1699 <li>Removes basic blocks with no predecessors.</li>
1700 <li>Merges a basic block into its predecessor if there is only one and the
1701 predecessor only has one successor.</li>
1702 <li>Eliminates PHI nodes for basic blocks with a single predecessor.</li>
1703 <li>Eliminates a basic block that only contains an unconditional
1708 <!-------------------------------------------------------------------------- -->
1710 <a name="sink">-sink: Code sinking</a>
1713 <p>This pass moves instructions into successor blocks, when possible, so that
1714 they aren't executed on paths where their results aren't needed.
1718 <!-------------------------------------------------------------------------- -->
1720 <a name="sretpromotion">-sretpromotion: Promote sret arguments to multiple ret values</a>
1724 This pass finds functions that return a struct (using a pointer to the struct
1725 as the first argument of the function, marked with the '<tt>sret</tt>' attribute) and
1726 replaces them with a new function that simply returns each of the elements of
1727 that struct (using multiple return values).
1731 This pass works under a number of conditions:
1735 <li>The returned struct must not contain other structs</li>
1736 <li>The returned struct must only be used to load values from</li>
1737 <li>The placeholder struct passed in is the result of an <tt>alloca</tt></li>
1741 <!-------------------------------------------------------------------------- -->
1743 <a name="strip">-strip: Strip all symbols from a module</a>
1747 performs code stripping. this transformation can delete:
1751 <li>names for virtual registers</li>
1752 <li>symbols for internal globals and functions</li>
1753 <li>debug information</li>
1757 note that this transformation makes code much less readable, so it should
1758 only be used in situations where the <tt>strip</tt> utility would be used,
1759 such as reducing code size or making it harder to reverse engineer code.
1763 <!-------------------------------------------------------------------------- -->
1765 <a name="strip-dead-debug-info">-strip-dead-debug-info: Strip debug info for unused symbols</a>
1769 performs code stripping. this transformation can delete:
1773 <li>names for virtual registers</li>
1774 <li>symbols for internal globals and functions</li>
1775 <li>debug information</li>
1779 note that this transformation makes code much less readable, so it should
1780 only be used in situations where the <tt>strip</tt> utility would be used,
1781 such as reducing code size or making it harder to reverse engineer code.
1785 <!-------------------------------------------------------------------------- -->
1787 <a name="strip-dead-prototypes">-strip-dead-prototypes: Strip Unused Function Prototypes</a>
1791 This pass loops over all of the functions in the input module, looking for
1792 dead declarations and removes them. Dead declarations are declarations of
1793 functions for which no implementation is available (i.e., declarations for
1794 unused library functions).
1798 <!-------------------------------------------------------------------------- -->
1800 <a name="strip-debug-declare">-strip-debug-declare: Strip all llvm.dbg.declare intrinsics</a>
1803 <p>This pass implements code stripping. Specifically, it can delete:</p>
1805 <li>names for virtual registers</li>
1806 <li>symbols for internal globals and functions</li>
1807 <li>debug information</li>
1810 Note that this transformation makes code much less readable, so it should
1811 only be used in situations where the 'strip' utility would be used, such as
1812 reducing code size or making it harder to reverse engineer code.
1816 <!-------------------------------------------------------------------------- -->
1818 <a name="strip-nondebug">-strip-nondebug: Strip all symbols, except dbg symbols, from a module</a>
1821 <p>This pass implements code stripping. Specifically, it can delete:</p>
1823 <li>names for virtual registers</li>
1824 <li>symbols for internal globals and functions</li>
1825 <li>debug information</li>
1828 Note that this transformation makes code much less readable, so it should
1829 only be used in situations where the 'strip' utility would be used, such as
1830 reducing code size or making it harder to reverse engineer code.
1834 <!-------------------------------------------------------------------------- -->
1836 <a name="tailcallelim">-tailcallelim: Tail Call Elimination</a>
1840 This file transforms calls of the current function (self recursion) followed
1841 by a return instruction with a branch to the entry of the function, creating
1842 a loop. This pass also implements the following extensions to the basic
1847 <li>Trivial instructions between the call and return do not prevent the
1848 transformation from taking place, though currently the analysis cannot
1849 support moving any really useful instructions (only dead ones).
1850 <li>This pass transforms functions that are prevented from being tail
1851 recursive by an associative expression to use an accumulator variable,
1852 thus compiling the typical naive factorial or <tt>fib</tt> implementation
1853 into efficient code.
1854 <li>TRE is performed if the function returns void, if the return
1855 returns the result returned by the call, or if the function returns a
1856 run-time constant on all exits from the function. It is possible, though
1857 unlikely, that the return returns something else (like constant 0), and
1858 can still be TRE'd. It can be TRE'd if <em>all other</em> return
1859 instructions in the function return the exact same value.
1860 <li>If it can prove that callees do not access theier caller stack frame,
1861 they are marked as eligible for tail call elimination (by the code
1866 <!-------------------------------------------------------------------------- -->
1868 <a name="tailduplicate">-tailduplicate: Tail Duplication</a>
1872 This pass performs a limited form of tail duplication, intended to simplify
1873 CFGs by removing some unconditional branches. This pass is necessary to
1874 straighten out loops created by the C front-end, but also is capable of
1875 making other code nicer. After this pass is run, the CFG simplify pass
1876 should be run to clean up the mess.
1882 <!-- ======================================================================= -->
1883 <h2><a name="utilities">Utility Passes</a></h2>
1885 <p>This section describes the LLVM Utility Passes.</p>
1887 <!-------------------------------------------------------------------------- -->
1889 <a name="deadarghaX0r">-deadarghaX0r: Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</a>
1893 Same as dead argument elimination, but deletes arguments to functions which
1894 are external. This is only for use by <a
1895 href="Bugpoint.html">bugpoint</a>.</p>
1898 <!-------------------------------------------------------------------------- -->
1900 <a name="extract-blocks">-extract-blocks: Extract Basic Blocks From Module (for bugpoint use)</a>
1904 This pass is used by bugpoint to extract all blocks from the module into their
1908 <!-------------------------------------------------------------------------- -->
1910 <a name="instnamer">-instnamer: Assign names to anonymous instructions</a>
1913 <p>This is a little utility pass that gives instructions names, this is mostly
1914 useful when diffing the effect of an optimization because deleting an
1915 unnamed instruction can change all other instruction numbering, making the
1920 <!-------------------------------------------------------------------------- -->
1922 <a name="preverify">-preverify: Preliminary module verification</a>
1926 Ensures that the module is in the form required by the <a
1927 href="#verifier">Module Verifier</a> pass.
1931 Running the verifier runs this pass automatically, so there should be no need
1936 <!-------------------------------------------------------------------------- -->
1938 <a name="verify">-verify: Module Verifier</a>
1942 Verifies an LLVM IR code. This is useful to run after an optimization which is
1943 undergoing testing. Note that <tt>llvm-as</tt> verifies its input before
1944 emitting bitcode, and also that malformed bitcode is likely to make LLVM
1945 crash. All language front-ends are therefore encouraged to verify their output
1946 before performing optimizing transformations.
1950 <li>Both of a binary operator's parameters are of the same type.</li>
1951 <li>Verify that the indices of mem access instructions match other
1953 <li>Verify that arithmetic and other things are only performed on
1954 first-class types. Verify that shifts and logicals only happen on
1956 <li>All of the constants in a switch statement are of the correct type.</li>
1957 <li>The code is in valid SSA form.</li>
1958 <li>It is illegal to put a label into any other type (like a structure) or
1960 <li>Only phi nodes can be self referential: <tt>%x = add i32 %x, %x</tt> is
1962 <li>PHI nodes must have an entry for each predecessor, with no extras.</li>
1963 <li>PHI nodes must be the first thing in a basic block, all grouped
1965 <li>PHI nodes must have at least one entry.</li>
1966 <li>All basic blocks should only end with terminator insts, not contain
1968 <li>The entry node to a function must not have predecessors.</li>
1969 <li>All Instructions must be embedded into a basic block.</li>
1970 <li>Functions cannot take a void-typed parameter.</li>
1971 <li>Verify that a function's argument list agrees with its declared
1973 <li>It is illegal to specify a name for a void value.</li>
1974 <li>It is illegal to have a internal global value with no initializer.</li>
1975 <li>It is illegal to have a ret instruction that returns a value that does
1976 not agree with the function return value type.</li>
1977 <li>Function call argument types match the function prototype.</li>
1978 <li>All other things that are tested by asserts spread about the code.</li>
1982 Note that this does not provide full security verification (like Java), but
1983 instead just tries to ensure that code is well-formed.
1987 <!-------------------------------------------------------------------------- -->
1989 <a name="view-cfg">-view-cfg: View CFG of function</a>
1993 Displays the control flow graph using the GraphViz tool.
1997 <!-------------------------------------------------------------------------- -->
1999 <a name="view-cfg-only">-view-cfg-only: View CFG of function (with no function bodies)</a>
2003 Displays the control flow graph using the GraphViz tool, but omitting function
2008 <!-------------------------------------------------------------------------- -->
2010 <a name="view-dom">-view-dom: View dominance tree of function</a>
2014 Displays the dominator tree using the GraphViz tool.
2018 <!-------------------------------------------------------------------------- -->
2020 <a name="view-dom-only">-view-dom-only: View dominance tree of function (with no function bodies)</a>
2024 Displays the dominator tree using the GraphViz tool, but omitting function
2029 <!-------------------------------------------------------------------------- -->
2031 <a name="view-postdom">-view-postdom: View postdominance tree of function</a>
2035 Displays the post dominator tree using the GraphViz tool.
2039 <!-------------------------------------------------------------------------- -->
2041 <a name="view-postdom-only">-view-postdom-only: View postdominance tree of function (with no function bodies)</a>
2045 Displays the post dominator tree using the GraphViz tool, but omitting
2052 <!-- *********************************************************************** -->
2056 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
2057 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
2058 <a href="http://validator.w3.org/check/referer"><img
2059 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
2061 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a><br>
2062 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
2063 Last modified: $Date$