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="#globalsmodref-aa">-globalsmodref-aa</a></td><td>Simple mod/ref analysis for globals</td></tr>
89 <tr><td><a href="#instcount">-instcount</a></td><td>Counts the various types of Instructions</td></tr>
90 <tr><td><a href="#intervals">-intervals</a></td><td>Interval Partition Construction</td></tr>
91 <tr><td><a href="#loops">-loops</a></td><td>Natural Loop Construction</td></tr>
92 <tr><td><a href="#memdep">-memdep</a></td><td>Memory Dependence Analysis</td></tr>
93 <tr><td><a href="#no-aa">-no-aa</a></td><td>No Alias Analysis (always returns 'may' alias)</td></tr>
94 <tr><td><a href="#no-profile">-no-profile</a></td><td>No Profile Information</td></tr>
95 <tr><td><a href="#postdomfrontier">-postdomfrontier</a></td><td>Post-Dominance Frontier Construction</td></tr>
96 <tr><td><a href="#postdomtree">-postdomtree</a></td><td>Post-Dominator Tree Construction</td></tr>
97 <tr><td><a href="#print-alias-sets">-print-alias-sets</a></td><td>Alias Set Printer</td></tr>
98 <tr><td><a href="#print-callgraph">-print-callgraph</a></td><td>Print a call graph</td></tr>
99 <tr><td><a href="#print-callgraph-sccs">-print-callgraph-sccs</a></td><td>Print SCCs of the Call Graph</td></tr>
100 <tr><td><a href="#print-cfg-sccs">-print-cfg-sccs</a></td><td>Print SCCs of each function CFG</td></tr>
101 <tr><td><a href="#print-externalfnconstants">-print-externalfnconstants</a></td><td>Print external fn callsites passed constants</td></tr>
102 <tr><td><a href="#print-function">-print-function</a></td><td>Print function to stderr</td></tr>
103 <tr><td><a href="#print-module">-print-module</a></td><td>Print module to stderr</td></tr>
104 <tr><td><a href="#print-used-types">-print-used-types</a></td><td>Find Used Types</td></tr>
105 <tr><td><a href="#profile-loader">-profile-loader</a></td><td>Load profile information from llvmprof.out</td></tr>
106 <tr><td><a href="#scalar-evolution">-scalar-evolution</a></td><td>Scalar Evolution Analysis</td></tr>
107 <tr><td><a href="#targetdata">-targetdata</a></td><td>Target Data Layout</td></tr>
110 <tr><th colspan="2"><b>TRANSFORM PASSES</b></th></tr>
111 <tr><th>Option</th><th>Name</th></tr>
112 <tr><td><a href="#adce">-adce</a></td><td>Aggressive Dead Code Elimination</td></tr>
113 <tr><td><a href="#argpromotion">-argpromotion</a></td><td>Promote 'by reference' arguments to scalars</td></tr>
114 <tr><td><a href="#block-placement">-block-placement</a></td><td>Profile Guided Basic Block Placement</td></tr>
115 <tr><td><a href="#break-crit-edges">-break-crit-edges</a></td><td>Break critical edges in CFG</td></tr>
116 <tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Prepare a function for code generation </td></tr>
117 <tr><td><a href="#constmerge">-constmerge</a></td><td>Merge Duplicate Global Constants</td></tr>
118 <tr><td><a href="#constprop">-constprop</a></td><td>Simple constant propagation</td></tr>
119 <tr><td><a href="#dce">-dce</a></td><td>Dead Code Elimination</td></tr>
120 <tr><td><a href="#deadargelim">-deadargelim</a></td><td>Dead Argument Elimination</td></tr>
121 <tr><td><a href="#deadtypeelim">-deadtypeelim</a></td><td>Dead Type Elimination</td></tr>
122 <tr><td><a href="#die">-die</a></td><td>Dead Instruction Elimination</td></tr>
123 <tr><td><a href="#dse">-dse</a></td><td>Dead Store Elimination</td></tr>
124 <tr><td><a href="#globaldce">-globaldce</a></td><td>Dead Global Elimination</td></tr>
125 <tr><td><a href="#globalopt">-globalopt</a></td><td>Global Variable Optimizer</td></tr>
126 <tr><td><a href="#gvn">-gvn</a></td><td>Global Value Numbering</td></tr>
127 <tr><td><a href="#indmemrem">-indmemrem</a></td><td>Indirect Malloc and Free Removal</td></tr>
128 <tr><td><a href="#indvars">-indvars</a></td><td>Canonicalize Induction Variables</td></tr>
129 <tr><td><a href="#inline">-inline</a></td><td>Function Integration/Inlining</td></tr>
130 <tr><td><a href="#insert-block-profiling">-insert-block-profiling</a></td><td>Insert instrumentation for block profiling</td></tr>
131 <tr><td><a href="#insert-edge-profiling">-insert-edge-profiling</a></td><td>Insert instrumentation for edge profiling</td></tr>
132 <tr><td><a href="#insert-function-profiling">-insert-function-profiling</a></td><td>Insert instrumentation for function profiling</td></tr>
133 <tr><td><a href="#insert-null-profiling-rs">-insert-null-profiling-rs</a></td><td>Measure profiling framework overhead</td></tr>
134 <tr><td><a href="#insert-rs-profiling-framework">-insert-rs-profiling-framework</a></td><td>Insert random sampling instrumentation framework</td></tr>
135 <tr><td><a href="#instcombine">-instcombine</a></td><td>Combine redundant instructions</td></tr>
136 <tr><td><a href="#internalize">-internalize</a></td><td>Internalize Global Symbols</td></tr>
137 <tr><td><a href="#ipconstprop">-ipconstprop</a></td><td>Interprocedural constant propagation</td></tr>
138 <tr><td><a href="#ipsccp">-ipsccp</a></td><td>Interprocedural Sparse Conditional Constant Propagation</td></tr>
139 <tr><td><a href="#jump-threading">-jump-threading</a></td><td>Thread control through conditional blocks </td></tr>
140 <tr><td><a href="#lcssa">-lcssa</a></td><td>Loop-Closed SSA Form Pass</td></tr>
141 <tr><td><a href="#licm">-licm</a></td><td>Loop Invariant Code Motion</td></tr>
142 <tr><td><a href="#loop-deletion">-loop-deletion</a></td><td>Dead Loop Deletion Pass </td></tr>
143 <tr><td><a href="#loop-extract">-loop-extract</a></td><td>Extract loops into new functions</td></tr>
144 <tr><td><a href="#loop-extract-single">-loop-extract-single</a></td><td>Extract at most one loop into a new function</td></tr>
145 <tr><td><a href="#loop-index-split">-loop-index-split</a></td><td>Index Split Loops</td></tr>
146 <tr><td><a href="#loop-reduce">-loop-reduce</a></td><td>Loop Strength Reduction</td></tr>
147 <tr><td><a href="#loop-rotate">-loop-rotate</a></td><td>Rotate Loops</td></tr>
148 <tr><td><a href="#loop-unroll">-loop-unroll</a></td><td>Unroll loops</td></tr>
149 <tr><td><a href="#loop-unswitch">-loop-unswitch</a></td><td>Unswitch loops</td></tr>
150 <tr><td><a href="#loopsimplify">-loopsimplify</a></td><td>Canonicalize natural loops</td></tr>
151 <tr><td><a href="#lowerallocs">-lowerallocs</a></td><td>Lower allocations from instructions to calls</td></tr>
152 <tr><td><a href="#lowerinvoke">-lowerinvoke</a></td><td>Lower invoke and unwind, for unwindless code generators</td></tr>
153 <tr><td><a href="#lowersetjmp">-lowersetjmp</a></td><td>Lower Set Jump</td></tr>
154 <tr><td><a href="#lowerswitch">-lowerswitch</a></td><td>Lower SwitchInst's to branches</td></tr>
155 <tr><td><a href="#mem2reg">-mem2reg</a></td><td>Promote Memory to Register</td></tr>
156 <tr><td><a href="#memcpyopt">-memcpyopt</a></td><td>Optimize use of memcpy and friends</td></tr>
157 <tr><td><a href="#mergereturn">-mergereturn</a></td><td>Unify function exit nodes</td></tr>
158 <tr><td><a href="#prune-eh">-prune-eh</a></td><td>Remove unused exception handling info</td></tr>
159 <tr><td><a href="#reassociate">-reassociate</a></td><td>Reassociate expressions</td></tr>
160 <tr><td><a href="#reg2mem">-reg2mem</a></td><td>Demote all values to stack slots</td></tr>
161 <tr><td><a href="#scalarrepl">-scalarrepl</a></td><td>Scalar Replacement of Aggregates</td></tr>
162 <tr><td><a href="#sccp">-sccp</a></td><td>Sparse Conditional Constant Propagation</td></tr>
163 <tr><td><a href="#simplify-libcalls">-simplify-libcalls</a></td><td>Simplify well-known library calls</td></tr>
164 <tr><td><a href="#simplifycfg">-simplifycfg</a></td><td>Simplify the CFG</td></tr>
165 <tr><td><a href="#strip">-strip</a></td><td>Strip all symbols from a module</td></tr>
166 <tr><td><a href="#strip-dead-prototypes">-strip-dead-prototypes</a></td><td>Remove unused function declarations</td></tr>
167 <tr><td><a href="#sretpromotion">-sretpromotion</a></td><td>Promote sret arguments</td></tr>
168 <tr><td><a href="#tailcallelim">-tailcallelim</a></td><td>Tail Call Elimination</td></tr>
169 <tr><td><a href="#tailduplicate">-tailduplicate</a></td><td>Tail Duplication</td></tr>
172 <tr><th colspan="2"><b>UTILITY PASSES</b></th></tr>
173 <tr><th>Option</th><th>Name</th></tr>
174 <tr><td><a href="#deadarghaX0r">-deadarghaX0r</a></td><td>Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</td></tr>
175 <tr><td><a href="#extract-blocks">-extract-blocks</a></td><td>Extract Basic Blocks From Module (for bugpoint use)</td></tr>
176 <tr><td><a href="#preverify">-preverify</a></td><td>Preliminary module verification</td></tr>
177 <tr><td><a href="#verify">-verify</a></td><td>Module Verifier</td></tr>
178 <tr><td><a href="#view-cfg">-view-cfg</a></td><td>View CFG of function</td></tr>
179 <tr><td><a href="#view-cfg-only">-view-cfg-only</a></td><td>View CFG of function (with no function bodies)</td></tr>
183 <!-- ======================================================================= -->
184 <div class="doc_section"> <a name="example">Analysis Passes</a></div>
185 <div class="doc_text">
186 <p>This section describes the LLVM Analysis Passes.</p>
189 <!-------------------------------------------------------------------------- -->
190 <div class="doc_subsection">
191 <a name="aa-eval">Exhaustive Alias Analysis Precision Evaluator</a>
193 <div class="doc_text">
194 <p>This is a simple N^2 alias analysis accuracy evaluator.
195 Basically, for each function in the program, it simply queries to see how the
196 alias analysis implementation answers alias queries between each pair of
197 pointers in the function.</p>
199 <p>This is inspired and adapted from code by: Naveen Neelakantam, Francesco
200 Spadini, and Wojciech Stryjewski.</p>
203 <!-------------------------------------------------------------------------- -->
204 <div class="doc_subsection">
205 <a name="basicaa">Basic Alias Analysis (default AA impl)</a>
207 <div class="doc_text">
209 This is the default implementation of the Alias Analysis interface
210 that simply implements a few identities (two different globals cannot alias,
211 etc), but otherwise does no analysis.
215 <!-------------------------------------------------------------------------- -->
216 <div class="doc_subsection">
217 <a name="basiccg">Basic CallGraph Construction</a>
219 <div class="doc_text">
220 <p>Yet to be written.</p>
223 <!-------------------------------------------------------------------------- -->
224 <div class="doc_subsection">
225 <a name="codegenprepare">Optimize for code generation</a>
227 <div class="doc_text">
229 This pass munges the code in the input function to better prepare it for
230 SelectionDAG-based code generation. This works around limitations in it's
231 basic-block-at-a-time approach. It should eventually be removed.
235 <!-------------------------------------------------------------------------- -->
236 <div class="doc_subsection">
237 <a name="count-aa">Count Alias Analysis Query Responses</a>
239 <div class="doc_text">
241 A pass which can be used to count how many alias queries
242 are being made and how the alias analysis implementation being used responds.
246 <!-------------------------------------------------------------------------- -->
247 <div class="doc_subsection">
248 <a name="debug-aa">AA use debugger</a>
250 <div class="doc_text">
252 This simple pass checks alias analysis users to ensure that if they
253 create a new value, they do not query AA without informing it of the value.
254 It acts as a shim over any other AA pass you want.
258 Yes keeping track of every value in the program is expensive, but this is
263 <!-------------------------------------------------------------------------- -->
264 <div class="doc_subsection">
265 <a name="domfrontier">Dominance Frontier Construction</a>
267 <div class="doc_text">
269 This pass is a simple dominator construction algorithm for finding forward
274 <!-------------------------------------------------------------------------- -->
275 <div class="doc_subsection">
276 <a name="domtree">Dominator Tree Construction</a>
278 <div class="doc_text">
280 This pass is a simple dominator construction algorithm for finding forward
285 <!-------------------------------------------------------------------------- -->
286 <div class="doc_subsection">
287 <a name="dot-callgraph">Print Call Graph to 'dot' file</a>
289 <div class="doc_text">
291 This pass, only available in <code>opt</code>, prints the call graph into a
292 <code>.dot</code> graph. This graph can then be processed with the "dot" tool
293 to convert it to postscript or some other suitable format.
297 <!-------------------------------------------------------------------------- -->
298 <div class="doc_subsection">
299 <a name="dot-cfg">Print CFG of function to 'dot' file</a>
301 <div class="doc_text">
303 This pass, only available in <code>opt</code>, prints the control flow graph
304 into a <code>.dot</code> graph. This graph can then be processed with the
305 "dot" tool to convert it to postscript or some other suitable format.
309 <!-------------------------------------------------------------------------- -->
310 <div class="doc_subsection">
311 <a name="dot-cfg-only">Print CFG of function to 'dot' file (with no function bodies)</a>
313 <div class="doc_text">
315 This pass, only available in <code>opt</code>, prints the control flow graph
316 into a <code>.dot</code> graph, omitting the function bodies. This graph can
317 then be processed with the "dot" tool to convert it to postscript or some
318 other suitable format.
322 <!-------------------------------------------------------------------------- -->
323 <div class="doc_subsection">
324 <a name="globalsmodref-aa">Simple mod/ref analysis for globals</a>
326 <div class="doc_text">
328 This simple pass provides alias and mod/ref information for global values
329 that do not have their address taken, and keeps track of whether functions
330 read or write memory (are "pure"). For this simple (but very common) case,
331 we can provide pretty accurate and useful information.
335 <!-------------------------------------------------------------------------- -->
336 <div class="doc_subsection">
337 <a name="instcount">Counts the various types of Instructions</a>
339 <div class="doc_text">
341 This pass collects the count of all instructions and reports them
345 <!-------------------------------------------------------------------------- -->
346 <div class="doc_subsection">
347 <a name="intervals">Interval Partition Construction</a>
349 <div class="doc_text">
351 This analysis calculates and represents the interval partition of a function,
352 or a preexisting interval partition.
356 In this way, the interval partition may be used to reduce a flow graph down
357 to its degenerate single node interval partition (unless it is irreducible).
361 <!-------------------------------------------------------------------------- -->
362 <div class="doc_subsection">
363 <a name="loops">Natural Loop Construction</a>
365 <div class="doc_text">
367 This analysis is used to identify natural loops and determine the loop depth
368 of various nodes of the CFG. Note that the loops identified may actually be
369 several natural loops that share the same header node... not just a single
374 <!-------------------------------------------------------------------------- -->
375 <div class="doc_subsection">
376 <a name="memdep">Memory Dependence Analysis</a>
378 <div class="doc_text">
380 An analysis that determines, for a given memory operation, what preceding
381 memory operations it depends on. It builds on alias analysis information, and
382 tries to provide a lazy, caching interface to a common kind of alias
387 <!-------------------------------------------------------------------------- -->
388 <div class="doc_subsection">
389 <a name="no-aa">No Alias Analysis (always returns 'may' alias)</a>
391 <div class="doc_text">
393 Always returns "I don't know" for alias queries. NoAA is unlike other alias
394 analysis implementations, in that it does not chain to a previous analysis. As
395 such it doesn't follow many of the rules that other alias analyses must.
399 <!-------------------------------------------------------------------------- -->
400 <div class="doc_subsection">
401 <a name="no-profile">No Profile Information</a>
403 <div class="doc_text">
405 The default "no profile" implementation of the abstract
406 <code>ProfileInfo</code> interface.
410 <!-------------------------------------------------------------------------- -->
411 <div class="doc_subsection">
412 <a name="postdomfrontier">Post-Dominance Frontier Construction</a>
414 <div class="doc_text">
416 This pass is a simple post-dominator construction algorithm for finding
417 post-dominator frontiers.
421 <!-------------------------------------------------------------------------- -->
422 <div class="doc_subsection">
423 <a name="postdomtree">Post-Dominator Tree Construction</a>
425 <div class="doc_text">
427 This pass is a simple post-dominator construction algorithm for finding
432 <!-------------------------------------------------------------------------- -->
433 <div class="doc_subsection">
434 <a name="print-alias-sets">Alias Set Printer</a>
436 <div class="doc_text">
437 <p>Yet to be written.</p>
440 <!-------------------------------------------------------------------------- -->
441 <div class="doc_subsection">
442 <a name="print-callgraph">Print a call graph</a>
444 <div class="doc_text">
446 This pass, only available in <code>opt</code>, prints the call graph to
447 standard output in a human-readable form.
451 <!-------------------------------------------------------------------------- -->
452 <div class="doc_subsection">
453 <a name="print-callgraph-sccs">Print SCCs of the Call Graph</a>
455 <div class="doc_text">
457 This pass, only available in <code>opt</code>, prints the SCCs of the call
458 graph to standard output in a human-readable form.
462 <!-------------------------------------------------------------------------- -->
463 <div class="doc_subsection">
464 <a name="print-cfg-sccs">Print SCCs of each function CFG</a>
466 <div class="doc_text">
468 This pass, only available in <code>opt</code>, prints the SCCs of each
469 function CFG to standard output in a human-readable form.
473 <!-------------------------------------------------------------------------- -->
474 <div class="doc_subsection">
475 <a name="print-externalfnconstants">Print external fn callsites passed constants</a>
477 <div class="doc_text">
479 This pass, only available in <code>opt</code>, prints out call sites to
480 external functions that are called with constant arguments. This can be
481 useful when looking for standard library functions we should constant fold
482 or handle in alias analyses.
486 <!-------------------------------------------------------------------------- -->
487 <div class="doc_subsection">
488 <a name="print-function">Print function to stderr</a>
490 <div class="doc_text">
492 The <code>PrintFunctionPass</code> class is designed to be pipelined with
493 other <code>FunctionPass</code>es, and prints out the functions of the module
494 as they are processed.
498 <!-------------------------------------------------------------------------- -->
499 <div class="doc_subsection">
500 <a name="print-module">Print module to stderr</a>
502 <div class="doc_text">
504 This pass simply prints out the entire module when it is executed.
508 <!-------------------------------------------------------------------------- -->
509 <div class="doc_subsection">
510 <a name="print-used-types">Find Used Types</a>
512 <div class="doc_text">
514 This pass is used to seek out all of the types in use by the program. Note
515 that this analysis explicitly does not include types only used by the symbol
519 <!-------------------------------------------------------------------------- -->
520 <div class="doc_subsection">
521 <a name="profile-loader">Load profile information from llvmprof.out</a>
523 <div class="doc_text">
525 A concrete implementation of profiling information that loads the information
526 from a profile dump file.
530 <!-------------------------------------------------------------------------- -->
531 <div class="doc_subsection">
532 <a name="scalar-evolution">Scalar Evolution Analysis</a>
534 <div class="doc_text">
536 The <code>ScalarEvolution</code> analysis can be used to analyze and
537 catagorize scalar expressions in loops. It specializes in recognizing general
538 induction variables, representing them with the abstract and opaque
539 <code>SCEV</code> class. Given this analysis, trip counts of loops and other
540 important properties can be obtained.
544 This analysis is primarily useful for induction variable substitution and
549 <!-------------------------------------------------------------------------- -->
550 <div class="doc_subsection">
551 <a name="targetdata">Target Data Layout</a>
553 <div class="doc_text">
554 <p>Provides other passes access to information on how the size and alignment
555 required by the the target ABI for various data types.</p>
558 <!-- ======================================================================= -->
559 <div class="doc_section"> <a name="transform">Transform Passes</a></div>
560 <div class="doc_text">
561 <p>This section describes the LLVM Transform Passes.</p>
564 <!-------------------------------------------------------------------------- -->
565 <div class="doc_subsection">
566 <a name="adce">Aggressive Dead Code Elimination</a>
568 <div class="doc_text">
569 <p>ADCE aggressively tries to eliminate code. This pass is similar to
570 <a href="#dce">DCE</a> but it assumes that values are dead until proven
571 otherwise. This is similar to <a href="#sccp">SCCP</a>, except applied to
572 the liveness of values.</p>
575 <!-------------------------------------------------------------------------- -->
576 <div class="doc_subsection">
577 <a name="argpromotion">Promote 'by reference' arguments to scalars</a>
579 <div class="doc_text">
581 This pass promotes "by reference" arguments to be "by value" arguments. In
582 practice, this means looking for internal functions that have pointer
583 arguments. If it can prove, through the use of alias analysis, that an
584 argument is *only* loaded, then it can pass the value into the function
585 instead of the address of the value. This can cause recursive simplification
586 of code and lead to the elimination of allocas (especially in C++ template
591 This pass also handles aggregate arguments that are passed into a function,
592 scalarizing them if the elements of the aggregate are only loaded. Note that
593 it refuses to scalarize aggregates which would require passing in more than
594 three operands to the function, because passing thousands of operands for a
595 large array or structure is unprofitable!
599 Note that this transformation could also be done for arguments that are only
600 stored to (returning the value instead), but does not currently. This case
601 would be best handled when and if LLVM starts supporting multiple return
602 values from functions.
606 <!-------------------------------------------------------------------------- -->
607 <div class="doc_subsection">
608 <a name="block-placement">Profile Guided Basic Block Placement</a>
610 <div class="doc_text">
611 <p>This pass is a very simple profile guided basic block placement algorithm.
612 The idea is to put frequently executed blocks together at the start of the
613 function and hopefully increase the number of fall-through conditional
614 branches. If there is no profile information for a particular function, this
615 pass basically orders blocks in depth-first order.</p>
618 <!-------------------------------------------------------------------------- -->
619 <div class="doc_subsection">
620 <a name="break-crit-edges">Break critical edges in CFG</a>
622 <div class="doc_text">
624 Break all of the critical edges in the CFG by inserting a dummy basic block.
625 It may be "required" by passes that cannot deal with critical edges. This
626 transformation obviously invalidates the CFG, but can update forward dominator
627 (set, immediate dominators, tree, and frontier) information.
631 <!-------------------------------------------------------------------------- -->
632 <div class="doc_subsection">
633 <a name="codegenprepare">Prepare a function for code generation</a>
635 <div class="doc_text">
636 This pass munges the code in the input function to better prepare it for
637 SelectionDAG-based code generation. This works around limitations in it's
638 basic-block-at-a-time approach. It should eventually be removed.
641 <!-------------------------------------------------------------------------- -->
642 <div class="doc_subsection">
643 <a name="constmerge">Merge Duplicate Global Constants</a>
645 <div class="doc_text">
647 Merges duplicate global constants together into a single constant that is
648 shared. This is useful because some passes (ie TraceValues) insert a lot of
649 string constants into the program, regardless of whether or not an existing
654 <!-------------------------------------------------------------------------- -->
655 <div class="doc_subsection">
656 <a name="constprop">Simple constant propagation</a>
658 <div class="doc_text">
659 <p>This file implements constant propagation and merging. It looks for
660 instructions involving only constant operands and replaces them with a
661 constant value instead of an instruction. For example:</p>
662 <blockquote><pre>add i32 1, 2</pre></blockquote>
664 <blockquote><pre>i32 3</pre></blockquote>
665 <p>NOTE: this pass has a habit of making definitions be dead. It is a good
666 idea to to run a <a href="#die">DIE</a> (Dead Instruction Elimination) pass
667 sometime after running this pass.</p>
670 <!-------------------------------------------------------------------------- -->
671 <div class="doc_subsection">
672 <a name="dce">Dead Code Elimination</a>
674 <div class="doc_text">
676 Dead code elimination is similar to <a href="#die">dead instruction
677 elimination</a>, but it rechecks instructions that were used by removed
678 instructions to see if they are newly dead.
682 <!-------------------------------------------------------------------------- -->
683 <div class="doc_subsection">
684 <a name="deadargelim">Dead Argument Elimination</a>
686 <div class="doc_text">
688 This pass deletes dead arguments from internal functions. Dead argument
689 elimination removes arguments which are directly dead, as well as arguments
690 only passed into function calls as dead arguments of other functions. This
691 pass also deletes dead arguments in a similar way.
695 This pass is often useful as a cleanup pass to run after aggressive
696 interprocedural passes, which add possibly-dead arguments.
700 <!-------------------------------------------------------------------------- -->
701 <div class="doc_subsection">
702 <a name="deadtypeelim">Dead Type Elimination</a>
704 <div class="doc_text">
706 This pass is used to cleanup the output of GCC. It eliminate names for types
707 that are unused in the entire translation unit, using the <a
708 href="#findusedtypes">find used types</a> pass.
712 <!-------------------------------------------------------------------------- -->
713 <div class="doc_subsection">
714 <a name="die">Dead Instruction Elimination</a>
716 <div class="doc_text">
718 Dead instruction elimination performs a single pass over the function,
719 removing instructions that are obviously dead.
723 <!-------------------------------------------------------------------------- -->
724 <div class="doc_subsection">
725 <a name="dse">Dead Store Elimination</a>
727 <div class="doc_text">
729 A trivial dead store elimination that only considers basic-block local
734 <!-------------------------------------------------------------------------- -->
735 <div class="doc_subsection">
736 <a name="globaldce">Dead Global Elimination</a>
738 <div class="doc_text">
740 This transform is designed to eliminate unreachable internal globals from the
741 program. It uses an aggressive algorithm, searching out globals that are
742 known to be alive. After it finds all of the globals which are needed, it
743 deletes whatever is left over. This allows it to delete recursive chunks of
744 the program which are unreachable.
748 <!-------------------------------------------------------------------------- -->
749 <div class="doc_subsection">
750 <a name="globalopt">Global Variable Optimizer</a>
752 <div class="doc_text">
754 This pass transforms simple global variables that never have their address
755 taken. If obviously true, it marks read/write globals as constant, deletes
756 variables only stored to, etc.
760 <!-------------------------------------------------------------------------- -->
761 <div class="doc_subsection">
762 <a name="gvn">Global Value Numbering</a>
764 <div class="doc_text">
766 This pass performs global value numbering to eliminate fully and partially
767 redundant instructions. It also performs redundant load elimination.
772 <!-------------------------------------------------------------------------- -->
773 <div class="doc_subsection">
774 <a name="indmemrem">Indirect Malloc and Free Removal</a>
776 <div class="doc_text">
778 This pass finds places where memory allocation functions may escape into
779 indirect land. Some transforms are much easier (aka possible) only if free
780 or malloc are not called indirectly.
784 Thus find places where the address of memory functions are taken and construct
785 bounce functions with direct calls of those functions.
789 <!-------------------------------------------------------------------------- -->
790 <div class="doc_subsection">
791 <a name="indvars">Canonicalize Induction Variables</a>
793 <div class="doc_text">
795 This transformation analyzes and transforms the induction variables (and
796 computations derived from them) into simpler forms suitable for subsequent
797 analysis and transformation.
801 This transformation makes the following changes to each loop with an
802 identifiable induction variable:
806 <li>All loops are transformed to have a <em>single</em> canonical
807 induction variable which starts at zero and steps by one.</li>
808 <li>The canonical induction variable is guaranteed to be the first PHI node
809 in the loop header block.</li>
810 <li>Any pointer arithmetic recurrences are raised to use array
815 If the trip count of a loop is computable, this pass also makes the following
820 <li>The exit condition for the loop is canonicalized to compare the
821 induction value against the exit value. This turns loops like:
822 <blockquote><pre>for (i = 7; i*i < 1000; ++i)</pre></blockquote>
824 <blockquote><pre>for (i = 0; i != 25; ++i)</pre></blockquote></li>
825 <li>Any use outside of the loop of an expression derived from the indvar
826 is changed to compute the derived value outside of the loop, eliminating
827 the dependence on the exit value of the induction variable. If the only
828 purpose of the loop is to compute the exit value of some derived
829 expression, this transformation will make the loop dead.</li>
833 This transformation should be followed by strength reduction after all of the
834 desired loop transformations have been performed. Additionally, on targets
835 where it is profitable, the loop could be transformed to count down to zero
836 (the "do loop" optimization).
840 <!-------------------------------------------------------------------------- -->
841 <div class="doc_subsection">
842 <a name="inline">Function Integration/Inlining</a>
844 <div class="doc_text">
846 Bottom-up inlining of functions into callees.
850 <!-------------------------------------------------------------------------- -->
851 <div class="doc_subsection">
852 <a name="insert-block-profiling">Insert instrumentation for block profiling</a>
854 <div class="doc_text">
856 This pass instruments the specified program with counters for basic block
857 profiling, which counts the number of times each basic block executes. This
858 is the most basic form of profiling, which can tell which blocks are hot, but
859 cannot reliably detect hot paths through the CFG.
863 Note that this implementation is very naïve. Control equivalent regions of
864 the CFG should not require duplicate counters, but it does put duplicate
869 <!-------------------------------------------------------------------------- -->
870 <div class="doc_subsection">
871 <a name="insert-edge-profiling">Insert instrumentation for edge profiling</a>
873 <div class="doc_text">
875 This pass instruments the specified program with counters for edge profiling.
876 Edge profiling can give a reasonable approximation of the hot paths through a
877 program, and is used for a wide variety of program transformations.
881 Note that this implementation is very naïve. It inserts a counter for
882 <em>every</em> edge in the program, instead of using control flow information
883 to prune the number of counters inserted.
887 <!-------------------------------------------------------------------------- -->
888 <div class="doc_subsection">
889 <a name="insert-function-profiling">Insert instrumentation for function profiling</a>
891 <div class="doc_text">
893 This pass instruments the specified program with counters for function
894 profiling, which counts the number of times each function is called.
898 <!-------------------------------------------------------------------------- -->
899 <div class="doc_subsection">
900 <a name="insert-null-profiling-rs">Measure profiling framework overhead</a>
902 <div class="doc_text">
904 The basic profiler that does nothing. It is the default profiler and thus
905 terminates <code>RSProfiler</code> chains. It is useful for measuring
910 <!-------------------------------------------------------------------------- -->
911 <div class="doc_subsection">
912 <a name="insert-rs-profiling-framework">Insert random sampling instrumentation framework</a>
914 <div class="doc_text">
916 The second stage of the random-sampling instrumentation framework, duplicates
917 all instructions in a function, ignoring the profiling code, then connects the
918 two versions together at the entry and at backedges. At each connection point
919 a choice is made as to whether to jump to the profiled code (take a sample) or
920 execute the unprofiled code.
924 After this pass, it is highly recommended to run<a href="#mem2reg">mem2reg</a>
925 and <a href="#adce">adce</a>. <a href="#instcombine">instcombine</a>,
926 <a href="#load-vn">load-vn</a>, <a href="#gdce">gdce</a>, and
927 <a href="#dse">dse</a> also are good to run afterwards.
931 <!-------------------------------------------------------------------------- -->
932 <div class="doc_subsection">
933 <a name="instcombine">Combine redundant instructions</a>
935 <div class="doc_text">
937 Combine instructions to form fewer, simple
938 instructions. This pass does not modify the CFG This pass is where algebraic
939 simplification happens.
943 This pass combines things like:
948 %Z = add i32 %Y, 1</pre></blockquote>
955 >%Z = add i32 %X, 2</pre></blockquote>
958 This is a simple worklist driven algorithm.
962 This pass guarantees that the following canonicalizations are performed on
967 <li>If a binary operator has a constant operand, it is moved to the right-
969 <li>Bitwise operators with constant operands are always grouped so that
970 shifts are performed first, then <code>or</code>s, then
971 <code>and</code>s, then <code>xor</code>s.</li>
972 <li>Compare instructions are converted from <code><</code>,
973 <code>></code>, <code>≤</code>, or <code>≥</code> to
974 <code>=</code> or <code>≠</code> if possible.</li>
975 <li>All <code>cmp</code> instructions on boolean values are replaced with
976 logical operations.</li>
977 <li><code>add <var>X</var>, <var>X</var></code> is represented as
978 <code>mul <var>X</var>, 2</code> ⇒ <code>shl <var>X</var>, 1</code></li>
979 <li>Multiplies with a constant power-of-two argument are transformed into
985 <!-------------------------------------------------------------------------- -->
986 <div class="doc_subsection">
987 <a name="internalize">Internalize Global Symbols</a>
989 <div class="doc_text">
991 This pass loops over all of the functions in the input module, looking for a
992 main function. If a main function is found, all other functions and all
993 global variables with initializers are marked as internal.
997 <!-------------------------------------------------------------------------- -->
998 <div class="doc_subsection">
999 <a name="ipconstprop">Interprocedural constant propagation</a>
1001 <div class="doc_text">
1003 This pass implements an <em>extremely</em> simple interprocedural constant
1004 propagation pass. It could certainly be improved in many different ways,
1005 like using a worklist. This pass makes arguments dead, but does not remove
1006 them. The existing dead argument elimination pass should be run after this
1007 to clean up the mess.
1011 <!-------------------------------------------------------------------------- -->
1012 <div class="doc_subsection">
1013 <a name="ipsccp">Interprocedural Sparse Conditional Constant Propagation</a>
1015 <div class="doc_text">
1017 An interprocedural variant of <a href="#sccp">Sparse Conditional Constant
1022 <!-------------------------------------------------------------------------- -->
1023 <div class="doc_subsection">
1024 <a name="jump-threading">Thread control through conditional blocks</a>
1026 <div class="doc_text">
1028 Jump threading tries to find distinct threads of control flow running through
1029 a basic block. This pass looks at blocks that have multiple predecessors and
1030 multiple successors. If one or more of the predecessors of the block can be
1031 proven to always cause a jump to one of the successors, we forward the edge
1032 from the predecessor to the successor by duplicating the contents of this
1036 An example of when this can occur is code like this:
1043 if (X < 3) {</pre>
1046 In this case, the unconditional branch at the end of the first if can be
1047 revectored to the false side of the second if.
1051 <!-------------------------------------------------------------------------- -->
1052 <div class="doc_subsection">
1053 <a name="lcssa">Loop-Closed SSA Form Pass</a>
1055 <div class="doc_text">
1057 This pass transforms loops by placing phi nodes at the end of the loops for
1058 all values that are live across the loop boundary. For example, it turns
1059 the left into the right code:
1063 >for (...) for (...)
1068 X3 = phi(X1, X2) X3 = phi(X1, X2)
1069 ... = X3 + 4 X4 = phi(X3)
1073 This is still valid LLVM; the extra phi nodes are purely redundant, and will
1074 be trivially eliminated by <code>InstCombine</code>. The major benefit of
1075 this transformation is that it makes many other loop optimizations, such as
1076 LoopUnswitching, simpler.
1080 <!-------------------------------------------------------------------------- -->
1081 <div class="doc_subsection">
1082 <a name="licm">Loop Invariant Code Motion</a>
1084 <div class="doc_text">
1086 This pass performs loop invariant code motion, attempting to remove as much
1087 code from the body of a loop as possible. It does this by either hoisting
1088 code into the preheader block, or by sinking code to the exit blocks if it is
1089 safe. This pass also promotes must-aliased memory locations in the loop to
1090 live in registers, thus hoisting and sinking "invariant" loads and stores.
1094 This pass uses alias analysis for two purposes:
1098 <li>Moving loop invariant loads and calls out of loops. If we can determine
1099 that a load or call inside of a loop never aliases anything stored to,
1100 we can hoist it or sink it like any other instruction.</li>
1101 <li>Scalar Promotion of Memory - If there is a store instruction inside of
1102 the loop, we try to move the store to happen AFTER the loop instead of
1103 inside of the loop. This can only happen if a few conditions are true:
1105 <li>The pointer stored through is loop invariant.</li>
1106 <li>There are no stores or loads in the loop which <em>may</em> alias
1107 the pointer. There are no calls in the loop which mod/ref the
1110 If these conditions are true, we can promote the loads and stores in the
1111 loop of the pointer to use a temporary alloca'd variable. We then use
1112 the mem2reg functionality to construct the appropriate SSA form for the
1116 <!-------------------------------------------------------------------------- -->
1117 <div class="doc_subsection">
1118 <a name="loop-deletion">Dead Loop Deletion Pass</a>
1120 <div class="doc_text">
1122 This file implements the Dead Loop Deletion Pass. This pass is responsible
1123 for eliminating loops with non-infinite computable trip counts that have no
1124 side effects or volatile instructions, and do not contribute to the
1125 computation of the function's return value.
1129 <!-------------------------------------------------------------------------- -->
1130 <div class="doc_subsection">
1131 <a name="loop-extract">Extract loops into new functions</a>
1133 <div class="doc_text">
1135 A pass wrapper around the <code>ExtractLoop()</code> scalar transformation to
1136 extract each top-level loop into its own new function. If the loop is the
1137 <em>only</em> loop in a given function, it is not touched. This is a pass most
1138 useful for debugging via bugpoint.
1142 <!-------------------------------------------------------------------------- -->
1143 <div class="doc_subsection">
1144 <a name="loop-extract-single">Extract at most one loop into a new function</a>
1146 <div class="doc_text">
1148 Similar to <a href="#loop-extract">Extract loops into new functions</a>,
1149 this pass extracts one natural loop from the program into a function if it
1150 can. This is used by bugpoint.
1154 <!-------------------------------------------------------------------------- -->
1155 <div class="doc_subsection">
1156 <a name="loop-index-split">Index Split Loops</a>
1158 <div class="doc_text">
1160 This pass divides loop's iteration range by spliting loop such that each
1161 individual loop is executed efficiently.
1165 <!-------------------------------------------------------------------------- -->
1166 <div class="doc_subsection">
1167 <a name="loop-reduce">Loop Strength Reduction</a>
1169 <div class="doc_text">
1171 This pass performs a strength reduction on array references inside loops that
1172 have as one or more of their components the loop induction variable. This is
1173 accomplished by creating a new value to hold the initial value of the array
1174 access for the first iteration, and then creating a new GEP instruction in
1175 the loop to increment the value by the appropriate amount.
1179 <!-------------------------------------------------------------------------- -->
1180 <div class="doc_subsection">
1181 <a name="loop-rotate">Rotate Loops</a>
1183 <div class="doc_text">
1184 <p>A simple loop rotation transformation.</p>
1187 <!-------------------------------------------------------------------------- -->
1188 <div class="doc_subsection">
1189 <a name="loop-unroll">Unroll loops</a>
1191 <div class="doc_text">
1193 This pass implements a simple loop unroller. It works best when loops have
1194 been canonicalized by the <a href="#indvars"><tt>-indvars</tt></a> pass,
1195 allowing it to determine the trip counts of loops easily.
1199 <!-------------------------------------------------------------------------- -->
1200 <div class="doc_subsection">
1201 <a name="loop-unswitch">Unswitch loops</a>
1203 <div class="doc_text">
1205 This pass transforms loops that contain branches on loop-invariant conditions
1206 to have multiple loops. For example, it turns the left into the right code:
1218 This can increase the size of the code exponentially (doubling it every time
1219 a loop is unswitched) so we only unswitch if the resultant code will be
1220 smaller than a threshold.
1224 This pass expects LICM to be run before it to hoist invariant conditions out
1225 of the loop, to make the unswitching opportunity obvious.
1229 <!-------------------------------------------------------------------------- -->
1230 <div class="doc_subsection">
1231 <a name="loopsimplify">Canonicalize natural loops</a>
1233 <div class="doc_text">
1235 This pass performs several transformations to transform natural loops into a
1236 simpler form, which makes subsequent analyses and transformations simpler and
1241 Loop pre-header insertion guarantees that there is a single, non-critical
1242 entry edge from outside of the loop to the loop header. This simplifies a
1243 number of analyses and transformations, such as LICM.
1247 Loop exit-block insertion guarantees that all exit blocks from the loop
1248 (blocks which are outside of the loop that have predecessors inside of the
1249 loop) only have predecessors from inside of the loop (and are thus dominated
1250 by the loop header). This simplifies transformations such as store-sinking
1251 that are built into LICM.
1255 This pass also guarantees that loops will have exactly one backedge.
1259 Note that the simplifycfg pass will clean up blocks which are split out but
1260 end up being unnecessary, so usage of this pass should not pessimize
1265 This pass obviously modifies the CFG, but updates loop information and
1266 dominator information.
1270 <!-------------------------------------------------------------------------- -->
1271 <div class="doc_subsection">
1272 <a name="lowerallocs">Lower allocations from instructions to calls</a>
1274 <div class="doc_text">
1276 Turn <tt>malloc</tt> and <tt>free</tt> instructions into <tt>@malloc</tt> and
1277 <tt>@free</tt> calls.
1281 This is a target-dependent tranformation because it depends on the size of
1282 data types and alignment constraints.
1286 <!-------------------------------------------------------------------------- -->
1287 <div class="doc_subsection">
1288 <a name="lowerinvoke">Lower invoke and unwind, for unwindless code generators</a>
1290 <div class="doc_text">
1292 This transformation is designed for use by code generators which do not yet
1293 support stack unwinding. This pass supports two models of exception handling
1294 lowering, the 'cheap' support and the 'expensive' support.
1298 'Cheap' exception handling support gives the program the ability to execute
1299 any program which does not "throw an exception", by turning 'invoke'
1300 instructions into calls and by turning 'unwind' instructions into calls to
1301 abort(). If the program does dynamically use the unwind instruction, the
1302 program will print a message then abort.
1306 'Expensive' exception handling support gives the full exception handling
1307 support to the program at the cost of making the 'invoke' instruction
1308 really expensive. It basically inserts setjmp/longjmp calls to emulate the
1309 exception handling as necessary.
1313 Because the 'expensive' support slows down programs a lot, and EH is only
1314 used for a subset of the programs, it must be specifically enabled by the
1315 <tt>-enable-correct-eh-support</tt> option.
1319 Note that after this pass runs the CFG is not entirely accurate (exceptional
1320 control flow edges are not correct anymore) so only very simple things should
1321 be done after the lowerinvoke pass has run (like generation of native code).
1322 This should not be used as a general purpose "my LLVM-to-LLVM pass doesn't
1323 support the invoke instruction yet" lowering pass.
1327 <!-------------------------------------------------------------------------- -->
1328 <div class="doc_subsection">
1329 <a name="lowersetjmp">Lower Set Jump</a>
1331 <div class="doc_text">
1333 Lowers <tt>setjmp</tt> and <tt>longjmp</tt> to use the LLVM invoke and unwind
1334 instructions as necessary.
1338 Lowering of <tt>longjmp</tt> is fairly trivial. We replace the call with a
1339 call to the LLVM library function <tt>__llvm_sjljeh_throw_longjmp()</tt>.
1340 This unwinds the stack for us calling all of the destructors for
1341 objects allocated on the stack.
1345 At a <tt>setjmp</tt> call, the basic block is split and the <tt>setjmp</tt>
1346 removed. The calls in a function that have a <tt>setjmp</tt> are converted to
1347 invoke where the except part checks to see if it's a <tt>longjmp</tt>
1348 exception and, if so, if it's handled in the function. If it is, then it gets
1349 the value returned by the <tt>longjmp</tt> and goes to where the basic block
1350 was split. <tt>invoke</tt> instructions are handled in a similar fashion with
1351 the original except block being executed if it isn't a <tt>longjmp</tt>
1352 except that is handled by that function.
1356 <!-------------------------------------------------------------------------- -->
1357 <div class="doc_subsection">
1358 <a name="lowerswitch">Lower SwitchInst's to branches</a>
1360 <div class="doc_text">
1362 Rewrites <tt>switch</tt> instructions with a sequence of branches, which
1363 allows targets to get away with not implementing the switch instruction until
1368 <!-------------------------------------------------------------------------- -->
1369 <div class="doc_subsection">
1370 <a name="mem2reg">Promote Memory to Register</a>
1372 <div class="doc_text">
1374 This file promotes memory references to be register references. It promotes
1375 <tt>alloca</tt> instructions which only have <tt>load</tt>s and
1376 <tt>store</tt>s as uses. An <tt>alloca</tt> is transformed by using dominator
1377 frontiers to place <tt>phi</tt> nodes, then traversing the function in
1378 depth-first order to rewrite <tt>load</tt>s and <tt>store</tt>s as
1379 appropriate. This is just the standard SSA construction algorithm to construct
1384 <!-------------------------------------------------------------------------- -->
1385 <div class="doc_subsection">
1386 <a name="memcpyopt">Optimize use of memcpy and friend</a>
1388 <div class="doc_text">
1390 This pass performs various transformations related to eliminating memcpy
1391 calls, or transforming sets of stores into memset's.
1395 <!-------------------------------------------------------------------------- -->
1396 <div class="doc_subsection">
1397 <a name="mergereturn">Unify function exit nodes</a>
1399 <div class="doc_text">
1401 Ensure that functions have at most one <tt>ret</tt> instruction in them.
1402 Additionally, it keeps track of which node is the new exit node of the CFG.
1406 <!-------------------------------------------------------------------------- -->
1407 <div class="doc_subsection">
1408 <a name="prune-eh">Remove unused exception handling info</a>
1410 <div class="doc_text">
1412 This file implements a simple interprocedural pass which walks the call-graph,
1413 turning <tt>invoke</tt> instructions into <tt>call</tt> instructions if and
1414 only if the callee cannot throw an exception. It implements this as a
1415 bottom-up traversal of the call-graph.
1419 <!-------------------------------------------------------------------------- -->
1420 <div class="doc_subsection">
1421 <a name="reassociate">Reassociate expressions</a>
1423 <div class="doc_text">
1425 This pass reassociates commutative expressions in an order that is designed
1426 to promote better constant propagation, GCSE, LICM, PRE, etc.
1430 For example: 4 + (<var>x</var> + 5) ⇒ <var>x</var> + (4 + 5)
1434 In the implementation of this algorithm, constants are assigned rank = 0,
1435 function arguments are rank = 1, and other values are assigned ranks
1436 corresponding to the reverse post order traversal of current function
1437 (starting at 2), which effectively gives values in deep loops higher rank
1438 than values not in loops.
1442 <!-------------------------------------------------------------------------- -->
1443 <div class="doc_subsection">
1444 <a name="reg2mem">Demote all values to stack slots</a>
1446 <div class="doc_text">
1448 This file demotes all registers to memory references. It is intented to be
1449 the inverse of <a href="#mem2reg"><tt>-mem2reg</tt></a>. By converting to
1450 <tt>load</tt> instructions, the only values live across basic blocks are
1451 <tt>alloca</tt> instructions and <tt>load</tt> instructions before
1452 <tt>phi</tt> nodes. It is intended that this should make CFG hacking much
1453 easier. To make later hacking easier, the entry block is split into two, such
1454 that all introduced <tt>alloca</tt> instructions (and nothing else) are in the
1459 <!-------------------------------------------------------------------------- -->
1460 <div class="doc_subsection">
1461 <a name="scalarrepl">Scalar Replacement of Aggregates</a>
1463 <div class="doc_text">
1465 The well-known scalar replacement of aggregates transformation. This
1466 transform breaks up <tt>alloca</tt> instructions of aggregate type (structure
1467 or array) into individual <tt>alloca</tt> instructions for each member if
1468 possible. Then, if possible, it transforms the individual <tt>alloca</tt>
1469 instructions into nice clean scalar SSA form.
1473 This combines a simple scalar replacement of aggregates algorithm with the <a
1474 href="#mem2reg"><tt>mem2reg</tt></a> algorithm because often interact,
1475 especially for C++ programs. As such, iterating between <tt>scalarrepl</tt>,
1476 then <a href="#mem2reg"><tt>mem2reg</tt></a> until we run out of things to
1481 <!-------------------------------------------------------------------------- -->
1482 <div class="doc_subsection">
1483 <a name="sccp">Sparse Conditional Constant Propagation</a>
1485 <div class="doc_text">
1487 Sparse conditional constant propagation and merging, which can be summarized
1492 <li>Assumes values are constant unless proven otherwise</li>
1493 <li>Assumes BasicBlocks are dead unless proven otherwise</li>
1494 <li>Proves values to be constant, and replaces them with constants</li>
1495 <li>Proves conditional branches to be unconditional</li>
1499 Note that this pass has a habit of making definitions be dead. It is a good
1500 idea to to run a DCE pass sometime after running this pass.
1504 <!-------------------------------------------------------------------------- -->
1505 <div class="doc_subsection">
1506 <a name="simplify-libcalls">Simplify well-known library calls</a>
1508 <div class="doc_text">
1510 Applies a variety of small optimizations for calls to specific well-known
1511 function calls (e.g. runtime library functions). For example, a call
1512 <tt>exit(3)</tt> that occurs within the <tt>main()</tt> function can be
1513 transformed into simply <tt>return 3</tt>.
1517 <!-------------------------------------------------------------------------- -->
1518 <div class="doc_subsection">
1519 <a name="simplifycfg">Simplify the CFG</a>
1521 <div class="doc_text">
1523 Performs dead code elimination and basic block merging. Specifically:
1527 <li>Removes basic blocks with no predecessors.</li>
1528 <li>Merges a basic block into its predecessor if there is only one and the
1529 predecessor only has one successor.</li>
1530 <li>Eliminates PHI nodes for basic blocks with a single predecessor.</li>
1531 <li>Eliminates a basic block that only contains an unconditional
1536 <!-------------------------------------------------------------------------- -->
1537 <div class="doc_subsection">
1538 <a name="strip">Strip all symbols from a module</a>
1540 <div class="doc_text">
1542 Performs code stripping. This transformation can delete:
1546 <li>names for virtual registers</li>
1547 <li>symbols for internal globals and functions</li>
1548 <li>debug information</li>
1552 Note that this transformation makes code much less readable, so it should
1553 only be used in situations where the <tt>strip</tt> utility would be used,
1554 such as reducing code size or making it harder to reverse engineer code.
1558 <!-------------------------------------------------------------------------- -->
1559 <div class="doc_subsection">
1560 <a name="strip-dead-prototypes">Remove unused function declarations</a>
1562 <div class="doc_text">
1564 This pass loops over all of the functions in the input module, looking for
1565 dead declarations and removes them. Dead declarations are declarations of
1566 functions for which no implementation is available (i.e., declarations for
1567 unused library functions).
1571 <!-------------------------------------------------------------------------- -->
1572 <div class="doc_subsection">
1573 <a name="sretpromotion">Promote sret arguments</a>
1575 <div class="doc_text">
1577 This pass finds functions that return a struct (using a pointer to the struct
1578 as the first argument of the function, marked with the '<tt>sret</tt>' attribute) and
1579 replaces them with a new function that simply returns each of the elements of
1580 that struct (using multiple return values).
1584 This pass works under a number of conditions:
1588 <li>The returned struct must not contain other structs</li>
1589 <li>The returned struct must only be used to load values from</li>
1590 <li>The placeholder struct passed in is the result of an <tt>alloca</tt></li>
1594 <!-------------------------------------------------------------------------- -->
1595 <div class="doc_subsection">
1596 <a name="tailcallelim">Tail Call Elimination</a>
1598 <div class="doc_text">
1600 This file transforms calls of the current function (self recursion) followed
1601 by a return instruction with a branch to the entry of the function, creating
1602 a loop. This pass also implements the following extensions to the basic
1607 <li>Trivial instructions between the call and return do not prevent the
1608 transformation from taking place, though currently the analysis cannot
1609 support moving any really useful instructions (only dead ones).
1610 <li>This pass transforms functions that are prevented from being tail
1611 recursive by an associative expression to use an accumulator variable,
1612 thus compiling the typical naive factorial or <tt>fib</tt> implementation
1613 into efficient code.
1614 <li>TRE is performed if the function returns void, if the return
1615 returns the result returned by the call, or if the function returns a
1616 run-time constant on all exits from the function. It is possible, though
1617 unlikely, that the return returns something else (like constant 0), and
1618 can still be TRE'd. It can be TRE'd if <em>all other</em> return
1619 instructions in the function return the exact same value.
1620 <li>If it can prove that callees do not access theier caller stack frame,
1621 they are marked as eligible for tail call elimination (by the code
1626 <!-------------------------------------------------------------------------- -->
1627 <div class="doc_subsection">
1628 <a name="tailduplicate">Tail Duplication</a>
1630 <div class="doc_text">
1632 This pass performs a limited form of tail duplication, intended to simplify
1633 CFGs by removing some unconditional branches. This pass is necessary to
1634 straighten out loops created by the C front-end, but also is capable of
1635 making other code nicer. After this pass is run, the CFG simplify pass
1636 should be run to clean up the mess.
1640 <!-- ======================================================================= -->
1641 <div class="doc_section"> <a name="transform">Utility Passes</a></div>
1642 <div class="doc_text">
1643 <p>This section describes the LLVM Utility Passes.</p>
1646 <!-------------------------------------------------------------------------- -->
1647 <div class="doc_subsection">
1648 <a name="deadarghaX0r">Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</a>
1650 <div class="doc_text">
1652 Same as dead argument elimination, but deletes arguments to functions which
1653 are external. This is only for use by <a
1654 href="Bugpoint.html">bugpoint</a>.</p>
1657 <!-------------------------------------------------------------------------- -->
1658 <div class="doc_subsection">
1659 <a name="extract-blocks">Extract Basic Blocks From Module (for bugpoint use)</a>
1661 <div class="doc_text">
1663 This pass is used by bugpoint to extract all blocks from the module into their
1667 <!-------------------------------------------------------------------------- -->
1668 <div class="doc_subsection">
1669 <a name="preverify">Preliminary module verification</a>
1671 <div class="doc_text">
1673 Ensures that the module is in the form required by the <a
1674 href="#verifier">Module Verifier</a> pass.
1678 Running the verifier runs this pass automatically, so there should be no need
1683 <!-------------------------------------------------------------------------- -->
1684 <div class="doc_subsection">
1685 <a name="verify">Module Verifier</a>
1687 <div class="doc_text">
1689 Verifies an LLVM IR code. This is useful to run after an optimization which is
1690 undergoing testing. Note that <tt>llvm-as</tt> verifies its input before
1691 emitting bitcode, and also that malformed bitcode is likely to make LLVM
1692 crash. All language front-ends are therefore encouraged to verify their output
1693 before performing optimizing transformations.
1697 <li>Both of a binary operator's parameters are of the same type.</li>
1698 <li>Verify that the indices of mem access instructions match other
1700 <li>Verify that arithmetic and other things are only performed on
1701 first-class types. Verify that shifts and logicals only happen on
1703 <li>All of the constants in a switch statement are of the correct type.</li>
1704 <li>The code is in valid SSA form.</li>
1705 <li>It is illegal to put a label into any other type (like a structure) or
1707 <li>Only phi nodes can be self referential: <tt>%x = add i32 %x, %x</tt> is
1709 <li>PHI nodes must have an entry for each predecessor, with no extras.</li>
1710 <li>PHI nodes must be the first thing in a basic block, all grouped
1712 <li>PHI nodes must have at least one entry.</li>
1713 <li>All basic blocks should only end with terminator insts, not contain
1715 <li>The entry node to a function must not have predecessors.</li>
1716 <li>All Instructions must be embedded into a basic block.</li>
1717 <li>Functions cannot take a void-typed parameter.</li>
1718 <li>Verify that a function's argument list agrees with its declared
1720 <li>It is illegal to specify a name for a void value.</li>
1721 <li>It is illegal to have a internal global value with no initializer.</li>
1722 <li>It is illegal to have a ret instruction that returns a value that does
1723 not agree with the function return value type.</li>
1724 <li>Function call argument types match the function prototype.</li>
1725 <li>All other things that are tested by asserts spread about the code.</li>
1729 Note that this does not provide full security verification (like Java), but
1730 instead just tries to ensure that code is well-formed.
1734 <!-------------------------------------------------------------------------- -->
1735 <div class="doc_subsection">
1736 <a name="view-cfg">View CFG of function</a>
1738 <div class="doc_text">
1740 Displays the control flow graph using the GraphViz tool.
1744 <!-------------------------------------------------------------------------- -->
1745 <div class="doc_subsection">
1746 <a name="view-cfg-only">View CFG of function (with no function bodies)</a>
1748 <div class="doc_text">
1750 Displays the control flow graph using the GraphViz tool, but omitting function
1755 <!-- *********************************************************************** -->
1759 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1760 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1761 <a href="http://validator.w3.org/check/referer"><img
1762 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1764 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a><br>
1765 <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
1766 Last modified: $Date$