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="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\">$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 <div class="doc_title">LLVM's Analysis and Transform Passes</div>
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 <div class="doc_section"> <a name="intro">Introduction</a> </div>
59 <div class="doc_text">
60 <p>This document serves as a high level summary of the optimization features
61 that LLVM provides. Optimizations are implemented as Passes that traverse some
62 portion of a program to either collect information or transform the program.
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>
73 <div class="doc_text" >
75 <tr><th colspan="2"><b>ANALYSIS PASSES</b></th></tr>
76 <tr><th>Option</th><th>Name</th></tr>
77 <tr><td><a href="#aa-eval">-aa-eval</a></td><td>Exhaustive Alias Analysis Precision Evaluator</td></tr>
78 <tr><td><a href="#basicaa">-basicaa</a></td><td>Basic Alias Analysis (default AA impl)</td></tr>
79 <tr><td><a href="#basiccg">-basiccg</a></td><td>Basic CallGraph Construction</td></tr>
80 <tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Optimize for code generation</td></tr>
81 <tr><td><a href="#count-aa">-count-aa</a></td><td>Count Alias Analysis Query Responses</td></tr>
82 <tr><td><a href="#debug-aa">-debug-aa</a></td><td>AA use debugger</td></tr>
83 <tr><td><a href="#domfrontier">-domfrontier</a></td><td>Dominance Frontier Construction</td></tr>
84 <tr><td><a href="#domtree">-domtree</a></td><td>Dominator Tree Construction</td></tr>
85 <tr><td><a href="#dot-callgraph">-dot-callgraph</a></td><td>Print Call Graph to 'dot' file</td></tr>
86 <tr><td><a href="#dot-cfg">-dot-cfg</a></td><td>Print CFG of function to 'dot' file</td></tr>
87 <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>
88 <tr><td><a href="#dot-dom">-dot-dom</a></td><td>Print dominator tree of function to 'dot' file</td></tr>
89 <tr><td><a href="#dot-dom-only">-dot-dom-only</a></td><td>Print dominator tree of function to 'dot' file (with no function bodies)</td></tr>
90 <tr><td><a href="#dot-postdom">-dot-postdom</a></td><td>Print post dominator tree of function to 'dot' file</td></tr>
91 <tr><td><a href="#dot-postdom-only">-dot-postdom-only</a></td><td>Print post dominator tree of function to 'dot' file (with no function bodies)</td></tr>
92 <tr><td><a href="#globalsmodref-aa">-globalsmodref-aa</a></td><td>Simple mod/ref analysis for globals</td></tr>
93 <tr><td><a href="#instcount">-instcount</a></td><td>Counts the various types of Instructions</td></tr>
94 <tr><td><a href="#intervals">-intervals</a></td><td>Interval Partition Construction</td></tr>
95 <tr><td><a href="#loops">-loops</a></td><td>Natural Loop Construction</td></tr>
96 <tr><td><a href="#memdep">-memdep</a></td><td>Memory Dependence Analysis</td></tr>
97 <tr><td><a href="#no-aa">-no-aa</a></td><td>No Alias Analysis (always returns 'may' alias)</td></tr>
98 <tr><td><a href="#no-profile">-no-profile</a></td><td>No Profile Information</td></tr>
99 <tr><td><a href="#postdomfrontier">-postdomfrontier</a></td><td>Post-Dominance Frontier Construction</td></tr>
100 <tr><td><a href="#postdomtree">-postdomtree</a></td><td>Post-Dominator Tree Construction</td></tr>
101 <tr><td><a href="#print-alias-sets">-print-alias-sets</a></td><td>Alias Set Printer</td></tr>
102 <tr><td><a href="#print-callgraph">-print-callgraph</a></td><td>Print a call graph</td></tr>
103 <tr><td><a href="#print-callgraph-sccs">-print-callgraph-sccs</a></td><td>Print SCCs of the Call Graph</td></tr>
104 <tr><td><a href="#print-cfg-sccs">-print-cfg-sccs</a></td><td>Print SCCs of each function CFG</td></tr>
105 <tr><td><a href="#print-externalfnconstants">-print-externalfnconstants</a></td><td>Print external fn callsites passed constants</td></tr>
106 <tr><td><a href="#print-function">-print-function</a></td><td>Print function to stderr</td></tr>
107 <tr><td><a href="#print-module">-print-module</a></td><td>Print module to stderr</td></tr>
108 <tr><td><a href="#print-used-types">-print-used-types</a></td><td>Find Used Types</td></tr>
109 <tr><td><a href="#profile-loader">-profile-loader</a></td><td>Load profile information from llvmprof.out</td></tr>
110 <tr><td><a href="#scalar-evolution">-scalar-evolution</a></td><td>Scalar Evolution Analysis</td></tr>
111 <tr><td><a href="#targetdata">-targetdata</a></td><td>Target Data Layout</td></tr>
114 <tr><th colspan="2"><b>TRANSFORM PASSES</b></th></tr>
115 <tr><th>Option</th><th>Name</th></tr>
116 <tr><td><a href="#adce">-adce</a></td><td>Aggressive Dead Code Elimination</td></tr>
117 <tr><td><a href="#argpromotion">-argpromotion</a></td><td>Promote 'by reference' arguments to scalars</td></tr>
118 <tr><td><a href="#block-placement">-block-placement</a></td><td>Profile Guided Basic Block Placement</td></tr>
119 <tr><td><a href="#break-crit-edges">-break-crit-edges</a></td><td>Break critical edges in CFG</td></tr>
120 <tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Prepare a function for code generation </td></tr>
121 <tr><td><a href="#constmerge">-constmerge</a></td><td>Merge Duplicate Global Constants</td></tr>
122 <tr><td><a href="#constprop">-constprop</a></td><td>Simple constant propagation</td></tr>
123 <tr><td><a href="#dce">-dce</a></td><td>Dead Code Elimination</td></tr>
124 <tr><td><a href="#deadargelim">-deadargelim</a></td><td>Dead Argument Elimination</td></tr>
125 <tr><td><a href="#deadtypeelim">-deadtypeelim</a></td><td>Dead Type Elimination</td></tr>
126 <tr><td><a href="#die">-die</a></td><td>Dead Instruction Elimination</td></tr>
127 <tr><td><a href="#dse">-dse</a></td><td>Dead Store Elimination</td></tr>
128 <tr><td><a href="#globaldce">-globaldce</a></td><td>Dead Global Elimination</td></tr>
129 <tr><td><a href="#globalopt">-globalopt</a></td><td>Global Variable Optimizer</td></tr>
130 <tr><td><a href="#gvn">-gvn</a></td><td>Global Value Numbering</td></tr>
131 <tr><td><a href="#indvars">-indvars</a></td><td>Canonicalize Induction Variables</td></tr>
132 <tr><td><a href="#inline">-inline</a></td><td>Function Integration/Inlining</td></tr>
133 <tr><td><a href="#insert-block-profiling">-insert-block-profiling</a></td><td>Insert instrumentation for block profiling</td></tr>
134 <tr><td><a href="#insert-edge-profiling">-insert-edge-profiling</a></td><td>Insert instrumentation for edge profiling</td></tr>
135 <tr><td><a href="#insert-function-profiling">-insert-function-profiling</a></td><td>Insert instrumentation for function profiling</td></tr>
136 <tr><td><a href="#insert-null-profiling-rs">-insert-null-profiling-rs</a></td><td>Measure profiling framework overhead</td></tr>
137 <tr><td><a href="#insert-rs-profiling-framework">-insert-rs-profiling-framework</a></td><td>Insert random sampling instrumentation framework</td></tr>
138 <tr><td><a href="#instcombine">-instcombine</a></td><td>Combine redundant instructions</td></tr>
139 <tr><td><a href="#internalize">-internalize</a></td><td>Internalize Global Symbols</td></tr>
140 <tr><td><a href="#ipconstprop">-ipconstprop</a></td><td>Interprocedural constant propagation</td></tr>
141 <tr><td><a href="#ipsccp">-ipsccp</a></td><td>Interprocedural Sparse Conditional Constant Propagation</td></tr>
142 <tr><td><a href="#jump-threading">-jump-threading</a></td><td>Thread control through conditional blocks </td></tr>
143 <tr><td><a href="#lcssa">-lcssa</a></td><td>Loop-Closed SSA Form Pass</td></tr>
144 <tr><td><a href="#licm">-licm</a></td><td>Loop Invariant Code Motion</td></tr>
145 <tr><td><a href="#loop-deletion">-loop-deletion</a></td><td>Dead Loop Deletion Pass </td></tr>
146 <tr><td><a href="#loop-extract">-loop-extract</a></td><td>Extract loops into new functions</td></tr>
147 <tr><td><a href="#loop-extract-single">-loop-extract-single</a></td><td>Extract at most one loop into a new function</td></tr>
148 <tr><td><a href="#loop-index-split">-loop-index-split</a></td><td>Index Split Loops</td></tr>
149 <tr><td><a href="#loop-reduce">-loop-reduce</a></td><td>Loop Strength Reduction</td></tr>
150 <tr><td><a href="#loop-rotate">-loop-rotate</a></td><td>Rotate Loops</td></tr>
151 <tr><td><a href="#loop-unroll">-loop-unroll</a></td><td>Unroll loops</td></tr>
152 <tr><td><a href="#loop-unswitch">-loop-unswitch</a></td><td>Unswitch loops</td></tr>
153 <tr><td><a href="#loopsimplify">-loopsimplify</a></td><td>Canonicalize natural loops</td></tr>
154 <tr><td><a href="#lowerinvoke">-lowerinvoke</a></td><td>Lower invoke and unwind, for unwindless code generators</td></tr>
155 <tr><td><a href="#lowersetjmp">-lowersetjmp</a></td><td>Lower Set Jump</td></tr>
156 <tr><td><a href="#lowerswitch">-lowerswitch</a></td><td>Lower SwitchInst's to branches</td></tr>
157 <tr><td><a href="#mem2reg">-mem2reg</a></td><td>Promote Memory to Register</td></tr>
158 <tr><td><a href="#memcpyopt">-memcpyopt</a></td><td>Optimize use of memcpy and friends</td></tr>
159 <tr><td><a href="#mergereturn">-mergereturn</a></td><td>Unify function exit nodes</td></tr>
160 <tr><td><a href="#prune-eh">-prune-eh</a></td><td>Remove unused exception handling info</td></tr>
161 <tr><td><a href="#reassociate">-reassociate</a></td><td>Reassociate expressions</td></tr>
162 <tr><td><a href="#reg2mem">-reg2mem</a></td><td>Demote all values to stack slots</td></tr>
163 <tr><td><a href="#scalarrepl">-scalarrepl</a></td><td>Scalar Replacement of Aggregates</td></tr>
164 <tr><td><a href="#sccp">-sccp</a></td><td>Sparse Conditional Constant Propagation</td></tr>
165 <tr><td><a href="#simplify-libcalls">-simplify-libcalls</a></td><td>Simplify well-known library calls</td></tr>
166 <tr><td><a href="#simplifycfg">-simplifycfg</a></td><td>Simplify the CFG</td></tr>
167 <tr><td><a href="#strip">-strip</a></td><td>Strip all symbols from a module</td></tr>
168 <tr><td><a href="#strip-dead-prototypes">-strip-dead-prototypes</a></td><td>Remove unused function declarations</td></tr>
169 <tr><td><a href="#sretpromotion">-sretpromotion</a></td><td>Promote sret arguments</td></tr>
170 <tr><td><a href="#tailcallelim">-tailcallelim</a></td><td>Tail Call Elimination</td></tr>
171 <tr><td><a href="#tailduplicate">-tailduplicate</a></td><td>Tail Duplication</td></tr>
174 <tr><th colspan="2"><b>UTILITY PASSES</b></th></tr>
175 <tr><th>Option</th><th>Name</th></tr>
176 <tr><td><a href="#deadarghaX0r">-deadarghaX0r</a></td><td>Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</td></tr>
177 <tr><td><a href="#extract-blocks">-extract-blocks</a></td><td>Extract Basic Blocks From Module (for bugpoint use)</td></tr>
178 <tr><td><a href="#preverify">-preverify</a></td><td>Preliminary module verification</td></tr>
179 <tr><td><a href="#verify">-verify</a></td><td>Module Verifier</td></tr>
180 <tr><td><a href="#view-cfg">-view-cfg</a></td><td>View CFG of function</td></tr>
181 <tr><td><a href="#view-cfg-only">-view-cfg-only</a></td><td>View CFG of function (with no function bodies)</td></tr>
182 <tr><td><a href="#view-dom">-view-dom</a></td><td>View dominator tree of function</td></tr>
183 <tr><td><a href="#view-dom-only">-view-dom-only</a></td><td>View dominator tree of function (with no function bodies)</td></tr>
184 <tr><td><a href="#view-postdom">-view-postdom</a></td><td>View post dominator tree of function</td></tr>
185 <tr><td><a href="#view-postdom-only">-view-postdom-only</a></td><td>View post dominator tree of function (with no function bodies)</td></tr>
189 <!-- ======================================================================= -->
190 <div class="doc_section"> <a name="example">Analysis Passes</a></div>
191 <div class="doc_text">
192 <p>This section describes the LLVM Analysis Passes.</p>
195 <!-------------------------------------------------------------------------- -->
196 <div class="doc_subsection">
197 <a name="aa-eval">Exhaustive Alias Analysis Precision Evaluator</a>
199 <div class="doc_text">
200 <p>This is a simple N^2 alias analysis accuracy evaluator.
201 Basically, for each function in the program, it simply queries to see how the
202 alias analysis implementation answers alias queries between each pair of
203 pointers in the function.</p>
205 <p>This is inspired and adapted from code by: Naveen Neelakantam, Francesco
206 Spadini, and Wojciech Stryjewski.</p>
209 <!-------------------------------------------------------------------------- -->
210 <div class="doc_subsection">
211 <a name="basicaa">Basic Alias Analysis (default AA impl)</a>
213 <div class="doc_text">
215 This is the default implementation of the Alias Analysis interface
216 that simply implements a few identities (two different globals cannot alias,
217 etc), but otherwise does no analysis.
221 <!-------------------------------------------------------------------------- -->
222 <div class="doc_subsection">
223 <a name="basiccg">Basic CallGraph Construction</a>
225 <div class="doc_text">
226 <p>Yet to be written.</p>
229 <!-------------------------------------------------------------------------- -->
230 <div class="doc_subsection">
231 <a name="codegenprepare">Optimize for code generation</a>
233 <div class="doc_text">
235 This pass munges the code in the input function to better prepare it for
236 SelectionDAG-based code generation. This works around limitations in it's
237 basic-block-at-a-time approach. It should eventually be removed.
241 <!-------------------------------------------------------------------------- -->
242 <div class="doc_subsection">
243 <a name="count-aa">Count Alias Analysis Query Responses</a>
245 <div class="doc_text">
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 <!-------------------------------------------------------------------------- -->
253 <div class="doc_subsection">
254 <a name="debug-aa">AA use debugger</a>
256 <div class="doc_text">
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 <!-------------------------------------------------------------------------- -->
270 <div class="doc_subsection">
271 <a name="domfrontier">Dominance Frontier Construction</a>
273 <div class="doc_text">
275 This pass is a simple dominator construction algorithm for finding forward
280 <!-------------------------------------------------------------------------- -->
281 <div class="doc_subsection">
282 <a name="domtree">Dominator Tree Construction</a>
284 <div class="doc_text">
286 This pass is a simple dominator construction algorithm for finding forward
291 <!-------------------------------------------------------------------------- -->
292 <div class="doc_subsection">
293 <a name="dot-callgraph">Print Call Graph to 'dot' file</a>
295 <div class="doc_text">
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 <!-------------------------------------------------------------------------- -->
304 <div class="doc_subsection">
305 <a name="dot-cfg">Print CFG of function to 'dot' file</a>
307 <div class="doc_text">
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 <!-------------------------------------------------------------------------- -->
316 <div class="doc_subsection">
317 <a name="dot-cfg-only">Print CFG of function to 'dot' file (with no function bodies)</a>
319 <div class="doc_text">
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 <!-------------------------------------------------------------------------- -->
329 <div class="doc_subsection">
330 <a name="dot-dom">Print dominator tree of function to 'dot' file</a>
332 <div class="doc_text">
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 <!-------------------------------------------------------------------------- -->
341 <div class="doc_subsection">
342 <a name="dot-dom-only">Print dominator tree of function to 'dot' file (with no
345 <div class="doc_text">
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 <!-------------------------------------------------------------------------- -->
355 <div class="doc_subsection">
356 <a name="dot-postdom">Print post dominator tree of function to 'dot' file</a>
358 <div class="doc_text">
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 <!-------------------------------------------------------------------------- -->
367 <div class="doc_subsection">
368 <a name="dot-postdom-only">Print post dominator tree of function to 'dot' file
369 (with no function bodies)</a>
371 <div class="doc_text">
373 This pass, only available in <code>opt</code>, prints the post dominator tree
374 into a <code>.dot</code> graph, omitting the function bodies. This graph can
375 then be processed with the "dot" tool to convert it to postscript or some
376 other suitable format.
380 <!-------------------------------------------------------------------------- -->
381 <div class="doc_subsection">
382 <a name="globalsmodref-aa">Simple mod/ref analysis for globals</a>
384 <div class="doc_text">
386 This simple pass provides alias and mod/ref information for global values
387 that do not have their address taken, and keeps track of whether functions
388 read or write memory (are "pure"). For this simple (but very common) case,
389 we can provide pretty accurate and useful information.
393 <!-------------------------------------------------------------------------- -->
394 <div class="doc_subsection">
395 <a name="instcount">Counts the various types of Instructions</a>
397 <div class="doc_text">
399 This pass collects the count of all instructions and reports them
403 <!-------------------------------------------------------------------------- -->
404 <div class="doc_subsection">
405 <a name="intervals">Interval Partition Construction</a>
407 <div class="doc_text">
409 This analysis calculates and represents the interval partition of a function,
410 or a preexisting interval partition.
414 In this way, the interval partition may be used to reduce a flow graph down
415 to its degenerate single node interval partition (unless it is irreducible).
419 <!-------------------------------------------------------------------------- -->
420 <div class="doc_subsection">
421 <a name="loops">Natural Loop Construction</a>
423 <div class="doc_text">
425 This analysis is used to identify natural loops and determine the loop depth
426 of various nodes of the CFG. Note that the loops identified may actually be
427 several natural loops that share the same header node... not just a single
432 <!-------------------------------------------------------------------------- -->
433 <div class="doc_subsection">
434 <a name="memdep">Memory Dependence Analysis</a>
436 <div class="doc_text">
438 An analysis that determines, for a given memory operation, what preceding
439 memory operations it depends on. It builds on alias analysis information, and
440 tries to provide a lazy, caching interface to a common kind of alias
445 <!-------------------------------------------------------------------------- -->
446 <div class="doc_subsection">
447 <a name="no-aa">No Alias Analysis (always returns 'may' alias)</a>
449 <div class="doc_text">
451 Always returns "I don't know" for alias queries. NoAA is unlike other alias
452 analysis implementations, in that it does not chain to a previous analysis. As
453 such it doesn't follow many of the rules that other alias analyses must.
457 <!-------------------------------------------------------------------------- -->
458 <div class="doc_subsection">
459 <a name="no-profile">No Profile Information</a>
461 <div class="doc_text">
463 The default "no profile" implementation of the abstract
464 <code>ProfileInfo</code> interface.
468 <!-------------------------------------------------------------------------- -->
469 <div class="doc_subsection">
470 <a name="postdomfrontier">Post-Dominance Frontier Construction</a>
472 <div class="doc_text">
474 This pass is a simple post-dominator construction algorithm for finding
475 post-dominator frontiers.
479 <!-------------------------------------------------------------------------- -->
480 <div class="doc_subsection">
481 <a name="postdomtree">Post-Dominator Tree Construction</a>
483 <div class="doc_text">
485 This pass is a simple post-dominator construction algorithm for finding
490 <!-------------------------------------------------------------------------- -->
491 <div class="doc_subsection">
492 <a name="print-alias-sets">Alias Set Printer</a>
494 <div class="doc_text">
495 <p>Yet to be written.</p>
498 <!-------------------------------------------------------------------------- -->
499 <div class="doc_subsection">
500 <a name="print-callgraph">Print a call graph</a>
502 <div class="doc_text">
504 This pass, only available in <code>opt</code>, prints the call graph to
505 standard output in a human-readable form.
509 <!-------------------------------------------------------------------------- -->
510 <div class="doc_subsection">
511 <a name="print-callgraph-sccs">Print SCCs of the Call Graph</a>
513 <div class="doc_text">
515 This pass, only available in <code>opt</code>, prints the SCCs of the call
516 graph to standard output in a human-readable form.
520 <!-------------------------------------------------------------------------- -->
521 <div class="doc_subsection">
522 <a name="print-cfg-sccs">Print SCCs of each function CFG</a>
524 <div class="doc_text">
526 This pass, only available in <code>opt</code>, prints the SCCs of each
527 function CFG to standard output in a human-readable form.
531 <!-------------------------------------------------------------------------- -->
532 <div class="doc_subsection">
533 <a name="print-externalfnconstants">Print external fn callsites passed constants</a>
535 <div class="doc_text">
537 This pass, only available in <code>opt</code>, prints out call sites to
538 external functions that are called with constant arguments. This can be
539 useful when looking for standard library functions we should constant fold
540 or handle in alias analyses.
544 <!-------------------------------------------------------------------------- -->
545 <div class="doc_subsection">
546 <a name="print-function">Print function to stderr</a>
548 <div class="doc_text">
550 The <code>PrintFunctionPass</code> class is designed to be pipelined with
551 other <code>FunctionPass</code>es, and prints out the functions of the module
552 as they are processed.
556 <!-------------------------------------------------------------------------- -->
557 <div class="doc_subsection">
558 <a name="print-module">Print module to stderr</a>
560 <div class="doc_text">
562 This pass simply prints out the entire module when it is executed.
566 <!-------------------------------------------------------------------------- -->
567 <div class="doc_subsection">
568 <a name="print-used-types">Find Used Types</a>
570 <div class="doc_text">
572 This pass is used to seek out all of the types in use by the program. Note
573 that this analysis explicitly does not include types only used by the symbol
577 <!-------------------------------------------------------------------------- -->
578 <div class="doc_subsection">
579 <a name="profile-loader">Load profile information from llvmprof.out</a>
581 <div class="doc_text">
583 A concrete implementation of profiling information that loads the information
584 from a profile dump file.
588 <!-------------------------------------------------------------------------- -->
589 <div class="doc_subsection">
590 <a name="scalar-evolution">Scalar Evolution Analysis</a>
592 <div class="doc_text">
594 The <code>ScalarEvolution</code> analysis can be used to analyze and
595 catagorize scalar expressions in loops. It specializes in recognizing general
596 induction variables, representing them with the abstract and opaque
597 <code>SCEV</code> class. Given this analysis, trip counts of loops and other
598 important properties can be obtained.
602 This analysis is primarily useful for induction variable substitution and
607 <!-------------------------------------------------------------------------- -->
608 <div class="doc_subsection">
609 <a name="targetdata">Target Data Layout</a>
611 <div class="doc_text">
612 <p>Provides other passes access to information on how the size and alignment
613 required by the the target ABI for various data types.</p>
616 <!-- ======================================================================= -->
617 <div class="doc_section"> <a name="transform">Transform Passes</a></div>
618 <div class="doc_text">
619 <p>This section describes the LLVM Transform Passes.</p>
622 <!-------------------------------------------------------------------------- -->
623 <div class="doc_subsection">
624 <a name="adce">Aggressive Dead Code Elimination</a>
626 <div class="doc_text">
627 <p>ADCE aggressively tries to eliminate code. This pass is similar to
628 <a href="#dce">DCE</a> but it assumes that values are dead until proven
629 otherwise. This is similar to <a href="#sccp">SCCP</a>, except applied to
630 the liveness of values.</p>
633 <!-------------------------------------------------------------------------- -->
634 <div class="doc_subsection">
635 <a name="argpromotion">Promote 'by reference' arguments to scalars</a>
637 <div class="doc_text">
639 This pass promotes "by reference" arguments to be "by value" arguments. In
640 practice, this means looking for internal functions that have pointer
641 arguments. If it can prove, through the use of alias analysis, that an
642 argument is *only* loaded, then it can pass the value into the function
643 instead of the address of the value. This can cause recursive simplification
644 of code and lead to the elimination of allocas (especially in C++ template
649 This pass also handles aggregate arguments that are passed into a function,
650 scalarizing them if the elements of the aggregate are only loaded. Note that
651 it refuses to scalarize aggregates which would require passing in more than
652 three operands to the function, because passing thousands of operands for a
653 large array or structure is unprofitable!
657 Note that this transformation could also be done for arguments that are only
658 stored to (returning the value instead), but does not currently. This case
659 would be best handled when and if LLVM starts supporting multiple return
660 values from functions.
664 <!-------------------------------------------------------------------------- -->
665 <div class="doc_subsection">
666 <a name="block-placement">Profile Guided Basic Block Placement</a>
668 <div class="doc_text">
669 <p>This pass is a very simple profile guided basic block placement algorithm.
670 The idea is to put frequently executed blocks together at the start of the
671 function and hopefully increase the number of fall-through conditional
672 branches. If there is no profile information for a particular function, this
673 pass basically orders blocks in depth-first order.</p>
676 <!-------------------------------------------------------------------------- -->
677 <div class="doc_subsection">
678 <a name="break-crit-edges">Break critical edges in CFG</a>
680 <div class="doc_text">
682 Break all of the critical edges in the CFG by inserting a dummy basic block.
683 It may be "required" by passes that cannot deal with critical edges. This
684 transformation obviously invalidates the CFG, but can update forward dominator
685 (set, immediate dominators, tree, and frontier) information.
689 <!-------------------------------------------------------------------------- -->
690 <div class="doc_subsection">
691 <a name="codegenprepare">Prepare a function for code generation</a>
693 <div class="doc_text">
694 This pass munges the code in the input function to better prepare it for
695 SelectionDAG-based code generation. This works around limitations in it's
696 basic-block-at-a-time approach. It should eventually be removed.
699 <!-------------------------------------------------------------------------- -->
700 <div class="doc_subsection">
701 <a name="constmerge">Merge Duplicate Global Constants</a>
703 <div class="doc_text">
705 Merges duplicate global constants together into a single constant that is
706 shared. This is useful because some passes (ie TraceValues) insert a lot of
707 string constants into the program, regardless of whether or not an existing
712 <!-------------------------------------------------------------------------- -->
713 <div class="doc_subsection">
714 <a name="constprop">Simple constant propagation</a>
716 <div class="doc_text">
717 <p>This file implements constant propagation and merging. It looks for
718 instructions involving only constant operands and replaces them with a
719 constant value instead of an instruction. For example:</p>
720 <blockquote><pre>add i32 1, 2</pre></blockquote>
722 <blockquote><pre>i32 3</pre></blockquote>
723 <p>NOTE: this pass has a habit of making definitions be dead. It is a good
724 idea to to run a <a href="#die">DIE</a> (Dead Instruction Elimination) pass
725 sometime after running this pass.</p>
728 <!-------------------------------------------------------------------------- -->
729 <div class="doc_subsection">
730 <a name="dce">Dead Code Elimination</a>
732 <div class="doc_text">
734 Dead code elimination is similar to <a href="#die">dead instruction
735 elimination</a>, but it rechecks instructions that were used by removed
736 instructions to see if they are newly dead.
740 <!-------------------------------------------------------------------------- -->
741 <div class="doc_subsection">
742 <a name="deadargelim">Dead Argument Elimination</a>
744 <div class="doc_text">
746 This pass deletes dead arguments from internal functions. Dead argument
747 elimination removes arguments which are directly dead, as well as arguments
748 only passed into function calls as dead arguments of other functions. This
749 pass also deletes dead arguments in a similar way.
753 This pass is often useful as a cleanup pass to run after aggressive
754 interprocedural passes, which add possibly-dead arguments.
758 <!-------------------------------------------------------------------------- -->
759 <div class="doc_subsection">
760 <a name="deadtypeelim">Dead Type Elimination</a>
762 <div class="doc_text">
764 This pass is used to cleanup the output of GCC. It eliminate names for types
765 that are unused in the entire translation unit, using the <a
766 href="#findusedtypes">find used types</a> pass.
770 <!-------------------------------------------------------------------------- -->
771 <div class="doc_subsection">
772 <a name="die">Dead Instruction Elimination</a>
774 <div class="doc_text">
776 Dead instruction elimination performs a single pass over the function,
777 removing instructions that are obviously dead.
781 <!-------------------------------------------------------------------------- -->
782 <div class="doc_subsection">
783 <a name="dse">Dead Store Elimination</a>
785 <div class="doc_text">
787 A trivial dead store elimination that only considers basic-block local
792 <!-------------------------------------------------------------------------- -->
793 <div class="doc_subsection">
794 <a name="globaldce">Dead Global Elimination</a>
796 <div class="doc_text">
798 This transform is designed to eliminate unreachable internal globals from the
799 program. It uses an aggressive algorithm, searching out globals that are
800 known to be alive. After it finds all of the globals which are needed, it
801 deletes whatever is left over. This allows it to delete recursive chunks of
802 the program which are unreachable.
806 <!-------------------------------------------------------------------------- -->
807 <div class="doc_subsection">
808 <a name="globalopt">Global Variable Optimizer</a>
810 <div class="doc_text">
812 This pass transforms simple global variables that never have their address
813 taken. If obviously true, it marks read/write globals as constant, deletes
814 variables only stored to, etc.
818 <!-------------------------------------------------------------------------- -->
819 <div class="doc_subsection">
820 <a name="gvn">Global Value Numbering</a>
822 <div class="doc_text">
824 This pass performs global value numbering to eliminate fully and partially
825 redundant instructions. It also performs redundant load elimination.
830 <!-------------------------------------------------------------------------- -->
831 <div class="doc_subsection">
832 <a name="indvars">Canonicalize Induction Variables</a>
834 <div class="doc_text">
836 This transformation analyzes and transforms the induction variables (and
837 computations derived from them) into simpler forms suitable for subsequent
838 analysis and transformation.
842 This transformation makes the following changes to each loop with an
843 identifiable induction variable:
847 <li>All loops are transformed to have a <em>single</em> canonical
848 induction variable which starts at zero and steps by one.</li>
849 <li>The canonical induction variable is guaranteed to be the first PHI node
850 in the loop header block.</li>
851 <li>Any pointer arithmetic recurrences are raised to use array
856 If the trip count of a loop is computable, this pass also makes the following
861 <li>The exit condition for the loop is canonicalized to compare the
862 induction value against the exit value. This turns loops like:
863 <blockquote><pre>for (i = 7; i*i < 1000; ++i)</pre></blockquote>
865 <blockquote><pre>for (i = 0; i != 25; ++i)</pre></blockquote></li>
866 <li>Any use outside of the loop of an expression derived from the indvar
867 is changed to compute the derived value outside of the loop, eliminating
868 the dependence on the exit value of the induction variable. If the only
869 purpose of the loop is to compute the exit value of some derived
870 expression, this transformation will make the loop dead.</li>
874 This transformation should be followed by strength reduction after all of the
875 desired loop transformations have been performed. Additionally, on targets
876 where it is profitable, the loop could be transformed to count down to zero
877 (the "do loop" optimization).
881 <!-------------------------------------------------------------------------- -->
882 <div class="doc_subsection">
883 <a name="inline">Function Integration/Inlining</a>
885 <div class="doc_text">
887 Bottom-up inlining of functions into callees.
891 <!-------------------------------------------------------------------------- -->
892 <div class="doc_subsection">
893 <a name="insert-block-profiling">Insert instrumentation for block profiling</a>
895 <div class="doc_text">
897 This pass instruments the specified program with counters for basic block
898 profiling, which counts the number of times each basic block executes. This
899 is the most basic form of profiling, which can tell which blocks are hot, but
900 cannot reliably detect hot paths through the CFG.
904 Note that this implementation is very naïve. Control equivalent regions of
905 the CFG should not require duplicate counters, but it does put duplicate
910 <!-------------------------------------------------------------------------- -->
911 <div class="doc_subsection">
912 <a name="insert-edge-profiling">Insert instrumentation for edge profiling</a>
914 <div class="doc_text">
916 This pass instruments the specified program with counters for edge profiling.
917 Edge profiling can give a reasonable approximation of the hot paths through a
918 program, and is used for a wide variety of program transformations.
922 Note that this implementation is very naïve. It inserts a counter for
923 <em>every</em> edge in the program, instead of using control flow information
924 to prune the number of counters inserted.
928 <!-------------------------------------------------------------------------- -->
929 <div class="doc_subsection">
930 <a name="insert-function-profiling">Insert instrumentation for function profiling</a>
932 <div class="doc_text">
934 This pass instruments the specified program with counters for function
935 profiling, which counts the number of times each function is called.
939 <!-------------------------------------------------------------------------- -->
940 <div class="doc_subsection">
941 <a name="insert-null-profiling-rs">Measure profiling framework overhead</a>
943 <div class="doc_text">
945 The basic profiler that does nothing. It is the default profiler and thus
946 terminates <code>RSProfiler</code> chains. It is useful for measuring
951 <!-------------------------------------------------------------------------- -->
952 <div class="doc_subsection">
953 <a name="insert-rs-profiling-framework">Insert random sampling instrumentation framework</a>
955 <div class="doc_text">
957 The second stage of the random-sampling instrumentation framework, duplicates
958 all instructions in a function, ignoring the profiling code, then connects the
959 two versions together at the entry and at backedges. At each connection point
960 a choice is made as to whether to jump to the profiled code (take a sample) or
961 execute the unprofiled code.
965 After this pass, it is highly recommended to run<a href="#mem2reg">mem2reg</a>
966 and <a href="#adce">adce</a>. <a href="#instcombine">instcombine</a>,
967 <a href="#load-vn">load-vn</a>, <a href="#gdce">gdce</a>, and
968 <a href="#dse">dse</a> also are good to run afterwards.
972 <!-------------------------------------------------------------------------- -->
973 <div class="doc_subsection">
974 <a name="instcombine">Combine redundant instructions</a>
976 <div class="doc_text">
978 Combine instructions to form fewer, simple
979 instructions. This pass does not modify the CFG This pass is where algebraic
980 simplification happens.
984 This pass combines things like:
989 %Z = add i32 %Y, 1</pre></blockquote>
996 >%Z = add i32 %X, 2</pre></blockquote>
999 This is a simple worklist driven algorithm.
1003 This pass guarantees that the following canonicalizations are performed on
1008 <li>If a binary operator has a constant operand, it is moved to the right-
1010 <li>Bitwise operators with constant operands are always grouped so that
1011 shifts are performed first, then <code>or</code>s, then
1012 <code>and</code>s, then <code>xor</code>s.</li>
1013 <li>Compare instructions are converted from <code><</code>,
1014 <code>></code>, <code>≤</code>, or <code>≥</code> to
1015 <code>=</code> or <code>≠</code> if possible.</li>
1016 <li>All <code>cmp</code> instructions on boolean values are replaced with
1017 logical operations.</li>
1018 <li><code>add <var>X</var>, <var>X</var></code> is represented as
1019 <code>mul <var>X</var>, 2</code> ⇒ <code>shl <var>X</var>, 1</code></li>
1020 <li>Multiplies with a constant power-of-two argument are transformed into
1026 <!-------------------------------------------------------------------------- -->
1027 <div class="doc_subsection">
1028 <a name="internalize">Internalize Global Symbols</a>
1030 <div class="doc_text">
1032 This pass loops over all of the functions in the input module, looking for a
1033 main function. If a main function is found, all other functions and all
1034 global variables with initializers are marked as internal.
1038 <!-------------------------------------------------------------------------- -->
1039 <div class="doc_subsection">
1040 <a name="ipconstprop">Interprocedural constant propagation</a>
1042 <div class="doc_text">
1044 This pass implements an <em>extremely</em> simple interprocedural constant
1045 propagation pass. It could certainly be improved in many different ways,
1046 like using a worklist. This pass makes arguments dead, but does not remove
1047 them. The existing dead argument elimination pass should be run after this
1048 to clean up the mess.
1052 <!-------------------------------------------------------------------------- -->
1053 <div class="doc_subsection">
1054 <a name="ipsccp">Interprocedural Sparse Conditional Constant Propagation</a>
1056 <div class="doc_text">
1058 An interprocedural variant of <a href="#sccp">Sparse Conditional Constant
1063 <!-------------------------------------------------------------------------- -->
1064 <div class="doc_subsection">
1065 <a name="jump-threading">Thread control through conditional blocks</a>
1067 <div class="doc_text">
1069 Jump threading tries to find distinct threads of control flow running through
1070 a basic block. This pass looks at blocks that have multiple predecessors and
1071 multiple successors. If one or more of the predecessors of the block can be
1072 proven to always cause a jump to one of the successors, we forward the edge
1073 from the predecessor to the successor by duplicating the contents of this
1077 An example of when this can occur is code like this:
1084 if (X < 3) {</pre>
1087 In this case, the unconditional branch at the end of the first if can be
1088 revectored to the false side of the second if.
1092 <!-------------------------------------------------------------------------- -->
1093 <div class="doc_subsection">
1094 <a name="lcssa">Loop-Closed SSA Form Pass</a>
1096 <div class="doc_text">
1098 This pass transforms loops by placing phi nodes at the end of the loops for
1099 all values that are live across the loop boundary. For example, it turns
1100 the left into the right code:
1104 >for (...) for (...)
1109 X3 = phi(X1, X2) X3 = phi(X1, X2)
1110 ... = X3 + 4 X4 = phi(X3)
1114 This is still valid LLVM; the extra phi nodes are purely redundant, and will
1115 be trivially eliminated by <code>InstCombine</code>. The major benefit of
1116 this transformation is that it makes many other loop optimizations, such as
1117 LoopUnswitching, simpler.
1121 <!-------------------------------------------------------------------------- -->
1122 <div class="doc_subsection">
1123 <a name="licm">Loop Invariant Code Motion</a>
1125 <div class="doc_text">
1127 This pass performs loop invariant code motion, attempting to remove as much
1128 code from the body of a loop as possible. It does this by either hoisting
1129 code into the preheader block, or by sinking code to the exit blocks if it is
1130 safe. This pass also promotes must-aliased memory locations in the loop to
1131 live in registers, thus hoisting and sinking "invariant" loads and stores.
1135 This pass uses alias analysis for two purposes:
1139 <li>Moving loop invariant loads and calls out of loops. If we can determine
1140 that a load or call inside of a loop never aliases anything stored to,
1141 we can hoist it or sink it like any other instruction.</li>
1142 <li>Scalar Promotion of Memory - If there is a store instruction inside of
1143 the loop, we try to move the store to happen AFTER the loop instead of
1144 inside of the loop. This can only happen if a few conditions are true:
1146 <li>The pointer stored through is loop invariant.</li>
1147 <li>There are no stores or loads in the loop which <em>may</em> alias
1148 the pointer. There are no calls in the loop which mod/ref the
1151 If these conditions are true, we can promote the loads and stores in the
1152 loop of the pointer to use a temporary alloca'd variable. We then use
1153 the mem2reg functionality to construct the appropriate SSA form for the
1157 <!-------------------------------------------------------------------------- -->
1158 <div class="doc_subsection">
1159 <a name="loop-deletion">Dead Loop Deletion Pass</a>
1161 <div class="doc_text">
1163 This file implements the Dead Loop Deletion Pass. This pass is responsible
1164 for eliminating loops with non-infinite computable trip counts that have no
1165 side effects or volatile instructions, and do not contribute to the
1166 computation of the function's return value.
1170 <!-------------------------------------------------------------------------- -->
1171 <div class="doc_subsection">
1172 <a name="loop-extract">Extract loops into new functions</a>
1174 <div class="doc_text">
1176 A pass wrapper around the <code>ExtractLoop()</code> scalar transformation to
1177 extract each top-level loop into its own new function. If the loop is the
1178 <em>only</em> loop in a given function, it is not touched. This is a pass most
1179 useful for debugging via bugpoint.
1183 <!-------------------------------------------------------------------------- -->
1184 <div class="doc_subsection">
1185 <a name="loop-extract-single">Extract at most one loop into a new function</a>
1187 <div class="doc_text">
1189 Similar to <a href="#loop-extract">Extract loops into new functions</a>,
1190 this pass extracts one natural loop from the program into a function if it
1191 can. This is used by bugpoint.
1195 <!-------------------------------------------------------------------------- -->
1196 <div class="doc_subsection">
1197 <a name="loop-index-split">Index Split Loops</a>
1199 <div class="doc_text">
1201 This pass divides loop's iteration range by spliting loop such that each
1202 individual loop is executed efficiently.
1206 <!-------------------------------------------------------------------------- -->
1207 <div class="doc_subsection">
1208 <a name="loop-reduce">Loop Strength Reduction</a>
1210 <div class="doc_text">
1212 This pass performs a strength reduction on array references inside loops that
1213 have as one or more of their components the loop induction variable. This is
1214 accomplished by creating a new value to hold the initial value of the array
1215 access for the first iteration, and then creating a new GEP instruction in
1216 the loop to increment the value by the appropriate amount.
1220 <!-------------------------------------------------------------------------- -->
1221 <div class="doc_subsection">
1222 <a name="loop-rotate">Rotate Loops</a>
1224 <div class="doc_text">
1225 <p>A simple loop rotation transformation.</p>
1228 <!-------------------------------------------------------------------------- -->
1229 <div class="doc_subsection">
1230 <a name="loop-unroll">Unroll loops</a>
1232 <div class="doc_text">
1234 This pass implements a simple loop unroller. It works best when loops have
1235 been canonicalized by the <a href="#indvars"><tt>-indvars</tt></a> pass,
1236 allowing it to determine the trip counts of loops easily.
1240 <!-------------------------------------------------------------------------- -->
1241 <div class="doc_subsection">
1242 <a name="loop-unswitch">Unswitch loops</a>
1244 <div class="doc_text">
1246 This pass transforms loops that contain branches on loop-invariant conditions
1247 to have multiple loops. For example, it turns the left into the right code:
1259 This can increase the size of the code exponentially (doubling it every time
1260 a loop is unswitched) so we only unswitch if the resultant code will be
1261 smaller than a threshold.
1265 This pass expects LICM to be run before it to hoist invariant conditions out
1266 of the loop, to make the unswitching opportunity obvious.
1270 <!-------------------------------------------------------------------------- -->
1271 <div class="doc_subsection">
1272 <a name="loopsimplify">Canonicalize natural loops</a>
1274 <div class="doc_text">
1276 This pass performs several transformations to transform natural loops into a
1277 simpler form, which makes subsequent analyses and transformations simpler and
1282 Loop pre-header insertion guarantees that there is a single, non-critical
1283 entry edge from outside of the loop to the loop header. This simplifies a
1284 number of analyses and transformations, such as LICM.
1288 Loop exit-block insertion guarantees that all exit blocks from the loop
1289 (blocks which are outside of the loop that have predecessors inside of the
1290 loop) only have predecessors from inside of the loop (and are thus dominated
1291 by the loop header). This simplifies transformations such as store-sinking
1292 that are built into LICM.
1296 This pass also guarantees that loops will have exactly one backedge.
1300 Note that the simplifycfg pass will clean up blocks which are split out but
1301 end up being unnecessary, so usage of this pass should not pessimize
1306 This pass obviously modifies the CFG, but updates loop information and
1307 dominator information.
1311 <!-------------------------------------------------------------------------- -->
1312 <div class="doc_subsection">
1313 <a name="lowerinvoke">Lower invoke and unwind, for unwindless code generators</a>
1315 <div class="doc_text">
1317 This transformation is designed for use by code generators which do not yet
1318 support stack unwinding. This pass supports two models of exception handling
1319 lowering, the 'cheap' support and the 'expensive' support.
1323 'Cheap' exception handling support gives the program the ability to execute
1324 any program which does not "throw an exception", by turning 'invoke'
1325 instructions into calls and by turning 'unwind' instructions into calls to
1326 abort(). If the program does dynamically use the unwind instruction, the
1327 program will print a message then abort.
1331 'Expensive' exception handling support gives the full exception handling
1332 support to the program at the cost of making the 'invoke' instruction
1333 really expensive. It basically inserts setjmp/longjmp calls to emulate the
1334 exception handling as necessary.
1338 Because the 'expensive' support slows down programs a lot, and EH is only
1339 used for a subset of the programs, it must be specifically enabled by the
1340 <tt>-enable-correct-eh-support</tt> option.
1344 Note that after this pass runs the CFG is not entirely accurate (exceptional
1345 control flow edges are not correct anymore) so only very simple things should
1346 be done after the lowerinvoke pass has run (like generation of native code).
1347 This should not be used as a general purpose "my LLVM-to-LLVM pass doesn't
1348 support the invoke instruction yet" lowering pass.
1352 <!-------------------------------------------------------------------------- -->
1353 <div class="doc_subsection">
1354 <a name="lowersetjmp">Lower Set Jump</a>
1356 <div class="doc_text">
1358 Lowers <tt>setjmp</tt> and <tt>longjmp</tt> to use the LLVM invoke and unwind
1359 instructions as necessary.
1363 Lowering of <tt>longjmp</tt> is fairly trivial. We replace the call with a
1364 call to the LLVM library function <tt>__llvm_sjljeh_throw_longjmp()</tt>.
1365 This unwinds the stack for us calling all of the destructors for
1366 objects allocated on the stack.
1370 At a <tt>setjmp</tt> call, the basic block is split and the <tt>setjmp</tt>
1371 removed. The calls in a function that have a <tt>setjmp</tt> are converted to
1372 invoke where the except part checks to see if it's a <tt>longjmp</tt>
1373 exception and, if so, if it's handled in the function. If it is, then it gets
1374 the value returned by the <tt>longjmp</tt> and goes to where the basic block
1375 was split. <tt>invoke</tt> instructions are handled in a similar fashion with
1376 the original except block being executed if it isn't a <tt>longjmp</tt>
1377 except that is handled by that function.
1381 <!-------------------------------------------------------------------------- -->
1382 <div class="doc_subsection">
1383 <a name="lowerswitch">Lower SwitchInst's to branches</a>
1385 <div class="doc_text">
1387 Rewrites <tt>switch</tt> instructions with a sequence of branches, which
1388 allows targets to get away with not implementing the switch instruction until
1393 <!-------------------------------------------------------------------------- -->
1394 <div class="doc_subsection">
1395 <a name="mem2reg">Promote Memory to Register</a>
1397 <div class="doc_text">
1399 This file promotes memory references to be register references. It promotes
1400 <tt>alloca</tt> instructions which only have <tt>load</tt>s and
1401 <tt>store</tt>s as uses. An <tt>alloca</tt> is transformed by using dominator
1402 frontiers to place <tt>phi</tt> nodes, then traversing the function in
1403 depth-first order to rewrite <tt>load</tt>s and <tt>store</tt>s as
1404 appropriate. This is just the standard SSA construction algorithm to construct
1409 <!-------------------------------------------------------------------------- -->
1410 <div class="doc_subsection">
1411 <a name="memcpyopt">Optimize use of memcpy and friend</a>
1413 <div class="doc_text">
1415 This pass performs various transformations related to eliminating memcpy
1416 calls, or transforming sets of stores into memset's.
1420 <!-------------------------------------------------------------------------- -->
1421 <div class="doc_subsection">
1422 <a name="mergereturn">Unify function exit nodes</a>
1424 <div class="doc_text">
1426 Ensure that functions have at most one <tt>ret</tt> instruction in them.
1427 Additionally, it keeps track of which node is the new exit node of the CFG.
1431 <!-------------------------------------------------------------------------- -->
1432 <div class="doc_subsection">
1433 <a name="prune-eh">Remove unused exception handling info</a>
1435 <div class="doc_text">
1437 This file implements a simple interprocedural pass which walks the call-graph,
1438 turning <tt>invoke</tt> instructions into <tt>call</tt> instructions if and
1439 only if the callee cannot throw an exception. It implements this as a
1440 bottom-up traversal of the call-graph.
1444 <!-------------------------------------------------------------------------- -->
1445 <div class="doc_subsection">
1446 <a name="reassociate">Reassociate expressions</a>
1448 <div class="doc_text">
1450 This pass reassociates commutative expressions in an order that is designed
1451 to promote better constant propagation, GCSE, LICM, PRE, etc.
1455 For example: 4 + (<var>x</var> + 5) ⇒ <var>x</var> + (4 + 5)
1459 In the implementation of this algorithm, constants are assigned rank = 0,
1460 function arguments are rank = 1, and other values are assigned ranks
1461 corresponding to the reverse post order traversal of current function
1462 (starting at 2), which effectively gives values in deep loops higher rank
1463 than values not in loops.
1467 <!-------------------------------------------------------------------------- -->
1468 <div class="doc_subsection">
1469 <a name="reg2mem">Demote all values to stack slots</a>
1471 <div class="doc_text">
1473 This file demotes all registers to memory references. It is intented to be
1474 the inverse of <a href="#mem2reg"><tt>-mem2reg</tt></a>. By converting to
1475 <tt>load</tt> instructions, the only values live across basic blocks are
1476 <tt>alloca</tt> instructions and <tt>load</tt> instructions before
1477 <tt>phi</tt> nodes. It is intended that this should make CFG hacking much
1478 easier. To make later hacking easier, the entry block is split into two, such
1479 that all introduced <tt>alloca</tt> instructions (and nothing else) are in the
1484 <!-------------------------------------------------------------------------- -->
1485 <div class="doc_subsection">
1486 <a name="scalarrepl">Scalar Replacement of Aggregates</a>
1488 <div class="doc_text">
1490 The well-known scalar replacement of aggregates transformation. This
1491 transform breaks up <tt>alloca</tt> instructions of aggregate type (structure
1492 or array) into individual <tt>alloca</tt> instructions for each member if
1493 possible. Then, if possible, it transforms the individual <tt>alloca</tt>
1494 instructions into nice clean scalar SSA form.
1498 This combines a simple scalar replacement of aggregates algorithm with the <a
1499 href="#mem2reg"><tt>mem2reg</tt></a> algorithm because often interact,
1500 especially for C++ programs. As such, iterating between <tt>scalarrepl</tt>,
1501 then <a href="#mem2reg"><tt>mem2reg</tt></a> until we run out of things to
1506 <!-------------------------------------------------------------------------- -->
1507 <div class="doc_subsection">
1508 <a name="sccp">Sparse Conditional Constant Propagation</a>
1510 <div class="doc_text">
1512 Sparse conditional constant propagation and merging, which can be summarized
1517 <li>Assumes values are constant unless proven otherwise</li>
1518 <li>Assumes BasicBlocks are dead unless proven otherwise</li>
1519 <li>Proves values to be constant, and replaces them with constants</li>
1520 <li>Proves conditional branches to be unconditional</li>
1524 Note that this pass has a habit of making definitions be dead. It is a good
1525 idea to to run a DCE pass sometime after running this pass.
1529 <!-------------------------------------------------------------------------- -->
1530 <div class="doc_subsection">
1531 <a name="simplify-libcalls">Simplify well-known library calls</a>
1533 <div class="doc_text">
1535 Applies a variety of small optimizations for calls to specific well-known
1536 function calls (e.g. runtime library functions). For example, a call
1537 <tt>exit(3)</tt> that occurs within the <tt>main()</tt> function can be
1538 transformed into simply <tt>return 3</tt>.
1542 <!-------------------------------------------------------------------------- -->
1543 <div class="doc_subsection">
1544 <a name="simplifycfg">Simplify the CFG</a>
1546 <div class="doc_text">
1548 Performs dead code elimination and basic block merging. Specifically:
1552 <li>Removes basic blocks with no predecessors.</li>
1553 <li>Merges a basic block into its predecessor if there is only one and the
1554 predecessor only has one successor.</li>
1555 <li>Eliminates PHI nodes for basic blocks with a single predecessor.</li>
1556 <li>Eliminates a basic block that only contains an unconditional
1561 <!-------------------------------------------------------------------------- -->
1562 <div class="doc_subsection">
1563 <a name="strip">Strip all symbols from a module</a>
1565 <div class="doc_text">
1567 Performs code stripping. This transformation can delete:
1571 <li>names for virtual registers</li>
1572 <li>symbols for internal globals and functions</li>
1573 <li>debug information</li>
1577 Note that this transformation makes code much less readable, so it should
1578 only be used in situations where the <tt>strip</tt> utility would be used,
1579 such as reducing code size or making it harder to reverse engineer code.
1583 <!-------------------------------------------------------------------------- -->
1584 <div class="doc_subsection">
1585 <a name="strip-dead-prototypes">Remove unused function declarations</a>
1587 <div class="doc_text">
1589 This pass loops over all of the functions in the input module, looking for
1590 dead declarations and removes them. Dead declarations are declarations of
1591 functions for which no implementation is available (i.e., declarations for
1592 unused library functions).
1596 <!-------------------------------------------------------------------------- -->
1597 <div class="doc_subsection">
1598 <a name="sretpromotion">Promote sret arguments</a>
1600 <div class="doc_text">
1602 This pass finds functions that return a struct (using a pointer to the struct
1603 as the first argument of the function, marked with the '<tt>sret</tt>' attribute) and
1604 replaces them with a new function that simply returns each of the elements of
1605 that struct (using multiple return values).
1609 This pass works under a number of conditions:
1613 <li>The returned struct must not contain other structs</li>
1614 <li>The returned struct must only be used to load values from</li>
1615 <li>The placeholder struct passed in is the result of an <tt>alloca</tt></li>
1619 <!-------------------------------------------------------------------------- -->
1620 <div class="doc_subsection">
1621 <a name="tailcallelim">Tail Call Elimination</a>
1623 <div class="doc_text">
1625 This file transforms calls of the current function (self recursion) followed
1626 by a return instruction with a branch to the entry of the function, creating
1627 a loop. This pass also implements the following extensions to the basic
1632 <li>Trivial instructions between the call and return do not prevent the
1633 transformation from taking place, though currently the analysis cannot
1634 support moving any really useful instructions (only dead ones).
1635 <li>This pass transforms functions that are prevented from being tail
1636 recursive by an associative expression to use an accumulator variable,
1637 thus compiling the typical naive factorial or <tt>fib</tt> implementation
1638 into efficient code.
1639 <li>TRE is performed if the function returns void, if the return
1640 returns the result returned by the call, or if the function returns a
1641 run-time constant on all exits from the function. It is possible, though
1642 unlikely, that the return returns something else (like constant 0), and
1643 can still be TRE'd. It can be TRE'd if <em>all other</em> return
1644 instructions in the function return the exact same value.
1645 <li>If it can prove that callees do not access theier caller stack frame,
1646 they are marked as eligible for tail call elimination (by the code
1651 <!-------------------------------------------------------------------------- -->
1652 <div class="doc_subsection">
1653 <a name="tailduplicate">Tail Duplication</a>
1655 <div class="doc_text">
1657 This pass performs a limited form of tail duplication, intended to simplify
1658 CFGs by removing some unconditional branches. This pass is necessary to
1659 straighten out loops created by the C front-end, but also is capable of
1660 making other code nicer. After this pass is run, the CFG simplify pass
1661 should be run to clean up the mess.
1665 <!-- ======================================================================= -->
1666 <div class="doc_section"> <a name="transform">Utility Passes</a></div>
1667 <div class="doc_text">
1668 <p>This section describes the LLVM Utility Passes.</p>
1671 <!-------------------------------------------------------------------------- -->
1672 <div class="doc_subsection">
1673 <a name="deadarghaX0r">Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</a>
1675 <div class="doc_text">
1677 Same as dead argument elimination, but deletes arguments to functions which
1678 are external. This is only for use by <a
1679 href="Bugpoint.html">bugpoint</a>.</p>
1682 <!-------------------------------------------------------------------------- -->
1683 <div class="doc_subsection">
1684 <a name="extract-blocks">Extract Basic Blocks From Module (for bugpoint use)</a>
1686 <div class="doc_text">
1688 This pass is used by bugpoint to extract all blocks from the module into their
1692 <!-------------------------------------------------------------------------- -->
1693 <div class="doc_subsection">
1694 <a name="preverify">Preliminary module verification</a>
1696 <div class="doc_text">
1698 Ensures that the module is in the form required by the <a
1699 href="#verifier">Module Verifier</a> pass.
1703 Running the verifier runs this pass automatically, so there should be no need
1708 <!-------------------------------------------------------------------------- -->
1709 <div class="doc_subsection">
1710 <a name="verify">Module Verifier</a>
1712 <div class="doc_text">
1714 Verifies an LLVM IR code. This is useful to run after an optimization which is
1715 undergoing testing. Note that <tt>llvm-as</tt> verifies its input before
1716 emitting bitcode, and also that malformed bitcode is likely to make LLVM
1717 crash. All language front-ends are therefore encouraged to verify their output
1718 before performing optimizing transformations.
1722 <li>Both of a binary operator's parameters are of the same type.</li>
1723 <li>Verify that the indices of mem access instructions match other
1725 <li>Verify that arithmetic and other things are only performed on
1726 first-class types. Verify that shifts and logicals only happen on
1728 <li>All of the constants in a switch statement are of the correct type.</li>
1729 <li>The code is in valid SSA form.</li>
1730 <li>It is illegal to put a label into any other type (like a structure) or
1732 <li>Only phi nodes can be self referential: <tt>%x = add i32 %x, %x</tt> is
1734 <li>PHI nodes must have an entry for each predecessor, with no extras.</li>
1735 <li>PHI nodes must be the first thing in a basic block, all grouped
1737 <li>PHI nodes must have at least one entry.</li>
1738 <li>All basic blocks should only end with terminator insts, not contain
1740 <li>The entry node to a function must not have predecessors.</li>
1741 <li>All Instructions must be embedded into a basic block.</li>
1742 <li>Functions cannot take a void-typed parameter.</li>
1743 <li>Verify that a function's argument list agrees with its declared
1745 <li>It is illegal to specify a name for a void value.</li>
1746 <li>It is illegal to have a internal global value with no initializer.</li>
1747 <li>It is illegal to have a ret instruction that returns a value that does
1748 not agree with the function return value type.</li>
1749 <li>Function call argument types match the function prototype.</li>
1750 <li>All other things that are tested by asserts spread about the code.</li>
1754 Note that this does not provide full security verification (like Java), but
1755 instead just tries to ensure that code is well-formed.
1759 <!-------------------------------------------------------------------------- -->
1760 <div class="doc_subsection">
1761 <a name="view-cfg">View CFG of function</a>
1763 <div class="doc_text">
1765 Displays the control flow graph using the GraphViz tool.
1769 <!-------------------------------------------------------------------------- -->
1770 <div class="doc_subsection">
1771 <a name="view-cfg-only">View CFG of function (with no function bodies)</a>
1773 <div class="doc_text">
1775 Displays the control flow graph using the GraphViz tool, but omitting function
1780 <!-------------------------------------------------------------------------- -->
1781 <div class="doc_subsection">
1782 <a name="view-dom">View dominator tree of function</a>
1784 <div class="doc_text">
1786 Displays the dominator tree using the GraphViz tool.
1790 <!-------------------------------------------------------------------------- -->
1791 <div class="doc_subsection">
1792 <a name="view-dom-only">View dominator tree of function (with no function
1796 <div class="doc_text">
1798 Displays the dominator tree using the GraphViz tool, but omitting function
1803 <!-------------------------------------------------------------------------- -->
1804 <div class="doc_subsection">
1805 <a name="view-postdom">View post dominator tree of function</a>
1807 <div class="doc_text">
1809 Displays the post dominator tree using the GraphViz tool.
1813 <!-------------------------------------------------------------------------- -->
1814 <div class="doc_subsection">
1815 <a name="view-postdom-only">View post dominator tree of function (with no
1819 <div class="doc_text">
1821 Displays the post dominator tree using the GraphViz tool, but omitting
1826 <!-- *********************************************************************** -->
1830 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1831 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1832 <a href="http://validator.w3.org/check/referer"><img
1833 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1835 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a><br>
1836 <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
1837 Last modified: $Date$