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="#postdomtree">-postdomtree</a></td><td>Post-Dominator Tree Construction</td></tr>
104 <tr><td><a href="#print-alias-sets">-print-alias-sets</a></td><td>Alias Set Printer</td></tr>
105 <tr><td><a href="#print-callgraph">-print-callgraph</a></td><td>Print a call graph</td></tr>
106 <tr><td><a href="#print-callgraph-sccs">-print-callgraph-sccs</a></td><td>Print SCCs of the Call Graph</td></tr>
107 <tr><td><a href="#print-cfg-sccs">-print-cfg-sccs</a></td><td>Print SCCs of each function CFG</td></tr>
108 <tr><td><a href="#print-dbginfo">-print-dbginfo</a></td><td>Print debug info in human readable form</td></tr>
109 <tr><td><a href="#print-dom-info">-print-dom-info</a></td><td>Dominator Info Printer</td></tr>
110 <tr><td><a href="#print-externalfnconstants">-print-externalfnconstants</a></td><td>Print external fn callsites passed constants</td></tr>
111 <tr><td><a href="#print-function">-print-function</a></td><td>Print function to stderr</td></tr>
112 <tr><td><a href="#print-module">-print-module</a></td><td>Print module to stderr</td></tr>
113 <tr><td><a href="#print-used-types">-print-used-types</a></td><td>Find Used Types</td></tr>
114 <tr><td><a href="#profile-estimator">-profile-estimator</a></td><td>Estimate profiling information</td></tr>
115 <tr><td><a href="#profile-loader">-profile-loader</a></td><td>Load profile information from llvmprof.out</td></tr>
116 <tr><td><a href="#profile-verifier">-profile-verifier</a></td><td>Verify profiling information</td></tr>
117 <tr><td><a href="#regions">-regions</a></td><td>Detect single entry single exit regions</td></tr>
118 <tr><td><a href="#scalar-evolution">-scalar-evolution</a></td><td>Scalar Evolution Analysis</td></tr>
119 <tr><td><a href="#scev-aa">-scev-aa</a></td><td>ScalarEvolution-based Alias Analysis</td></tr>
120 <tr><td><a href="#targetdata">-targetdata</a></td><td>Target Data Layout</td></tr>
123 <tr><th colspan="2"><b>TRANSFORM PASSES</b></th></tr>
124 <tr><th>Option</th><th>Name</th></tr>
125 <tr><td><a href="#adce">-adce</a></td><td>Aggressive Dead Code Elimination</td></tr>
126 <tr><td><a href="#always-inline">-always-inline</a></td><td>Inliner for always_inline functions</td></tr>
127 <tr><td><a href="#argpromotion">-argpromotion</a></td><td>Promote 'by reference' arguments to scalars</td></tr>
128 <tr><td><a href="#bb-vectorize">-bb-vectorize</a></td><td>Combine instructions to form vector instructions within basic blocks</td></tr>
129 <tr><td><a href="#block-placement">-block-placement</a></td><td>Profile Guided Basic Block Placement</td></tr>
130 <tr><td><a href="#break-crit-edges">-break-crit-edges</a></td><td>Break critical edges in CFG</td></tr>
131 <tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Optimize for code generation</td></tr>
132 <tr><td><a href="#constmerge">-constmerge</a></td><td>Merge Duplicate Global Constants</td></tr>
133 <tr><td><a href="#constprop">-constprop</a></td><td>Simple constant propagation</td></tr>
134 <tr><td><a href="#dce">-dce</a></td><td>Dead Code Elimination</td></tr>
135 <tr><td><a href="#deadargelim">-deadargelim</a></td><td>Dead Argument Elimination</td></tr>
136 <tr><td><a href="#deadtypeelim">-deadtypeelim</a></td><td>Dead Type Elimination</td></tr>
137 <tr><td><a href="#die">-die</a></td><td>Dead Instruction Elimination</td></tr>
138 <tr><td><a href="#dse">-dse</a></td><td>Dead Store Elimination</td></tr>
139 <tr><td><a href="#functionattrs">-functionattrs</a></td><td>Deduce function attributes</td></tr>
140 <tr><td><a href="#globaldce">-globaldce</a></td><td>Dead Global Elimination</td></tr>
141 <tr><td><a href="#globalopt">-globalopt</a></td><td>Global Variable Optimizer</td></tr>
142 <tr><td><a href="#gvn">-gvn</a></td><td>Global Value Numbering</td></tr>
143 <tr><td><a href="#indvars">-indvars</a></td><td>Canonicalize Induction Variables</td></tr>
144 <tr><td><a href="#inline">-inline</a></td><td>Function Integration/Inlining</td></tr>
145 <tr><td><a href="#insert-edge-profiling">-insert-edge-profiling</a></td><td>Insert instrumentation for edge profiling</td></tr>
146 <tr><td><a href="#insert-optimal-edge-profiling">-insert-optimal-edge-profiling</a></td><td>Insert optimal instrumentation for edge profiling</td></tr>
147 <tr><td><a href="#instcombine">-instcombine</a></td><td>Combine redundant instructions</td></tr>
148 <tr><td><a href="#internalize">-internalize</a></td><td>Internalize Global Symbols</td></tr>
149 <tr><td><a href="#ipconstprop">-ipconstprop</a></td><td>Interprocedural constant propagation</td></tr>
150 <tr><td><a href="#ipsccp">-ipsccp</a></td><td>Interprocedural Sparse Conditional Constant Propagation</td></tr>
151 <tr><td><a href="#jump-threading">-jump-threading</a></td><td>Jump Threading</td></tr>
152 <tr><td><a href="#lcssa">-lcssa</a></td><td>Loop-Closed SSA Form Pass</td></tr>
153 <tr><td><a href="#licm">-licm</a></td><td>Loop Invariant Code Motion</td></tr>
154 <tr><td><a href="#loop-deletion">-loop-deletion</a></td><td>Delete dead loops</td></tr>
155 <tr><td><a href="#loop-extract">-loop-extract</a></td><td>Extract loops into new functions</td></tr>
156 <tr><td><a href="#loop-extract-single">-loop-extract-single</a></td><td>Extract at most one loop into a new function</td></tr>
157 <tr><td><a href="#loop-reduce">-loop-reduce</a></td><td>Loop Strength Reduction</td></tr>
158 <tr><td><a href="#loop-rotate">-loop-rotate</a></td><td>Rotate Loops</td></tr>
159 <tr><td><a href="#loop-simplify">-loop-simplify</a></td><td>Canonicalize natural loops</td></tr>
160 <tr><td><a href="#loop-unroll">-loop-unroll</a></td><td>Unroll loops</td></tr>
161 <tr><td><a href="#loop-unswitch">-loop-unswitch</a></td><td>Unswitch loops</td></tr>
162 <tr><td><a href="#loweratomic">-loweratomic</a></td><td>Lower atomic intrinsics to non-atomic form</td></tr>
163 <tr><td><a href="#lowerinvoke">-lowerinvoke</a></td><td>Lower invoke and unwind, for unwindless code generators</td></tr>
164 <tr><td><a href="#lowerswitch">-lowerswitch</a></td><td>Lower SwitchInst's to branches</td></tr>
165 <tr><td><a href="#mem2reg">-mem2reg</a></td><td>Promote Memory to Register</td></tr>
166 <tr><td><a href="#memcpyopt">-memcpyopt</a></td><td>MemCpy Optimization</td></tr>
167 <tr><td><a href="#mergefunc">-mergefunc</a></td><td>Merge Functions</td></tr>
168 <tr><td><a href="#mergereturn">-mergereturn</a></td><td>Unify function exit nodes</td></tr>
169 <tr><td><a href="#partial-inliner">-partial-inliner</a></td><td>Partial Inliner</td></tr>
170 <tr><td><a href="#prune-eh">-prune-eh</a></td><td>Remove unused exception handling info</td></tr>
171 <tr><td><a href="#reassociate">-reassociate</a></td><td>Reassociate expressions</td></tr>
172 <tr><td><a href="#reg2mem">-reg2mem</a></td><td>Demote all values to stack slots</td></tr>
173 <tr><td><a href="#scalarrepl">-scalarrepl</a></td><td>Scalar Replacement of Aggregates (DT)</td></tr>
174 <tr><td><a href="#sccp">-sccp</a></td><td>Sparse Conditional Constant Propagation</td></tr>
175 <tr><td><a href="#simplify-libcalls">-simplify-libcalls</a></td><td>Simplify well-known library calls</td></tr>
176 <tr><td><a href="#simplifycfg">-simplifycfg</a></td><td>Simplify the CFG</td></tr>
177 <tr><td><a href="#sink">-sink</a></td><td>Code sinking</td></tr>
178 <tr><td><a href="#sretpromotion">-sretpromotion</a></td><td>Promote sret arguments to multiple ret values</td></tr>
179 <tr><td><a href="#strip">-strip</a></td><td>Strip all symbols from a module</td></tr>
180 <tr><td><a href="#strip-dead-debug-info">-strip-dead-debug-info</a></td><td>Strip debug info for unused symbols</td></tr>
181 <tr><td><a href="#strip-dead-prototypes">-strip-dead-prototypes</a></td><td>Strip Unused Function Prototypes</td></tr>
182 <tr><td><a href="#strip-debug-declare">-strip-debug-declare</a></td><td>Strip all llvm.dbg.declare intrinsics</td></tr>
183 <tr><td><a href="#strip-nondebug">-strip-nondebug</a></td><td>Strip all symbols, except dbg symbols, from a module</td></tr>
184 <tr><td><a href="#tailcallelim">-tailcallelim</a></td><td>Tail Call Elimination</td></tr>
185 <tr><td><a href="#tailduplicate">-tailduplicate</a></td><td>Tail Duplication</td></tr>
188 <tr><th colspan="2"><b>UTILITY PASSES</b></th></tr>
189 <tr><th>Option</th><th>Name</th></tr>
190 <tr><td><a href="#deadarghaX0r">-deadarghaX0r</a></td><td>Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</td></tr>
191 <tr><td><a href="#extract-blocks">-extract-blocks</a></td><td>Extract Basic Blocks From Module (for bugpoint use)</td></tr>
192 <tr><td><a href="#instnamer">-instnamer</a></td><td>Assign names to anonymous instructions</td></tr>
193 <tr><td><a href="#preverify">-preverify</a></td><td>Preliminary module verification</td></tr>
194 <tr><td><a href="#verify">-verify</a></td><td>Module Verifier</td></tr>
195 <tr><td><a href="#view-cfg">-view-cfg</a></td><td>View CFG of function</td></tr>
196 <tr><td><a href="#view-cfg-only">-view-cfg-only</a></td><td>View CFG of function (with no function bodies)</td></tr>
197 <tr><td><a href="#view-dom">-view-dom</a></td><td>View dominance tree of function</td></tr>
198 <tr><td><a href="#view-dom-only">-view-dom-only</a></td><td>View dominance tree of function (with no function bodies)</td></tr>
199 <tr><td><a href="#view-postdom">-view-postdom</a></td><td>View postdominance tree of function</td></tr>
200 <tr><td><a href="#view-postdom-only">-view-postdom-only</a></td><td>View postdominance tree of function (with no function bodies)</td></tr>
205 <!-- ======================================================================= -->
206 <h2><a name="analyses">Analysis Passes</a></h2>
208 <p>This section describes the LLVM Analysis Passes.</p>
210 <!-------------------------------------------------------------------------- -->
212 <a name="aa-eval">-aa-eval: Exhaustive Alias Analysis Precision Evaluator</a>
215 <p>This is a simple N^2 alias analysis accuracy evaluator.
216 Basically, for each function in the program, it simply queries to see how the
217 alias analysis implementation answers alias queries between each pair of
218 pointers in the function.</p>
220 <p>This is inspired and adapted from code by: Naveen Neelakantam, Francesco
221 Spadini, and Wojciech Stryjewski.</p>
224 <!-------------------------------------------------------------------------- -->
226 <a name="basicaa">-basicaa: Basic Alias Analysis (stateless AA impl)</a>
229 <p>A basic alias analysis pass that implements identities (two different
230 globals cannot alias, etc), but does no stateful analysis.</p>
233 <!-------------------------------------------------------------------------- -->
235 <a name="basiccg">-basiccg: Basic CallGraph Construction</a>
238 <p>Yet to be written.</p>
241 <!-------------------------------------------------------------------------- -->
243 <a name="count-aa">-count-aa: Count Alias Analysis Query Responses</a>
247 A pass which can be used to count how many alias queries
248 are being made and how the alias analysis implementation being used responds.
252 <!-------------------------------------------------------------------------- -->
254 <a name="debug-aa">-debug-aa: AA use debugger</a>
258 This simple pass checks alias analysis users to ensure that if they
259 create a new value, they do not query AA without informing it of the value.
260 It acts as a shim over any other AA pass you want.
264 Yes keeping track of every value in the program is expensive, but this is
269 <!-------------------------------------------------------------------------- -->
271 <a name="domfrontier">-domfrontier: Dominance Frontier Construction</a>
275 This pass is a simple dominator construction algorithm for finding forward
280 <!-------------------------------------------------------------------------- -->
282 <a name="domtree">-domtree: Dominator Tree Construction</a>
286 This pass is a simple dominator construction algorithm for finding forward
291 <!-------------------------------------------------------------------------- -->
293 <a name="dot-callgraph">-dot-callgraph: Print Call Graph to 'dot' file</a>
297 This pass, only available in <code>opt</code>, prints the call graph into a
298 <code>.dot</code> graph. This graph can then be processed with the "dot" tool
299 to convert it to postscript or some other suitable format.
303 <!-------------------------------------------------------------------------- -->
305 <a name="dot-cfg">-dot-cfg: Print CFG of function to 'dot' file</a>
309 This pass, only available in <code>opt</code>, prints the control flow graph
310 into a <code>.dot</code> graph. This graph can then be processed with the
311 "dot" tool to convert it to postscript or some other suitable format.
315 <!-------------------------------------------------------------------------- -->
317 <a name="dot-cfg-only">-dot-cfg-only: Print CFG of function to 'dot' file (with no function bodies)</a>
321 This pass, only available in <code>opt</code>, prints the control flow graph
322 into a <code>.dot</code> graph, omitting the function bodies. This graph can
323 then be processed with the "dot" tool to convert it to postscript or some
324 other suitable format.
328 <!-------------------------------------------------------------------------- -->
330 <a name="dot-dom">-dot-dom: Print dominance tree of function to 'dot' file</a>
334 This pass, only available in <code>opt</code>, prints the dominator tree
335 into a <code>.dot</code> graph. This graph can then be processed with the
336 "dot" tool to convert it to postscript or some other suitable format.
340 <!-------------------------------------------------------------------------- -->
342 <a name="dot-dom-only">-dot-dom-only: Print dominance tree of function to 'dot' file (with no function bodies)</a>
346 This pass, only available in <code>opt</code>, prints the dominator tree
347 into a <code>.dot</code> graph, omitting the function bodies. This graph can
348 then be processed with the "dot" tool to convert it to postscript or some
349 other suitable format.
353 <!-------------------------------------------------------------------------- -->
355 <a name="dot-postdom">-dot-postdom: Print postdominance tree of function to 'dot' file</a>
359 This pass, only available in <code>opt</code>, prints the post dominator tree
360 into a <code>.dot</code> graph. This graph can then be processed with the
361 "dot" tool to convert it to postscript or some other suitable format.
365 <!-------------------------------------------------------------------------- -->
367 <a name="dot-postdom-only">-dot-postdom-only: Print postdominance tree of function to 'dot' file (with no function bodies)</a>
371 This pass, only available in <code>opt</code>, prints the post dominator tree
372 into a <code>.dot</code> graph, omitting the function bodies. This graph can
373 then be processed with the "dot" tool to convert it to postscript or some
374 other suitable format.
378 <!-------------------------------------------------------------------------- -->
380 <a name="globalsmodref-aa">-globalsmodref-aa: Simple mod/ref analysis for globals</a>
384 This simple pass provides alias and mod/ref information for global values
385 that do not have their address taken, and keeps track of whether functions
386 read or write memory (are "pure"). For this simple (but very common) case,
387 we can provide pretty accurate and useful information.
391 <!-------------------------------------------------------------------------- -->
393 <a name="instcount">-instcount: Counts the various types of Instructions</a>
397 This pass collects the count of all instructions and reports them
401 <!-------------------------------------------------------------------------- -->
403 <a name="intervals">-intervals: Interval Partition Construction</a>
407 This analysis calculates and represents the interval partition of a function,
408 or a preexisting interval partition.
412 In this way, the interval partition may be used to reduce a flow graph down
413 to its degenerate single node interval partition (unless it is irreducible).
417 <!-------------------------------------------------------------------------- -->
419 <a name="iv-users">-iv-users: Induction Variable Users</a>
422 <p>Bookkeeping for "interesting" users of expressions computed from
423 induction variables.</p>
426 <!-------------------------------------------------------------------------- -->
428 <a name="lazy-value-info">-lazy-value-info: Lazy Value Information Analysis</a>
431 <p>Interface for lazy computation of value constraint information.</p>
434 <!-------------------------------------------------------------------------- -->
436 <a name="lda">-lda: Loop Dependence Analysis</a>
439 <p>Loop dependence analysis framework, which is used to detect dependences in
440 memory accesses in loops.</p>
443 <!-------------------------------------------------------------------------- -->
445 <a name="libcall-aa">-libcall-aa: LibCall Alias Analysis</a>
448 <p>LibCall Alias Analysis.</p>
451 <!-------------------------------------------------------------------------- -->
453 <a name="lint">-lint: Statically lint-checks LLVM IR</a>
456 <p>This pass statically checks for common and easily-identified constructs
457 which produce undefined or likely unintended behavior in LLVM IR.</p>
459 <p>It is not a guarantee of correctness, in two ways. First, it isn't
460 comprehensive. There are checks which could be done statically which are
461 not yet implemented. Some of these are indicated by TODO comments, but
462 those aren't comprehensive either. Second, many conditions cannot be
463 checked statically. This pass does no dynamic instrumentation, so it
464 can't check for all possible problems.</p>
466 <p>Another limitation is that it assumes all code will be executed. A store
467 through a null pointer in a basic block which is never reached is harmless,
468 but this pass will warn about it anyway.</p>
470 <p>Optimization passes may make conditions that this pass checks for more or
471 less obvious. If an optimization pass appears to be introducing a warning,
472 it may be that the optimization pass is merely exposing an existing
473 condition in the code.</p>
475 <p>This code may be run before instcombine. In many cases, instcombine checks
476 for the same kinds of things and turns instructions with undefined behavior
477 into unreachable (or equivalent). Because of this, this pass makes some
478 effort to look through bitcasts and so on.
482 <!-------------------------------------------------------------------------- -->
484 <a name="loops">-loops: Natural Loop Information</a>
488 This analysis is used to identify natural loops and determine the loop depth
489 of various nodes of the CFG. Note that the loops identified may actually be
490 several natural loops that share the same header node... not just a single
495 <!-------------------------------------------------------------------------- -->
497 <a name="memdep">-memdep: Memory Dependence Analysis</a>
501 An analysis that determines, for a given memory operation, what preceding
502 memory operations it depends on. It builds on alias analysis information, and
503 tries to provide a lazy, caching interface to a common kind of alias
508 <!-------------------------------------------------------------------------- -->
510 <a name="module-debuginfo">-module-debuginfo: Decodes module-level debug info</a>
513 <p>This pass decodes the debug info metadata in a module and prints in a
514 (sufficiently-prepared-) human-readable form.
516 For example, run this pass from opt along with the -analyze option, and
517 it'll print to standard output.
521 <!-------------------------------------------------------------------------- -->
523 <a name="no-aa">-no-aa: No Alias Analysis (always returns 'may' alias)</a>
527 This is the default implementation of the Alias Analysis interface. It always
528 returns "I don't know" for alias queries. NoAA is unlike other alias analysis
529 implementations, in that it does not chain to a previous analysis. As such it
530 doesn't follow many of the rules that other alias analyses must.
534 <!-------------------------------------------------------------------------- -->
536 <a name="no-profile">-no-profile: No Profile Information</a>
540 The default "no profile" implementation of the abstract
541 <code>ProfileInfo</code> interface.
545 <!-------------------------------------------------------------------------- -->
547 <a name="postdomfrontier">-postdomfrontier: Post-Dominance Frontier Construction</a>
551 This pass is a simple post-dominator construction algorithm for finding
552 post-dominator frontiers.
556 <!-------------------------------------------------------------------------- -->
558 <a name="postdomtree">-postdomtree: Post-Dominator Tree Construction</a>
562 This pass is a simple post-dominator construction algorithm for finding
567 <!-------------------------------------------------------------------------- -->
569 <a name="print-alias-sets">-print-alias-sets: Alias Set Printer</a>
572 <p>Yet to be written.</p>
575 <!-------------------------------------------------------------------------- -->
577 <a name="print-callgraph">-print-callgraph: Print a call graph</a>
581 This pass, only available in <code>opt</code>, prints the call graph to
582 standard error in a human-readable form.
586 <!-------------------------------------------------------------------------- -->
588 <a name="print-callgraph-sccs">-print-callgraph-sccs: Print SCCs of the Call Graph</a>
592 This pass, only available in <code>opt</code>, prints the SCCs of the call
593 graph to standard error in a human-readable form.
597 <!-------------------------------------------------------------------------- -->
599 <a name="print-cfg-sccs">-print-cfg-sccs: Print SCCs of each function CFG</a>
603 This pass, only available in <code>opt</code>, prints the SCCs of each
604 function CFG to standard error in a human-readable form.
608 <!-------------------------------------------------------------------------- -->
610 <a name="print-dbginfo">-print-dbginfo: Print debug info in human readable form</a>
613 <p>Pass that prints instructions, and associated debug info:</p>
616 <li>source/line/col information</li>
617 <li>original variable name</li>
618 <li>original type name</li>
622 <!-------------------------------------------------------------------------- -->
624 <a name="print-dom-info">-print-dom-info: Dominator Info Printer</a>
627 <p>Dominator Info Printer.</p>
630 <!-------------------------------------------------------------------------- -->
632 <a name="print-externalfnconstants">-print-externalfnconstants: Print external fn callsites passed constants</a>
636 This pass, only available in <code>opt</code>, prints out call sites to
637 external functions that are called with constant arguments. This can be
638 useful when looking for standard library functions we should constant fold
639 or handle in alias analyses.
643 <!-------------------------------------------------------------------------- -->
645 <a name="print-function">-print-function: Print function to stderr</a>
649 The <code>PrintFunctionPass</code> class is designed to be pipelined with
650 other <code>FunctionPass</code>es, and prints out the functions of the module
651 as they are processed.
655 <!-------------------------------------------------------------------------- -->
657 <a name="print-module">-print-module: Print module to stderr</a>
661 This pass simply prints out the entire module when it is executed.
665 <!-------------------------------------------------------------------------- -->
667 <a name="print-used-types">-print-used-types: Find Used Types</a>
671 This pass is used to seek out all of the types in use by the program. Note
672 that this analysis explicitly does not include types only used by the symbol
676 <!-------------------------------------------------------------------------- -->
678 <a name="profile-estimator">-profile-estimator: Estimate profiling information</a>
681 <p>Profiling information that estimates the profiling information
682 in a very crude and unimaginative way.
686 <!-------------------------------------------------------------------------- -->
688 <a name="profile-loader">-profile-loader: Load profile information from llvmprof.out</a>
692 A concrete implementation of profiling information that loads the information
693 from a profile dump file.
697 <!-------------------------------------------------------------------------- -->
699 <a name="profile-verifier">-profile-verifier: Verify profiling information</a>
702 <p>Pass that checks profiling information for plausibility.</p>
705 <a name="regions">-regions: Detect single entry single exit regions</a>
709 The <code>RegionInfo</code> pass detects single entry single exit regions in a
710 function, where a region is defined as any subgraph that is connected to the
711 remaining graph at only two spots. Furthermore, an hierarchical region tree is
716 <!-------------------------------------------------------------------------- -->
718 <a name="scalar-evolution">-scalar-evolution: Scalar Evolution Analysis</a>
722 The <code>ScalarEvolution</code> analysis can be used to analyze and
723 catagorize scalar expressions in loops. It specializes in recognizing general
724 induction variables, representing them with the abstract and opaque
725 <code>SCEV</code> class. Given this analysis, trip counts of loops and other
726 important properties can be obtained.
730 This analysis is primarily useful for induction variable substitution and
735 <!-------------------------------------------------------------------------- -->
737 <a name="scev-aa">-scev-aa: ScalarEvolution-based Alias Analysis</a>
740 <p>Simple alias analysis implemented in terms of ScalarEvolution queries.
742 This differs from traditional loop dependence analysis in that it tests
743 for dependencies within a single iteration of a loop, rather than
744 dependencies between different iterations.
746 ScalarEvolution has a more complete understanding of pointer arithmetic
747 than BasicAliasAnalysis' collection of ad-hoc analyses.
751 <!-------------------------------------------------------------------------- -->
753 <a name="targetdata">-targetdata: Target Data Layout</a>
756 <p>Provides other passes access to information on how the size and alignment
757 required by the target ABI for various data types.</p>
762 <!-- ======================================================================= -->
763 <h2><a name="transforms">Transform Passes</a></h2>
765 <p>This section describes the LLVM Transform Passes.</p>
767 <!-------------------------------------------------------------------------- -->
769 <a name="adce">-adce: Aggressive Dead Code Elimination</a>
772 <p>ADCE aggressively tries to eliminate code. This pass is similar to
773 <a href="#dce">DCE</a> but it assumes that values are dead until proven
774 otherwise. This is similar to <a href="#sccp">SCCP</a>, except applied to
775 the liveness of values.</p>
778 <!-------------------------------------------------------------------------- -->
780 <a name="always-inline">-always-inline: Inliner for always_inline functions</a>
783 <p>A custom inliner that handles only functions that are marked as
787 <!-------------------------------------------------------------------------- -->
789 <a name="argpromotion">-argpromotion: Promote 'by reference' arguments to scalars</a>
793 This pass promotes "by reference" arguments to be "by value" arguments. In
794 practice, this means looking for internal functions that have pointer
795 arguments. If it can prove, through the use of alias analysis, that an
796 argument is *only* loaded, then it can pass the value into the function
797 instead of the address of the value. This can cause recursive simplification
798 of code and lead to the elimination of allocas (especially in C++ template
803 This pass also handles aggregate arguments that are passed into a function,
804 scalarizing them if the elements of the aggregate are only loaded. Note that
805 it refuses to scalarize aggregates which would require passing in more than
806 three operands to the function, because passing thousands of operands for a
807 large array or structure is unprofitable!
811 Note that this transformation could also be done for arguments that are only
812 stored to (returning the value instead), but does not currently. This case
813 would be best handled when and if LLVM starts supporting multiple return
814 values from functions.
818 <!-------------------------------------------------------------------------- -->
820 <a name="bb-vectorize">-bb-vectorize: Basic-Block Vectorization</a>
823 <p>This pass combines instructions inside basic blocks to form vector
824 instructions. It iterates over each basic block, attempting to pair
825 compatible instructions, repeating this process until no additional
826 pairs are selected for vectorization. When the outputs of some pair
827 of compatible instructions are used as inputs by some other pair of
828 compatible instructions, those pairs are part of a potential
829 vectorization chain. Instruction pairs are only fused into vector
830 instructions when they are part of a chain longer than some
831 threshold length. Moreover, the pass attempts to find the best
832 possible chain for each pair of compatible instructions. These
833 heuristics are intended to prevent vectorization in cases where
834 it would not yield a performance increase of the resulting code.
838 <!-------------------------------------------------------------------------- -->
840 <a name="block-placement">-block-placement: Profile Guided Basic Block Placement</a>
843 <p>This pass is a very simple profile guided basic block placement algorithm.
844 The idea is to put frequently executed blocks together at the start of the
845 function and hopefully increase the number of fall-through conditional
846 branches. If there is no profile information for a particular function, this
847 pass basically orders blocks in depth-first order.</p>
850 <!-------------------------------------------------------------------------- -->
852 <a name="break-crit-edges">-break-crit-edges: Break critical edges in CFG</a>
856 Break all of the critical edges in the CFG by inserting a dummy basic block.
857 It may be "required" by passes that cannot deal with critical edges. This
858 transformation obviously invalidates the CFG, but can update forward dominator
859 (set, immediate dominators, tree, and frontier) information.
863 <!-------------------------------------------------------------------------- -->
865 <a name="codegenprepare">-codegenprepare: Optimize for code generation</a>
868 This pass munges the code in the input function to better prepare it for
869 SelectionDAG-based code generation. This works around limitations in it's
870 basic-block-at-a-time approach. It should eventually be removed.
873 <!-------------------------------------------------------------------------- -->
875 <a name="constmerge">-constmerge: Merge Duplicate Global Constants</a>
879 Merges duplicate global constants together into a single constant that is
880 shared. This is useful because some passes (ie TraceValues) insert a lot of
881 string constants into the program, regardless of whether or not an existing
886 <!-------------------------------------------------------------------------- -->
888 <a name="constprop">-constprop: Simple constant propagation</a>
891 <p>This file implements constant propagation and merging. It looks for
892 instructions involving only constant operands and replaces them with a
893 constant value instead of an instruction. For example:</p>
894 <blockquote><pre>add i32 1, 2</pre></blockquote>
896 <blockquote><pre>i32 3</pre></blockquote>
897 <p>NOTE: this pass has a habit of making definitions be dead. It is a good
898 idea to to run a <a href="#die">DIE</a> (Dead Instruction Elimination) pass
899 sometime after running this pass.</p>
902 <!-------------------------------------------------------------------------- -->
904 <a name="dce">-dce: Dead Code Elimination</a>
908 Dead code elimination is similar to <a href="#die">dead instruction
909 elimination</a>, but it rechecks instructions that were used by removed
910 instructions to see if they are newly dead.
914 <!-------------------------------------------------------------------------- -->
916 <a name="deadargelim">-deadargelim: Dead Argument Elimination</a>
920 This pass deletes dead arguments from internal functions. Dead argument
921 elimination removes arguments which are directly dead, as well as arguments
922 only passed into function calls as dead arguments of other functions. This
923 pass also deletes dead arguments in a similar way.
927 This pass is often useful as a cleanup pass to run after aggressive
928 interprocedural passes, which add possibly-dead arguments.
932 <!-------------------------------------------------------------------------- -->
934 <a name="deadtypeelim">-deadtypeelim: Dead Type Elimination</a>
938 This pass is used to cleanup the output of GCC. It eliminate names for types
939 that are unused in the entire translation unit, using the <a
940 href="#findusedtypes">find used types</a> pass.
944 <!-------------------------------------------------------------------------- -->
946 <a name="die">-die: Dead Instruction Elimination</a>
950 Dead instruction elimination performs a single pass over the function,
951 removing instructions that are obviously dead.
955 <!-------------------------------------------------------------------------- -->
957 <a name="dse">-dse: Dead Store Elimination</a>
961 A trivial dead store elimination that only considers basic-block local
966 <!-------------------------------------------------------------------------- -->
968 <a name="functionattrs">-functionattrs: Deduce function attributes</a>
971 <p>A simple interprocedural pass which walks the call-graph, looking for
972 functions which do not access or only read non-local memory, and marking them
973 readnone/readonly. In addition, it marks function arguments (of pointer type)
974 'nocapture' if a call to the function does not create any copies of the pointer
975 value that outlive the call. This more or less means that the pointer is only
976 dereferenced, and not returned from the function or stored in a global.
977 This pass is implemented as a bottom-up traversal of the call-graph.
981 <!-------------------------------------------------------------------------- -->
983 <a name="globaldce">-globaldce: Dead Global Elimination</a>
987 This transform is designed to eliminate unreachable internal globals from the
988 program. It uses an aggressive algorithm, searching out globals that are
989 known to be alive. After it finds all of the globals which are needed, it
990 deletes whatever is left over. This allows it to delete recursive chunks of
991 the program which are unreachable.
995 <!-------------------------------------------------------------------------- -->
997 <a name="globalopt">-globalopt: Global Variable Optimizer</a>
1001 This pass transforms simple global variables that never have their address
1002 taken. If obviously true, it marks read/write globals as constant, deletes
1003 variables only stored to, etc.
1007 <!-------------------------------------------------------------------------- -->
1009 <a name="gvn">-gvn: Global Value Numbering</a>
1013 This pass performs global value numbering to eliminate fully and partially
1014 redundant instructions. It also performs redundant load elimination.
1018 <!-------------------------------------------------------------------------- -->
1020 <a name="indvars">-indvars: Canonicalize Induction Variables</a>
1024 This transformation analyzes and transforms the induction variables (and
1025 computations derived from them) into simpler forms suitable for subsequent
1026 analysis and transformation.
1030 This transformation makes the following changes to each loop with an
1031 identifiable induction variable:
1035 <li>All loops are transformed to have a <em>single</em> canonical
1036 induction variable which starts at zero and steps by one.</li>
1037 <li>The canonical induction variable is guaranteed to be the first PHI node
1038 in the loop header block.</li>
1039 <li>Any pointer arithmetic recurrences are raised to use array
1044 If the trip count of a loop is computable, this pass also makes the following
1049 <li>The exit condition for the loop is canonicalized to compare the
1050 induction value against the exit value. This turns loops like:
1051 <blockquote><pre>for (i = 7; i*i < 1000; ++i)</pre></blockquote>
1053 <blockquote><pre>for (i = 0; i != 25; ++i)</pre></blockquote></li>
1054 <li>Any use outside of the loop of an expression derived from the indvar
1055 is changed to compute the derived value outside of the loop, eliminating
1056 the dependence on the exit value of the induction variable. If the only
1057 purpose of the loop is to compute the exit value of some derived
1058 expression, this transformation will make the loop dead.</li>
1062 This transformation should be followed by strength reduction after all of the
1063 desired loop transformations have been performed. Additionally, on targets
1064 where it is profitable, the loop could be transformed to count down to zero
1065 (the "do loop" optimization).
1069 <!-------------------------------------------------------------------------- -->
1071 <a name="inline">-inline: Function Integration/Inlining</a>
1075 Bottom-up inlining of functions into callees.
1079 <!-------------------------------------------------------------------------- -->
1081 <a name="insert-edge-profiling">-insert-edge-profiling: Insert instrumentation for edge profiling</a>
1085 This pass instruments the specified program with counters for edge profiling.
1086 Edge profiling can give a reasonable approximation of the hot paths through a
1087 program, and is used for a wide variety of program transformations.
1091 Note that this implementation is very naïve. It inserts a counter for
1092 <em>every</em> edge in the program, instead of using control flow information
1093 to prune the number of counters inserted.
1097 <!-------------------------------------------------------------------------- -->
1099 <a name="insert-optimal-edge-profiling">-insert-optimal-edge-profiling: Insert optimal instrumentation for edge profiling</a>
1102 <p>This pass instruments the specified program with counters for edge profiling.
1103 Edge profiling can give a reasonable approximation of the hot paths through a
1104 program, and is used for a wide variety of program transformations.
1108 <!-------------------------------------------------------------------------- -->
1110 <a name="instcombine">-instcombine: Combine redundant instructions</a>
1114 Combine instructions to form fewer, simple
1115 instructions. This pass does not modify the CFG This pass is where algebraic
1116 simplification happens.
1120 This pass combines things like:
1125 %Z = add i32 %Y, 1</pre></blockquote>
1132 >%Z = add i32 %X, 2</pre></blockquote>
1135 This is a simple worklist driven algorithm.
1139 This pass guarantees that the following canonicalizations are performed on
1144 <li>If a binary operator has a constant operand, it is moved to the right-
1146 <li>Bitwise operators with constant operands are always grouped so that
1147 shifts are performed first, then <code>or</code>s, then
1148 <code>and</code>s, then <code>xor</code>s.</li>
1149 <li>Compare instructions are converted from <code><</code>,
1150 <code>></code>, <code>≤</code>, or <code>≥</code> to
1151 <code>=</code> or <code>≠</code> if possible.</li>
1152 <li>All <code>cmp</code> instructions on boolean values are replaced with
1153 logical operations.</li>
1154 <li><code>add <var>X</var>, <var>X</var></code> is represented as
1155 <code>mul <var>X</var>, 2</code> ⇒ <code>shl <var>X</var>, 1</code></li>
1156 <li>Multiplies with a constant power-of-two argument are transformed into
1162 <!-------------------------------------------------------------------------- -->
1164 <a name="internalize">-internalize: Internalize Global Symbols</a>
1168 This pass loops over all of the functions in the input module, looking for a
1169 main function. If a main function is found, all other functions and all
1170 global variables with initializers are marked as internal.
1174 <!-------------------------------------------------------------------------- -->
1176 <a name="ipconstprop">-ipconstprop: Interprocedural constant propagation</a>
1180 This pass implements an <em>extremely</em> simple interprocedural constant
1181 propagation pass. It could certainly be improved in many different ways,
1182 like using a worklist. This pass makes arguments dead, but does not remove
1183 them. The existing dead argument elimination pass should be run after this
1184 to clean up the mess.
1188 <!-------------------------------------------------------------------------- -->
1190 <a name="ipsccp">-ipsccp: Interprocedural Sparse Conditional Constant Propagation</a>
1194 An interprocedural variant of <a href="#sccp">Sparse Conditional Constant
1199 <!-------------------------------------------------------------------------- -->
1201 <a name="jump-threading">-jump-threading: Jump Threading</a>
1205 Jump threading tries to find distinct threads of control flow running through
1206 a basic block. This pass looks at blocks that have multiple predecessors and
1207 multiple successors. If one or more of the predecessors of the block can be
1208 proven to always cause a jump to one of the successors, we forward the edge
1209 from the predecessor to the successor by duplicating the contents of this
1213 An example of when this can occur is code like this:
1220 if (X < 3) {</pre>
1223 In this case, the unconditional branch at the end of the first if can be
1224 revectored to the false side of the second if.
1228 <!-------------------------------------------------------------------------- -->
1230 <a name="lcssa">-lcssa: Loop-Closed SSA Form Pass</a>
1234 This pass transforms loops by placing phi nodes at the end of the loops for
1235 all values that are live across the loop boundary. For example, it turns
1236 the left into the right code:
1240 >for (...) for (...)
1245 X3 = phi(X1, X2) X3 = phi(X1, X2)
1246 ... = X3 + 4 X4 = phi(X3)
1250 This is still valid LLVM; the extra phi nodes are purely redundant, and will
1251 be trivially eliminated by <code>InstCombine</code>. The major benefit of
1252 this transformation is that it makes many other loop optimizations, such as
1253 LoopUnswitching, simpler.
1257 <!-------------------------------------------------------------------------- -->
1259 <a name="licm">-licm: Loop Invariant Code Motion</a>
1263 This pass performs loop invariant code motion, attempting to remove as much
1264 code from the body of a loop as possible. It does this by either hoisting
1265 code into the preheader block, or by sinking code to the exit blocks if it is
1266 safe. This pass also promotes must-aliased memory locations in the loop to
1267 live in registers, thus hoisting and sinking "invariant" loads and stores.
1271 This pass uses alias analysis for two purposes:
1275 <li>Moving loop invariant loads and calls out of loops. If we can determine
1276 that a load or call inside of a loop never aliases anything stored to,
1277 we can hoist it or sink it like any other instruction.</li>
1278 <li>Scalar Promotion of Memory - If there is a store instruction inside of
1279 the loop, we try to move the store to happen AFTER the loop instead of
1280 inside of the loop. This can only happen if a few conditions are true:
1282 <li>The pointer stored through is loop invariant.</li>
1283 <li>There are no stores or loads in the loop which <em>may</em> alias
1284 the pointer. There are no calls in the loop which mod/ref the
1287 If these conditions are true, we can promote the loads and stores in the
1288 loop of the pointer to use a temporary alloca'd variable. We then use
1289 the mem2reg functionality to construct the appropriate SSA form for the
1294 <!-------------------------------------------------------------------------- -->
1296 <a name="loop-deletion">-loop-deletion: Delete dead loops</a>
1300 This file implements the Dead Loop Deletion Pass. This pass is responsible
1301 for eliminating loops with non-infinite computable trip counts that have no
1302 side effects or volatile instructions, and do not contribute to the
1303 computation of the function's return value.
1307 <!-------------------------------------------------------------------------- -->
1309 <a name="loop-extract">-loop-extract: Extract loops into new functions</a>
1313 A pass wrapper around the <code>ExtractLoop()</code> scalar transformation to
1314 extract each top-level loop into its own new function. If the loop is the
1315 <em>only</em> loop in a given function, it is not touched. This is a pass most
1316 useful for debugging via bugpoint.
1320 <!-------------------------------------------------------------------------- -->
1322 <a name="loop-extract-single">-loop-extract-single: Extract at most one loop into a new function</a>
1326 Similar to <a href="#loop-extract">Extract loops into new functions</a>,
1327 this pass extracts one natural loop from the program into a function if it
1328 can. This is used by bugpoint.
1332 <!-------------------------------------------------------------------------- -->
1334 <a name="loop-reduce">-loop-reduce: Loop Strength Reduction</a>
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.
1346 <!-------------------------------------------------------------------------- -->
1348 <a name="loop-rotate">-loop-rotate: Rotate Loops</a>
1351 <p>A simple loop rotation transformation.</p>
1354 <!-------------------------------------------------------------------------- -->
1356 <a name="loop-simplify">-loop-simplify: Canonicalize natural loops</a>
1360 This pass performs several transformations to transform natural loops into a
1361 simpler form, which makes subsequent analyses and transformations simpler and
1366 Loop pre-header insertion guarantees that there is a single, non-critical
1367 entry edge from outside of the loop to the loop header. This simplifies a
1368 number of analyses and transformations, such as LICM.
1372 Loop exit-block insertion guarantees that all exit blocks from the loop
1373 (blocks which are outside of the loop that have predecessors inside of the
1374 loop) only have predecessors from inside of the loop (and are thus dominated
1375 by the loop header). This simplifies transformations such as store-sinking
1376 that are built into LICM.
1380 This pass also guarantees that loops will have exactly one backedge.
1384 Note that the simplifycfg pass will clean up blocks which are split out but
1385 end up being unnecessary, so usage of this pass should not pessimize
1390 This pass obviously modifies the CFG, but updates loop information and
1391 dominator information.
1395 <!-------------------------------------------------------------------------- -->
1397 <a name="loop-unroll">-loop-unroll: Unroll loops</a>
1401 This pass implements a simple loop unroller. It works best when loops have
1402 been canonicalized by the <a href="#indvars"><tt>-indvars</tt></a> pass,
1403 allowing it to determine the trip counts of loops easily.
1407 <!-------------------------------------------------------------------------- -->
1409 <a name="loop-unswitch">-loop-unswitch: Unswitch loops</a>
1413 This pass transforms loops that contain branches on loop-invariant conditions
1414 to have multiple loops. For example, it turns the left into the right code:
1426 This can increase the size of the code exponentially (doubling it every time
1427 a loop is unswitched) so we only unswitch if the resultant code will be
1428 smaller than a threshold.
1432 This pass expects LICM to be run before it to hoist invariant conditions out
1433 of the loop, to make the unswitching opportunity obvious.
1437 <!-------------------------------------------------------------------------- -->
1439 <a name="loweratomic">-loweratomic: Lower atomic intrinsics to non-atomic form</a>
1443 This pass lowers atomic intrinsics to non-atomic form for use in a known
1444 non-preemptible environment.
1448 The pass does not verify that the environment is non-preemptible (in
1449 general this would require knowledge of the entire call graph of the
1450 program including any libraries which may not be available in bitcode form);
1451 it simply lowers every atomic intrinsic.
1455 <!-------------------------------------------------------------------------- -->
1457 <a name="lowerinvoke">-lowerinvoke: Lower invoke and unwind, for unwindless code generators</a>
1461 This transformation is designed for use by code generators which do not yet
1462 support stack unwinding. This pass supports two models of exception handling
1463 lowering, the 'cheap' support and the 'expensive' support.
1467 'Cheap' exception handling support gives the program the ability to execute
1468 any program which does not "throw an exception", by turning 'invoke'
1469 instructions into calls and by turning 'unwind' instructions into calls to
1470 abort(). If the program does dynamically use the unwind instruction, the
1471 program will print a message then abort.
1475 'Expensive' exception handling support gives the full exception handling
1476 support to the program at the cost of making the 'invoke' instruction
1477 really expensive. It basically inserts setjmp/longjmp calls to emulate the
1478 exception handling as necessary.
1482 Because the 'expensive' support slows down programs a lot, and EH is only
1483 used for a subset of the programs, it must be specifically enabled by the
1484 <tt>-enable-correct-eh-support</tt> option.
1488 Note that after this pass runs the CFG is not entirely accurate (exceptional
1489 control flow edges are not correct anymore) so only very simple things should
1490 be done after the lowerinvoke pass has run (like generation of native code).
1491 This should not be used as a general purpose "my LLVM-to-LLVM pass doesn't
1492 support the invoke instruction yet" lowering pass.
1496 <!-------------------------------------------------------------------------- -->
1498 <a name="lowerswitch">-lowerswitch: Lower SwitchInst's to branches</a>
1502 Rewrites <tt>switch</tt> instructions with a sequence of branches, which
1503 allows targets to get away with not implementing the switch instruction until
1508 <!-------------------------------------------------------------------------- -->
1510 <a name="mem2reg">-mem2reg: Promote Memory to Register</a>
1514 This file promotes memory references to be register references. It promotes
1515 <tt>alloca</tt> instructions which only have <tt>load</tt>s and
1516 <tt>store</tt>s as uses. An <tt>alloca</tt> is transformed by using dominator
1517 frontiers to place <tt>phi</tt> nodes, then traversing the function in
1518 depth-first order to rewrite <tt>load</tt>s and <tt>store</tt>s as
1519 appropriate. This is just the standard SSA construction algorithm to construct
1524 <!-------------------------------------------------------------------------- -->
1526 <a name="memcpyopt">-memcpyopt: MemCpy Optimization</a>
1530 This pass performs various transformations related to eliminating memcpy
1531 calls, or transforming sets of stores into memset's.
1535 <!-------------------------------------------------------------------------- -->
1537 <a name="mergefunc">-mergefunc: Merge Functions</a>
1540 <p>This pass looks for equivalent functions that are mergable and folds them.
1542 A hash is computed from the function, based on its type and number of
1545 Once all hashes are computed, we perform an expensive equality comparison
1546 on each function pair. This takes n^2/2 comparisons per bucket, so it's
1547 important that the hash function be high quality. The equality comparison
1548 iterates through each instruction in each basic block.
1550 When a match is found the functions are folded. If both functions are
1551 overridable, we move the functionality into a new internal function and
1552 leave two overridable thunks to it.
1556 <!-------------------------------------------------------------------------- -->
1558 <a name="mergereturn">-mergereturn: Unify function exit nodes</a>
1562 Ensure that functions have at most one <tt>ret</tt> instruction in them.
1563 Additionally, it keeps track of which node is the new exit node of the CFG.
1567 <!-------------------------------------------------------------------------- -->
1569 <a name="partial-inliner">-partial-inliner: Partial Inliner</a>
1572 <p>This pass performs partial inlining, typically by inlining an if
1573 statement that surrounds the body of the function.
1577 <!-------------------------------------------------------------------------- -->
1579 <a name="prune-eh">-prune-eh: Remove unused exception handling info</a>
1583 This file implements a simple interprocedural pass which walks the call-graph,
1584 turning <tt>invoke</tt> instructions into <tt>call</tt> instructions if and
1585 only if the callee cannot throw an exception. It implements this as a
1586 bottom-up traversal of the call-graph.
1590 <!-------------------------------------------------------------------------- -->
1592 <a name="reassociate">-reassociate: Reassociate expressions</a>
1596 This pass reassociates commutative expressions in an order that is designed
1597 to promote better constant propagation, GCSE, LICM, PRE, etc.
1601 For example: 4 + (<var>x</var> + 5) ⇒ <var>x</var> + (4 + 5)
1605 In the implementation of this algorithm, constants are assigned rank = 0,
1606 function arguments are rank = 1, and other values are assigned ranks
1607 corresponding to the reverse post order traversal of current function
1608 (starting at 2), which effectively gives values in deep loops higher rank
1609 than values not in loops.
1613 <!-------------------------------------------------------------------------- -->
1615 <a name="reg2mem">-reg2mem: Demote all values to stack slots</a>
1619 This file demotes all registers to memory references. It is intended to be
1620 the inverse of <a href="#mem2reg"><tt>-mem2reg</tt></a>. By converting to
1621 <tt>load</tt> instructions, the only values live across basic blocks are
1622 <tt>alloca</tt> instructions and <tt>load</tt> instructions before
1623 <tt>phi</tt> nodes. It is intended that this should make CFG hacking much
1624 easier. To make later hacking easier, the entry block is split into two, such
1625 that all introduced <tt>alloca</tt> instructions (and nothing else) are in the
1630 <!-------------------------------------------------------------------------- -->
1632 <a name="scalarrepl">-scalarrepl: Scalar Replacement of Aggregates (DT)</a>
1636 The well-known scalar replacement of aggregates transformation. This
1637 transform breaks up <tt>alloca</tt> instructions of aggregate type (structure
1638 or array) into individual <tt>alloca</tt> instructions for each member if
1639 possible. Then, if possible, it transforms the individual <tt>alloca</tt>
1640 instructions into nice clean scalar SSA form.
1644 This combines a simple scalar replacement of aggregates algorithm with the <a
1645 href="#mem2reg"><tt>mem2reg</tt></a> algorithm because often interact,
1646 especially for C++ programs. As such, iterating between <tt>scalarrepl</tt>,
1647 then <a href="#mem2reg"><tt>mem2reg</tt></a> until we run out of things to
1652 <!-------------------------------------------------------------------------- -->
1654 <a name="sccp">-sccp: Sparse Conditional Constant Propagation</a>
1658 Sparse conditional constant propagation and merging, which can be summarized
1663 <li>Assumes values are constant unless proven otherwise</li>
1664 <li>Assumes BasicBlocks are dead unless proven otherwise</li>
1665 <li>Proves values to be constant, and replaces them with constants</li>
1666 <li>Proves conditional branches to be unconditional</li>
1670 Note that this pass has a habit of making definitions be dead. It is a good
1671 idea to to run a DCE pass sometime after running this pass.
1675 <!-------------------------------------------------------------------------- -->
1677 <a name="simplify-libcalls">-simplify-libcalls: Simplify well-known library calls</a>
1681 Applies a variety of small optimizations for calls to specific well-known
1682 function calls (e.g. runtime library functions). For example, a call
1683 <tt>exit(3)</tt> that occurs within the <tt>main()</tt> function can be
1684 transformed into simply <tt>return 3</tt>.
1688 <!-------------------------------------------------------------------------- -->
1690 <a name="simplifycfg">-simplifycfg: Simplify the CFG</a>
1694 Performs dead code elimination and basic block merging. Specifically:
1698 <li>Removes basic blocks with no predecessors.</li>
1699 <li>Merges a basic block into its predecessor if there is only one and the
1700 predecessor only has one successor.</li>
1701 <li>Eliminates PHI nodes for basic blocks with a single predecessor.</li>
1702 <li>Eliminates a basic block that only contains an unconditional
1707 <!-------------------------------------------------------------------------- -->
1709 <a name="sink">-sink: Code sinking</a>
1712 <p>This pass moves instructions into successor blocks, when possible, so that
1713 they aren't executed on paths where their results aren't needed.
1717 <!-------------------------------------------------------------------------- -->
1719 <a name="sretpromotion">-sretpromotion: Promote sret arguments to multiple ret values</a>
1723 This pass finds functions that return a struct (using a pointer to the struct
1724 as the first argument of the function, marked with the '<tt>sret</tt>' attribute) and
1725 replaces them with a new function that simply returns each of the elements of
1726 that struct (using multiple return values).
1730 This pass works under a number of conditions:
1734 <li>The returned struct must not contain other structs</li>
1735 <li>The returned struct must only be used to load values from</li>
1736 <li>The placeholder struct passed in is the result of an <tt>alloca</tt></li>
1740 <!-------------------------------------------------------------------------- -->
1742 <a name="strip">-strip: Strip all symbols from a module</a>
1746 performs code stripping. this transformation can delete:
1750 <li>names for virtual registers</li>
1751 <li>symbols for internal globals and functions</li>
1752 <li>debug information</li>
1756 note that this transformation makes code much less readable, so it should
1757 only be used in situations where the <tt>strip</tt> utility would be used,
1758 such as reducing code size or making it harder to reverse engineer code.
1762 <!-------------------------------------------------------------------------- -->
1764 <a name="strip-dead-debug-info">-strip-dead-debug-info: Strip debug info for unused symbols</a>
1768 performs code stripping. this transformation can delete:
1772 <li>names for virtual registers</li>
1773 <li>symbols for internal globals and functions</li>
1774 <li>debug information</li>
1778 note that this transformation makes code much less readable, so it should
1779 only be used in situations where the <tt>strip</tt> utility would be used,
1780 such as reducing code size or making it harder to reverse engineer code.
1784 <!-------------------------------------------------------------------------- -->
1786 <a name="strip-dead-prototypes">-strip-dead-prototypes: Strip Unused Function Prototypes</a>
1790 This pass loops over all of the functions in the input module, looking for
1791 dead declarations and removes them. Dead declarations are declarations of
1792 functions for which no implementation is available (i.e., declarations for
1793 unused library functions).
1797 <!-------------------------------------------------------------------------- -->
1799 <a name="strip-debug-declare">-strip-debug-declare: Strip all llvm.dbg.declare intrinsics</a>
1802 <p>This pass implements code stripping. Specifically, it can delete:</p>
1804 <li>names for virtual registers</li>
1805 <li>symbols for internal globals and functions</li>
1806 <li>debug information</li>
1809 Note that this transformation makes code much less readable, so it should
1810 only be used in situations where the 'strip' utility would be used, such as
1811 reducing code size or making it harder to reverse engineer code.
1815 <!-------------------------------------------------------------------------- -->
1817 <a name="strip-nondebug">-strip-nondebug: Strip all symbols, except dbg symbols, from a module</a>
1820 <p>This pass implements code stripping. Specifically, it can delete:</p>
1822 <li>names for virtual registers</li>
1823 <li>symbols for internal globals and functions</li>
1824 <li>debug information</li>
1827 Note that this transformation makes code much less readable, so it should
1828 only be used in situations where the 'strip' utility would be used, such as
1829 reducing code size or making it harder to reverse engineer code.
1833 <!-------------------------------------------------------------------------- -->
1835 <a name="tailcallelim">-tailcallelim: Tail Call Elimination</a>
1839 This file transforms calls of the current function (self recursion) followed
1840 by a return instruction with a branch to the entry of the function, creating
1841 a loop. This pass also implements the following extensions to the basic
1846 <li>Trivial instructions between the call and return do not prevent the
1847 transformation from taking place, though currently the analysis cannot
1848 support moving any really useful instructions (only dead ones).
1849 <li>This pass transforms functions that are prevented from being tail
1850 recursive by an associative expression to use an accumulator variable,
1851 thus compiling the typical naive factorial or <tt>fib</tt> implementation
1852 into efficient code.
1853 <li>TRE is performed if the function returns void, if the return
1854 returns the result returned by the call, or if the function returns a
1855 run-time constant on all exits from the function. It is possible, though
1856 unlikely, that the return returns something else (like constant 0), and
1857 can still be TRE'd. It can be TRE'd if <em>all other</em> return
1858 instructions in the function return the exact same value.
1859 <li>If it can prove that callees do not access theier caller stack frame,
1860 they are marked as eligible for tail call elimination (by the code
1865 <!-------------------------------------------------------------------------- -->
1867 <a name="tailduplicate">-tailduplicate: Tail Duplication</a>
1871 This pass performs a limited form of tail duplication, intended to simplify
1872 CFGs by removing some unconditional branches. This pass is necessary to
1873 straighten out loops created by the C front-end, but also is capable of
1874 making other code nicer. After this pass is run, the CFG simplify pass
1875 should be run to clean up the mess.
1881 <!-- ======================================================================= -->
1882 <h2><a name="utilities">Utility Passes</a></h2>
1884 <p>This section describes the LLVM Utility Passes.</p>
1886 <!-------------------------------------------------------------------------- -->
1888 <a name="deadarghaX0r">-deadarghaX0r: Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</a>
1892 Same as dead argument elimination, but deletes arguments to functions which
1893 are external. This is only for use by <a
1894 href="Bugpoint.html">bugpoint</a>.</p>
1897 <!-------------------------------------------------------------------------- -->
1899 <a name="extract-blocks">-extract-blocks: Extract Basic Blocks From Module (for bugpoint use)</a>
1903 This pass is used by bugpoint to extract all blocks from the module into their
1907 <!-------------------------------------------------------------------------- -->
1909 <a name="instnamer">-instnamer: Assign names to anonymous instructions</a>
1912 <p>This is a little utility pass that gives instructions names, this is mostly
1913 useful when diffing the effect of an optimization because deleting an
1914 unnamed instruction can change all other instruction numbering, making the
1919 <!-------------------------------------------------------------------------- -->
1921 <a name="preverify">-preverify: Preliminary module verification</a>
1925 Ensures that the module is in the form required by the <a
1926 href="#verifier">Module Verifier</a> pass.
1930 Running the verifier runs this pass automatically, so there should be no need
1935 <!-------------------------------------------------------------------------- -->
1937 <a name="verify">-verify: Module Verifier</a>
1941 Verifies an LLVM IR code. This is useful to run after an optimization which is
1942 undergoing testing. Note that <tt>llvm-as</tt> verifies its input before
1943 emitting bitcode, and also that malformed bitcode is likely to make LLVM
1944 crash. All language front-ends are therefore encouraged to verify their output
1945 before performing optimizing transformations.
1949 <li>Both of a binary operator's parameters are of the same type.</li>
1950 <li>Verify that the indices of mem access instructions match other
1952 <li>Verify that arithmetic and other things are only performed on
1953 first-class types. Verify that shifts and logicals only happen on
1955 <li>All of the constants in a switch statement are of the correct type.</li>
1956 <li>The code is in valid SSA form.</li>
1957 <li>It is illegal to put a label into any other type (like a structure) or
1959 <li>Only phi nodes can be self referential: <tt>%x = add i32 %x, %x</tt> is
1961 <li>PHI nodes must have an entry for each predecessor, with no extras.</li>
1962 <li>PHI nodes must be the first thing in a basic block, all grouped
1964 <li>PHI nodes must have at least one entry.</li>
1965 <li>All basic blocks should only end with terminator insts, not contain
1967 <li>The entry node to a function must not have predecessors.</li>
1968 <li>All Instructions must be embedded into a basic block.</li>
1969 <li>Functions cannot take a void-typed parameter.</li>
1970 <li>Verify that a function's argument list agrees with its declared
1972 <li>It is illegal to specify a name for a void value.</li>
1973 <li>It is illegal to have an internal global value with no initializer.</li>
1974 <li>It is illegal to have a ret instruction that returns a value that does
1975 not agree with the function return value type.</li>
1976 <li>Function call argument types match the function prototype.</li>
1977 <li>All other things that are tested by asserts spread about the code.</li>
1981 Note that this does not provide full security verification (like Java), but
1982 instead just tries to ensure that code is well-formed.
1986 <!-------------------------------------------------------------------------- -->
1988 <a name="view-cfg">-view-cfg: View CFG of function</a>
1992 Displays the control flow graph using the GraphViz tool.
1996 <!-------------------------------------------------------------------------- -->
1998 <a name="view-cfg-only">-view-cfg-only: View CFG of function (with no function bodies)</a>
2002 Displays the control flow graph using the GraphViz tool, but omitting function
2007 <!-------------------------------------------------------------------------- -->
2009 <a name="view-dom">-view-dom: View dominance tree of function</a>
2013 Displays the dominator tree using the GraphViz tool.
2017 <!-------------------------------------------------------------------------- -->
2019 <a name="view-dom-only">-view-dom-only: View dominance tree of function (with no function bodies)</a>
2023 Displays the dominator tree using the GraphViz tool, but omitting function
2028 <!-------------------------------------------------------------------------- -->
2030 <a name="view-postdom">-view-postdom: View postdominance tree of function</a>
2034 Displays the post dominator tree using the GraphViz tool.
2038 <!-------------------------------------------------------------------------- -->
2040 <a name="view-postdom-only">-view-postdom-only: View postdominance tree of function (with no function bodies)</a>
2044 Displays the post dominator tree using the GraphViz tool, but omitting
2051 <!-- *********************************************************************** -->
2055 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
2056 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
2057 <a href="http://validator.w3.org/check/referer"><img
2058 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
2060 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a><br>
2061 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
2062 Last modified: $Date$