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="#anders-aa">-anders-aa</a></td><td>Andersen's Interprocedural Alias Analysis</td></tr>
79 <tr><td><a href="#basicaa">-basicaa</a></td><td>Basic Alias Analysis (default AA impl)</td></tr>
80 <tr><td><a href="#basiccg">-basiccg</a></td><td>Basic CallGraph Construction</td></tr>
81 <tr><td><a href="#basicvn">-basicvn</a></td><td>Basic Value Numbering (default GVN impl)</td></tr>
82 <tr><td><a href="#callgraph">-callgraph</a></td><td>Print a call graph</td></tr>
83 <tr><td><a href="#callscc">-callscc</a></td><td>Print SCCs of the Call Graph</td></tr>
84 <tr><td><a href="#cfgscc">-cfgscc</a></td><td>Print SCCs of each function CFG</td></tr>
85 <tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Optimize for code generation</td></tr>
86 <tr><td><a href="#count-aa">-count-aa</a></td><td>Count Alias Analysis Query Responses</td></tr>
87 <tr><td><a href="#debug-aa">-debug-aa</a></td><td>AA use debugger</td></tr>
88 <tr><td><a href="#domfrontier">-domfrontier</a></td><td>Dominance Frontier Construction</td></tr>
89 <tr><td><a href="#domtree">-domtree</a></td><td>Dominator Tree Construction</td></tr>
90 <tr><td><a href="#externalfnconstants">-externalfnconstants</a></td><td>Print external fn callsites passed constants</td></tr>
91 <tr><td><a href="#globalsmodref-aa">-globalsmodref-aa</a></td><td>Simple mod/ref analysis for globals</td></tr>
92 <tr><td><a href="#instcount">-instcount</a></td><td>Counts the various types of Instructions</td></tr>
93 <tr><td><a href="#intervals">-intervals</a></td><td>Interval Partition Construction</td></tr>
94 <tr><td><a href="#load-vn">-load-vn</a></td><td>Load Value Numbering</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">-print</a></td><td>Print function to stderr</td></tr>
102 <tr><td><a href="#print-alias-sets">-print-alias-sets</a></td><td>Alias Set Printer</td></tr>
103 <tr><td><a href="#print-callgraph">-print-callgraph</a></td><td>Print Call Graph to 'dot' file</td></tr>
104 <tr><td><a href="#print-cfg">-print-cfg</a></td><td>Print CFG of function to 'dot' file</td></tr>
105 <tr><td><a href="#print-cfg-only">-print-cfg-only</a></td><td>Print CFG of function to 'dot' file (with no function bodies)</td></tr>
106 <tr><td><a href="#printm">-printm</a></td><td>Print module to stderr</td></tr>
107 <tr><td><a href="#printusedtypes">-printusedtypes</a></td><td>Find Used Types</td></tr>
108 <tr><td><a href="#profile-loader">-profile-loader</a></td><td>Load profile information from llvmprof.out</td></tr>
109 <tr><td><a href="#scalar-evolution">-scalar-evolution</a></td><td>Scalar Evolution Analysis</td></tr>
110 <tr><td><a href="#targetdata">-targetdata</a></td><td>Target Data Layout</td></tr>
113 <tr><th colspan="2"><b>TRANSFORM PASSES</b></th></tr>
114 <tr><th>Option</th><th>Name</th></tr>
115 <tr><td><a href="#adce">-adce</a></td><td>Aggressive Dead Code Elimination</td></tr>
116 <tr><td><a href="#argpromotion">-argpromotion</a></td><td>Promote 'by reference' arguments to scalars</td></tr>
117 <tr><td><a href="#block-placement">-block-placement</a></td><td>Profile Guided Basic Block Placement</td></tr>
118 <tr><td><a href="#break-crit-edges">-break-crit-edges</a></td><td>Break critical edges in CFG</td></tr>
119 <tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Prepare a function for code generation </td></tr>
120 <tr><td><a href="#condprop">-condprop</a></td><td>Conditional Propagation</td></tr>
121 <tr><td><a href="#constmerge">-constmerge</a></td><td>Merge Duplicate Global Constants</td></tr>
122 <tr><td><a href="#constprop">-constprop</a></td><td>Simple constant propagation</td></tr>
123 <tr><td><a href="#dce">-dce</a></td><td>Dead Code Elimination</td></tr>
124 <tr><td><a href="#deadargelim">-deadargelim</a></td><td>Dead Argument Elimination</td></tr>
125 <tr><td><a href="#deadtypeelim">-deadtypeelim</a></td><td>Dead Type Elimination</td></tr>
126 <tr><td><a href="#die">-die</a></td><td>Dead Instruction Elimination</td></tr>
127 <tr><td><a href="#dse">-dse</a></td><td>Dead Store Elimination</td></tr>
128 <tr><td><a href="#gcse">-gcse</a></td><td>Global Common Subexpression Elimination</td></tr>
129 <tr><td><a href="#globaldce">-globaldce</a></td><td>Dead Global Elimination</td></tr>
130 <tr><td><a href="#globalopt">-globalopt</a></td><td>Global Variable Optimizer</td></tr>
131 <tr><td><a href="#gvn">-gvn</a></td><td>Global Value Numbering</td></tr>
132 <tr><td><a href="#gvnpre">-gvnpre</a></td><td>Global Value Numbering/Partial Redundancy Elimination</td></tr>
133 <tr><td><a href="#indmemrem">-indmemrem</a></td><td>Indirect Malloc and Free Removal</td></tr>
134 <tr><td><a href="#indvars">-indvars</a></td><td>Canonicalize Induction Variables</td></tr>
135 <tr><td><a href="#inline">-inline</a></td><td>Function Integration/Inlining</td></tr>
136 <tr><td><a href="#insert-block-profiling">-insert-block-profiling</a></td><td>Insert instrumentation for block profiling</td></tr>
137 <tr><td><a href="#insert-edge-profiling">-insert-edge-profiling</a></td><td>Insert instrumentation for edge profiling</td></tr>
138 <tr><td><a href="#insert-function-profiling">-insert-function-profiling</a></td><td>Insert instrumentation for function profiling</td></tr>
139 <tr><td><a href="#insert-null-profiling-rs">-insert-null-profiling-rs</a></td><td>Measure profiling framework overhead</td></tr>
140 <tr><td><a href="#insert-rs-profiling-framework">-insert-rs-profiling-framework</a></td><td>Insert random sampling instrumentation framework</td></tr>
141 <tr><td><a href="#instcombine">-instcombine</a></td><td>Combine redundant instructions</td></tr>
142 <tr><td><a href="#internalize">-internalize</a></td><td>Internalize Global Symbols</td></tr>
143 <tr><td><a href="#ipconstprop">-ipconstprop</a></td><td>Interprocedural constant propagation</td></tr>
144 <tr><td><a href="#ipsccp">-ipsccp</a></td><td>Interprocedural Sparse Conditional Constant Propagation</td></tr>
145 <tr><td><a href="#jump-threading">-jump-threading</a></td><td>Thread control through conditional blocks </td></tr>
146 <tr><td><a href="#lcssa">-lcssa</a></td><td>Loop-Closed SSA Form Pass</td></tr>
147 <tr><td><a href="#licm">-licm</a></td><td>Loop Invariant Code Motion</td></tr>
148 <tr><td><a href="#loop-deletion">-loop-deletion</a></td><td>Dead Loop Deletion Pass </td></tr>
149 <tr><td><a href="#loop-extract">-loop-extract</a></td><td>Extract loops into new functions</td></tr>
150 <tr><td><a href="#loop-extract-single">-loop-extract-single</a></td><td>Extract at most one loop into a new function</td></tr>
151 <tr><td><a href="#loop-index-split">-loop-index-split</a></td><td>Index Split Loops</td></tr>
152 <tr><td><a href="#loop-reduce">-loop-reduce</a></td><td>Loop Strength Reduction</td></tr>
153 <tr><td><a href="#loop-rotate">-loop-rotate</a></td><td>Rotate Loops</td></tr>
154 <tr><td><a href="#loop-unroll">-loop-unroll</a></td><td>Unroll loops</td></tr>
155 <tr><td><a href="#loop-unswitch">-loop-unswitch</a></td><td>Unswitch loops</td></tr>
156 <tr><td><a href="#loopsimplify">-loopsimplify</a></td><td>Canonicalize natural loops</td></tr>
157 <tr><td><a href="#lowerallocs">-lowerallocs</a></td><td>Lower allocations from instructions to calls</td></tr>
158 <tr><td><a href="#lowerinvoke">-lowerinvoke</a></td><td>Lower invoke and unwind, for unwindless code generators</td></tr>
159 <tr><td><a href="#lowersetjmp">-lowersetjmp</a></td><td>Lower Set Jump</td></tr>
160 <tr><td><a href="#lowerswitch">-lowerswitch</a></td><td>Lower SwitchInst's to branches</td></tr>
161 <tr><td><a href="#mem2reg">-mem2reg</a></td><td>Promote Memory to Register</td></tr>
162 <tr><td><a href="#memcpyopt">-memcpyopt</a></td><td>Optimize use of memcpy and friends</td></tr>
163 <tr><td><a href="#mergereturn">-mergereturn</a></td><td>Unify function exit nodes</td></tr>
164 <tr><td><a href="#predsimplify">-predsimplify</a></td><td>Predicate Simplifier</td></tr>
165 <tr><td><a href="#prune-eh">-prune-eh</a></td><td>Remove unused exception handling info</td></tr>
166 <tr><td><a href="#raiseallocs">-raiseallocs</a></td><td>Raise allocations from calls to instructions</td></tr>
167 <tr><td><a href="#reassociate">-reassociate</a></td><td>Reassociate expressions</td></tr>
168 <tr><td><a href="#reg2mem">-reg2mem</a></td><td>Demote all values to stack slots</td></tr>
169 <tr><td><a href="#scalarrepl">-scalarrepl</a></td><td>Scalar Replacement of Aggregates</td></tr>
170 <tr><td><a href="#sccp">-sccp</a></td><td>Sparse Conditional Constant Propagation</td></tr>
171 <tr><td><a href="#simplify-libcalls">-simplify-libcalls</a></td><td>Simplify well-known library calls</td></tr>
172 <tr><td><a href="#simplifycfg">-simplifycfg</a></td><td>Simplify the CFG</td></tr>
173 <tr><td><a href="#strip">-strip</a></td><td>Strip all symbols from a module</td></tr>
174 <tr><td><a href="#strip-dead-prototypes">-strip-dead-prototypes</a></td><td>Remove unused function declarations</td></tr>
175 <tr><td><a href="#sretpromotion">-sretpromotion</a></td><td>Promote sret arguments</td></tr>
176 <tr><td><a href="#tailcallelim">-tailcallelim</a></td><td>Tail Call Elimination</td></tr>
177 <tr><td><a href="#tailduplicate">-tailduplicate</a></td><td>Tail Duplication</td></tr>
180 <tr><th colspan="2"><b>UTILITY PASSES</b></th></tr>
181 <tr><th>Option</th><th>Name</th></tr>
182 <tr><td><a href="#deadarghaX0r">-deadarghaX0r</a></td><td>Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</td></tr>
183 <tr><td><a href="#extract-blocks">-extract-blocks</a></td><td>Extract Basic Blocks From Module (for bugpoint use)</td></tr>
184 <tr><td><a href="#preverify">-preverify</a></td><td>Preliminary module verification</td></tr>
185 <tr><td><a href="#verify">-verify</a></td><td>Module Verifier</td></tr>
186 <tr><td><a href="#view-cfg">-view-cfg</a></td><td>View CFG of function</td></tr>
187 <tr><td><a href="#view-cfg-only">-view-cfg-only</a></td><td>View CFG of function (with no function bodies)</td></tr>
191 <!-- ======================================================================= -->
192 <div class="doc_section"> <a name="example">Analysis Passes</a></div>
193 <div class="doc_text">
194 <p>This section describes the LLVM Analysis Passes.</p>
197 <!-------------------------------------------------------------------------- -->
198 <div class="doc_subsection">
199 <a name="aa-eval">Exhaustive Alias Analysis Precision Evaluator</a>
201 <div class="doc_text">
202 <p>This is a simple N^2 alias analysis accuracy evaluator.
203 Basically, for each function in the program, it simply queries to see how the
204 alias analysis implementation answers alias queries between each pair of
205 pointers in the function.</p>
207 <p>This is inspired and adapted from code by: Naveen Neelakantam, Francesco
208 Spadini, and Wojciech Stryjewski.</p>
211 <!-------------------------------------------------------------------------- -->
212 <div class="doc_subsection">
213 <a name="anders-aa">Andersen's Interprocedural Alias Analysis</a>
215 <div class="doc_text">
217 This is an implementation of Andersen's interprocedural alias
222 In pointer analysis terms, this is a subset-based, flow-insensitive,
223 field-sensitive, and context-insensitive algorithm pointer algorithm.
227 This algorithm is implemented as three stages:
231 <li>Object identification.</li>
232 <li>Inclusion constraint identification.</li>
233 <li>Offline constraint graph optimization.</li>
234 <li>Inclusion constraint solving.</li>
238 The object identification stage identifies all of the memory objects in the
239 program, which includes globals, heap allocated objects, and stack allocated
244 The inclusion constraint identification stage finds all inclusion constraints
245 in the program by scanning the program, looking for pointer assignments and
246 other statements that effect the points-to graph. For a statement like
247 <code><var>A</var> = <var>B</var></code>, this statement is processed to
248 indicate that <var>A</var> can point to anything that <var>B</var> can point
249 to. Constraints can handle copies, loads, and stores, and address taking.
253 The offline constraint graph optimization portion includes offline variable
254 substitution algorithms intended to computer pointer and location
255 equivalences. Pointer equivalences are those pointers that will have the
256 same points-to sets, and location equivalences are those variables that
257 always appear together in points-to sets.
261 The inclusion constraint solving phase iteratively propagates the inclusion
262 constraints until a fixed point is reached. This is an O(<var>n</var>³)
267 Function constraints are handled as if they were structs with <var>X</var>
268 fields. Thus, an access to argument <var>X</var> of function <var>Y</var> is
269 an access to node index <code>getNode(<var>Y</var>) + <var>X</var></code>.
270 This representation allows handling of indirect calls without any issues. To
271 wit, an indirect call <code><var>Y</var>(<var>a</var>,<var>b</var>)</code> is
272 equivalent to <code>*(<var>Y</var> + 1) = <var>a</var>, *(<var>Y</var> + 2) =
273 <var>b</var></code>. The return node for a function <var>F</var> is always
274 located at <code>getNode(<var>F</var>) + CallReturnPos</code>. The arguments
275 start at <code>getNode(<var>F</var>) + CallArgPos</code>.
279 <!-------------------------------------------------------------------------- -->
280 <div class="doc_subsection">
281 <a name="basicaa">Basic Alias Analysis (default AA impl)</a>
283 <div class="doc_text">
285 This is the default implementation of the Alias Analysis interface
286 that simply implements a few identities (two different globals cannot alias,
287 etc), but otherwise does no analysis.
291 <!-------------------------------------------------------------------------- -->
292 <div class="doc_subsection">
293 <a name="basiccg">Basic CallGraph Construction</a>
295 <div class="doc_text">
296 <p>Yet to be written.</p>
299 <!-------------------------------------------------------------------------- -->
300 <div class="doc_subsection">
301 <a name="basicvn">Basic Value Numbering (default GVN impl)</a>
303 <div class="doc_text">
305 This is the default implementation of the <code>ValueNumbering</code>
306 interface. It walks the SSA def-use chains to trivially identify
307 lexically identical expressions. This does not require any ahead of time
308 analysis, so it is a very fast default implementation.
312 <!-------------------------------------------------------------------------- -->
313 <div class="doc_subsection">
314 <a name="callgraph">Print a call graph</a>
316 <div class="doc_text">
318 This pass, only available in <code>opt</code>, prints the call graph to
319 standard output in a human-readable form.
323 <!-------------------------------------------------------------------------- -->
324 <div class="doc_subsection">
325 <a name="callscc">Print SCCs of the Call Graph</a>
327 <div class="doc_text">
329 This pass, only available in <code>opt</code>, prints the SCCs of the call
330 graph to standard output in a human-readable form.
334 <!-------------------------------------------------------------------------- -->
335 <div class="doc_subsection">
336 <a name="cfgscc">Print SCCs of each function CFG</a>
338 <div class="doc_text">
340 This pass, only available in <code>opt</code>, prints the SCCs of each
341 function CFG to standard output in a human-readable form.
345 <!-------------------------------------------------------------------------- -->
346 <div class="doc_subsection">
347 <a name="codegenprepare">Optimize for code generation</a>
349 <div class="doc_text">
351 This pass munges the code in the input function to better prepare it for
352 SelectionDAG-based code generation. This works around limitations in it's
353 basic-block-at-a-time approach. It should eventually be removed.
357 <!-------------------------------------------------------------------------- -->
358 <div class="doc_subsection">
359 <a name="count-aa">Count Alias Analysis Query Responses</a>
361 <div class="doc_text">
363 A pass which can be used to count how many alias queries
364 are being made and how the alias analysis implementation being used responds.
368 <!-------------------------------------------------------------------------- -->
369 <div class="doc_subsection">
370 <a name="debug-aa">AA use debugger</a>
372 <div class="doc_text">
374 This simple pass checks alias analysis users to ensure that if they
375 create a new value, they do not query AA without informing it of the value.
376 It acts as a shim over any other AA pass you want.
380 Yes keeping track of every value in the program is expensive, but this is
385 <!-------------------------------------------------------------------------- -->
386 <div class="doc_subsection">
387 <a name="domfrontier">Dominance Frontier Construction</a>
389 <div class="doc_text">
391 This pass is a simple dominator construction algorithm for finding forward
396 <!-------------------------------------------------------------------------- -->
397 <div class="doc_subsection">
398 <a name="domtree">Dominator Tree Construction</a>
400 <div class="doc_text">
402 This pass is a simple dominator construction algorithm for finding forward
407 <!-------------------------------------------------------------------------- -->
408 <div class="doc_subsection">
409 <a name="externalfnconstants">Print external fn callsites passed constants</a>
411 <div class="doc_text">
413 This pass, only available in <code>opt</code>, prints out call sites to
414 external functions that are called with constant arguments. This can be
415 useful when looking for standard library functions we should constant fold
416 or handle in alias analyses.
420 <!-------------------------------------------------------------------------- -->
421 <div class="doc_subsection">
422 <a name="globalsmodref-aa">Simple mod/ref analysis for globals</a>
424 <div class="doc_text">
426 This simple pass provides alias and mod/ref information for global values
427 that do not have their address taken, and keeps track of whether functions
428 read or write memory (are "pure"). For this simple (but very common) case,
429 we can provide pretty accurate and useful information.
433 <!-------------------------------------------------------------------------- -->
434 <div class="doc_subsection">
435 <a name="instcount">Counts the various types of Instructions</a>
437 <div class="doc_text">
439 This pass collects the count of all instructions and reports them
443 <!-------------------------------------------------------------------------- -->
444 <div class="doc_subsection">
445 <a name="intervals">Interval Partition Construction</a>
447 <div class="doc_text">
449 This analysis calculates and represents the interval partition of a function,
450 or a preexisting interval partition.
454 In this way, the interval partition may be used to reduce a flow graph down
455 to its degenerate single node interval partition (unless it is irreducible).
459 <!-------------------------------------------------------------------------- -->
460 <div class="doc_subsection">
461 <a name="load-vn">Load Value Numbering</a>
463 <div class="doc_text">
465 This pass value numbers load and call instructions. To do this, it finds
466 lexically identical load instructions, and uses alias analysis to determine
467 which loads are guaranteed to produce the same value. To value number call
468 instructions, it looks for calls to functions that do not write to memory
469 which do not have intervening instructions that clobber the memory that is
474 This pass builds off of another value numbering pass to implement value
475 numbering for non-load and non-call instructions. It uses Alias Analysis so
476 that it can disambiguate the load instructions. The more powerful these base
477 analyses are, the more powerful the resultant value numbering will be.
481 <!-------------------------------------------------------------------------- -->
482 <div class="doc_subsection">
483 <a name="loops">Natural Loop Construction</a>
485 <div class="doc_text">
487 This analysis is used to identify natural loops and determine the loop depth
488 of various nodes of the CFG. Note that the loops identified may actually be
489 several natural loops that share the same header node... not just a single
494 <!-------------------------------------------------------------------------- -->
495 <div class="doc_subsection">
496 <a name="memdep">Memory Dependence Analysis</a>
498 <div class="doc_text">
500 An analysis that determines, for a given memory operation, what preceding
501 memory operations it depends on. It builds on alias analysis information, and
502 tries to provide a lazy, caching interface to a common kind of alias
507 <!-------------------------------------------------------------------------- -->
508 <div class="doc_subsection">
509 <a name="no-aa">No Alias Analysis (always returns 'may' alias)</a>
511 <div class="doc_text">
513 Always returns "I don't know" for alias queries. NoAA is unlike other alias
514 analysis implementations, in that it does not chain to a previous analysis. As
515 such it doesn't follow many of the rules that other alias analyses must.
519 <!-------------------------------------------------------------------------- -->
520 <div class="doc_subsection">
521 <a name="no-profile">No Profile Information</a>
523 <div class="doc_text">
525 The default "no profile" implementation of the abstract
526 <code>ProfileInfo</code> interface.
530 <!-------------------------------------------------------------------------- -->
531 <div class="doc_subsection">
532 <a name="postdomfrontier">Post-Dominance Frontier Construction</a>
534 <div class="doc_text">
536 This pass is a simple post-dominator construction algorithm for finding
537 post-dominator frontiers.
541 <!-------------------------------------------------------------------------- -->
542 <div class="doc_subsection">
543 <a name="postdomtree">Post-Dominator Tree Construction</a>
545 <div class="doc_text">
547 This pass is a simple post-dominator construction algorithm for finding
552 <!-------------------------------------------------------------------------- -->
553 <div class="doc_subsection">
554 <a name="print">Print function to stderr</a>
556 <div class="doc_text">
558 The <code>PrintFunctionPass</code> class is designed to be pipelined with
559 other <code>FunctionPass</code>es, and prints out the functions of the module
560 as they are processed.
564 <!-------------------------------------------------------------------------- -->
565 <div class="doc_subsection">
566 <a name="print-alias-sets">Alias Set Printer</a>
568 <div class="doc_text">
569 <p>Yet to be written.</p>
572 <!-------------------------------------------------------------------------- -->
573 <div class="doc_subsection">
574 <a name="print-callgraph">Print Call Graph to 'dot' file</a>
576 <div class="doc_text">
578 This pass, only available in <code>opt</code>, prints the call graph into a
579 <code>.dot</code> graph. This graph can then be processed with the "dot" tool
580 to convert it to postscript or some other suitable format.
584 <!-------------------------------------------------------------------------- -->
585 <div class="doc_subsection">
586 <a name="print-cfg">Print CFG of function to 'dot' file</a>
588 <div class="doc_text">
590 This pass, only available in <code>opt</code>, prints the control flow graph
591 into a <code>.dot</code> graph. This graph can then be processed with the
592 "dot" tool to convert it to postscript or some other suitable format.
596 <!-------------------------------------------------------------------------- -->
597 <div class="doc_subsection">
598 <a name="print-cfg-only">Print CFG of function to 'dot' file (with no function bodies)</a>
600 <div class="doc_text">
602 This pass, only available in <code>opt</code>, prints the control flow graph
603 into a <code>.dot</code> graph, omitting the function bodies. This graph can
604 then be processed with the "dot" tool to convert it to postscript or some
605 other suitable format.
609 <!-------------------------------------------------------------------------- -->
610 <div class="doc_subsection">
611 <a name="printm">Print module to stderr</a>
613 <div class="doc_text">
615 This pass simply prints out the entire module when it is executed.
619 <!-------------------------------------------------------------------------- -->
620 <div class="doc_subsection">
621 <a name="printusedtypes">Find Used Types</a>
623 <div class="doc_text">
625 This pass is used to seek out all of the types in use by the program. Note
626 that this analysis explicitly does not include types only used by the symbol
630 <!-------------------------------------------------------------------------- -->
631 <div class="doc_subsection">
632 <a name="profile-loader">Load profile information from llvmprof.out</a>
634 <div class="doc_text">
636 A concrete implementation of profiling information that loads the information
637 from a profile dump file.
641 <!-------------------------------------------------------------------------- -->
642 <div class="doc_subsection">
643 <a name="scalar-evolution">Scalar Evolution Analysis</a>
645 <div class="doc_text">
647 The <code>ScalarEvolution</code> analysis can be used to analyze and
648 catagorize scalar expressions in loops. It specializes in recognizing general
649 induction variables, representing them with the abstract and opaque
650 <code>SCEV</code> class. Given this analysis, trip counts of loops and other
651 important properties can be obtained.
655 This analysis is primarily useful for induction variable substitution and
660 <!-------------------------------------------------------------------------- -->
661 <div class="doc_subsection">
662 <a name="targetdata">Target Data Layout</a>
664 <div class="doc_text">
665 <p>Provides other passes access to information on how the size and alignment
666 required by the the target ABI for various data types.</p>
669 <!-- ======================================================================= -->
670 <div class="doc_section"> <a name="transform">Transform Passes</a></div>
671 <div class="doc_text">
672 <p>This section describes the LLVM Transform Passes.</p>
675 <!-------------------------------------------------------------------------- -->
676 <div class="doc_subsection">
677 <a name="adce">Aggressive Dead Code Elimination</a>
679 <div class="doc_text">
680 <p>ADCE aggressively tries to eliminate code. This pass is similar to
681 <a href="#dce">DCE</a> but it assumes that values are dead until proven
682 otherwise. This is similar to <a href="#sccp">SCCP</a>, except applied to
683 the liveness of values.</p>
686 <!-------------------------------------------------------------------------- -->
687 <div class="doc_subsection">
688 <a name="argpromotion">Promote 'by reference' arguments to scalars</a>
690 <div class="doc_text">
692 This pass promotes "by reference" arguments to be "by value" arguments. In
693 practice, this means looking for internal functions that have pointer
694 arguments. If it can prove, through the use of alias analysis, that an
695 argument is *only* loaded, then it can pass the value into the function
696 instead of the address of the value. This can cause recursive simplification
697 of code and lead to the elimination of allocas (especially in C++ template
702 This pass also handles aggregate arguments that are passed into a function,
703 scalarizing them if the elements of the aggregate are only loaded. Note that
704 it refuses to scalarize aggregates which would require passing in more than
705 three operands to the function, because passing thousands of operands for a
706 large array or structure is unprofitable!
710 Note that this transformation could also be done for arguments that are only
711 stored to (returning the value instead), but does not currently. This case
712 would be best handled when and if LLVM starts supporting multiple return
713 values from functions.
717 <!-------------------------------------------------------------------------- -->
718 <div class="doc_subsection">
719 <a name="block-placement">Profile Guided Basic Block Placement</a>
721 <div class="doc_text">
722 <p>This pass is a very simple profile guided basic block placement algorithm.
723 The idea is to put frequently executed blocks together at the start of the
724 function and hopefully increase the number of fall-through conditional
725 branches. If there is no profile information for a particular function, this
726 pass basically orders blocks in depth-first order.</p>
729 <!-------------------------------------------------------------------------- -->
730 <div class="doc_subsection">
731 <a name="break-crit-edges">Break critical edges in CFG</a>
733 <div class="doc_text">
735 Break all of the critical edges in the CFG by inserting a dummy basic block.
736 It may be "required" by passes that cannot deal with critical edges. This
737 transformation obviously invalidates the CFG, but can update forward dominator
738 (set, immediate dominators, tree, and frontier) information.
742 <!-------------------------------------------------------------------------- -->
743 <div class="doc_subsection">
744 <a name="codegenprepare">Prepare a function for code generation</a>
746 <div class="doc_text">
747 This pass munges the code in the input function to better prepare it for
748 SelectionDAG-based code generation. This works around limitations in it's
749 basic-block-at-a-time approach. It should eventually be removed.
752 <!-------------------------------------------------------------------------- -->
753 <div class="doc_subsection">
754 <a name="condprop">Conditional Propagation</a>
756 <div class="doc_text">
757 <p>This pass propagates information about conditional expressions through the
758 program, allowing it to eliminate conditional branches in some cases.</p>
761 <!-------------------------------------------------------------------------- -->
762 <div class="doc_subsection">
763 <a name="constmerge">Merge Duplicate Global Constants</a>
765 <div class="doc_text">
767 Merges duplicate global constants together into a single constant that is
768 shared. This is useful because some passes (ie TraceValues) insert a lot of
769 string constants into the program, regardless of whether or not an existing
774 <!-------------------------------------------------------------------------- -->
775 <div class="doc_subsection">
776 <a name="constprop">Simple constant propagation</a>
778 <div class="doc_text">
779 <p>This file implements constant propagation and merging. It looks for
780 instructions involving only constant operands and replaces them with a
781 constant value instead of an instruction. For example:</p>
782 <blockquote><pre>add i32 1, 2</pre></blockquote>
784 <blockquote><pre>i32 3</pre></blockquote>
785 <p>NOTE: this pass has a habit of making definitions be dead. It is a good
786 idea to to run a <a href="#die">DIE</a> (Dead Instruction Elimination) pass
787 sometime after running this pass.</p>
790 <!-------------------------------------------------------------------------- -->
791 <div class="doc_subsection">
792 <a name="dce">Dead Code Elimination</a>
794 <div class="doc_text">
796 Dead code elimination is similar to <a href="#die">dead instruction
797 elimination</a>, but it rechecks instructions that were used by removed
798 instructions to see if they are newly dead.
802 <!-------------------------------------------------------------------------- -->
803 <div class="doc_subsection">
804 <a name="deadargelim">Dead Argument Elimination</a>
806 <div class="doc_text">
808 This pass deletes dead arguments from internal functions. Dead argument
809 elimination removes arguments which are directly dead, as well as arguments
810 only passed into function calls as dead arguments of other functions. This
811 pass also deletes dead arguments in a similar way.
815 This pass is often useful as a cleanup pass to run after aggressive
816 interprocedural passes, which add possibly-dead arguments.
820 <!-------------------------------------------------------------------------- -->
821 <div class="doc_subsection">
822 <a name="deadtypeelim">Dead Type Elimination</a>
824 <div class="doc_text">
826 This pass is used to cleanup the output of GCC. It eliminate names for types
827 that are unused in the entire translation unit, using the <a
828 href="#findusedtypes">find used types</a> pass.
832 <!-------------------------------------------------------------------------- -->
833 <div class="doc_subsection">
834 <a name="die">Dead Instruction Elimination</a>
836 <div class="doc_text">
838 Dead instruction elimination performs a single pass over the function,
839 removing instructions that are obviously dead.
843 <!-------------------------------------------------------------------------- -->
844 <div class="doc_subsection">
845 <a name="dse">Dead Store Elimination</a>
847 <div class="doc_text">
849 A trivial dead store elimination that only considers basic-block local
854 <!-------------------------------------------------------------------------- -->
855 <div class="doc_subsection">
856 <a name="gcse">Global Common Subexpression Elimination</a>
858 <div class="doc_text">
860 This pass is designed to be a very quick global transformation that
861 eliminates global common subexpressions from a function. It does this by
862 using an existing value numbering implementation to identify the common
863 subexpressions, eliminating them when possible.
867 <!-------------------------------------------------------------------------- -->
868 <div class="doc_subsection">
869 <a name="globaldce">Dead Global Elimination</a>
871 <div class="doc_text">
873 This transform is designed to eliminate unreachable internal globals from the
874 program. It uses an aggressive algorithm, searching out globals that are
875 known to be alive. After it finds all of the globals which are needed, it
876 deletes whatever is left over. This allows it to delete recursive chunks of
877 the program which are unreachable.
881 <!-------------------------------------------------------------------------- -->
882 <div class="doc_subsection">
883 <a name="globalopt">Global Variable Optimizer</a>
885 <div class="doc_text">
887 This pass transforms simple global variables that never have their address
888 taken. If obviously true, it marks read/write globals as constant, deletes
889 variables only stored to, etc.
893 <!-------------------------------------------------------------------------- -->
894 <div class="doc_subsection">
895 <a name="gvn">Global Value Numbering</a>
897 <div class="doc_text">
899 This pass performs global value numbering to eliminate fully redundant
900 instructions. It also performs simple dead load elimination.
904 <!-------------------------------------------------------------------------- -->
905 <div class="doc_subsection">
906 <a name="gvnpre">Global Value Numbering/Partial Redundancy Elimination</a>
908 <div class="doc_text">
910 This pass performs a hybrid of global value numbering and partial redundancy
911 elimination, known as GVN-PRE. It performs partial redundancy elimination on
912 values, rather than lexical expressions, allowing a more comprehensive view
913 the optimization. It replaces redundant values with uses of earlier
914 occurences of the same value. While this is beneficial in that it eliminates
915 unneeded computation, it also increases register pressure by creating large
916 live ranges, and should be used with caution on platforms that are very
917 sensitive to register pressure.
921 <!-------------------------------------------------------------------------- -->
922 <div class="doc_subsection">
923 <a name="indmemrem">Indirect Malloc and Free Removal</a>
925 <div class="doc_text">
927 This pass finds places where memory allocation functions may escape into
928 indirect land. Some transforms are much easier (aka possible) only if free
929 or malloc are not called indirectly.
933 Thus find places where the address of memory functions are taken and construct
934 bounce functions with direct calls of those functions.
938 <!-------------------------------------------------------------------------- -->
939 <div class="doc_subsection">
940 <a name="indvars">Canonicalize Induction Variables</a>
942 <div class="doc_text">
944 This transformation analyzes and transforms the induction variables (and
945 computations derived from them) into simpler forms suitable for subsequent
946 analysis and transformation.
950 This transformation makes the following changes to each loop with an
951 identifiable induction variable:
955 <li>All loops are transformed to have a <em>single</em> canonical
956 induction variable which starts at zero and steps by one.</li>
957 <li>The canonical induction variable is guaranteed to be the first PHI node
958 in the loop header block.</li>
959 <li>Any pointer arithmetic recurrences are raised to use array
964 If the trip count of a loop is computable, this pass also makes the following
969 <li>The exit condition for the loop is canonicalized to compare the
970 induction value against the exit value. This turns loops like:
971 <blockquote><pre>for (i = 7; i*i < 1000; ++i)</pre></blockquote>
973 <blockquote><pre>for (i = 0; i != 25; ++i)</pre></blockquote></li>
974 <li>Any use outside of the loop of an expression derived from the indvar
975 is changed to compute the derived value outside of the loop, eliminating
976 the dependence on the exit value of the induction variable. If the only
977 purpose of the loop is to compute the exit value of some derived
978 expression, this transformation will make the loop dead.</li>
982 This transformation should be followed by strength reduction after all of the
983 desired loop transformations have been performed. Additionally, on targets
984 where it is profitable, the loop could be transformed to count down to zero
985 (the "do loop" optimization).
989 <!-------------------------------------------------------------------------- -->
990 <div class="doc_subsection">
991 <a name="inline">Function Integration/Inlining</a>
993 <div class="doc_text">
995 Bottom-up inlining of functions into callees.
999 <!-------------------------------------------------------------------------- -->
1000 <div class="doc_subsection">
1001 <a name="insert-block-profiling">Insert instrumentation for block profiling</a>
1003 <div class="doc_text">
1005 This pass instruments the specified program with counters for basic block
1006 profiling, which counts the number of times each basic block executes. This
1007 is the most basic form of profiling, which can tell which blocks are hot, but
1008 cannot reliably detect hot paths through the CFG.
1012 Note that this implementation is very naïve. Control equivalent regions of
1013 the CFG should not require duplicate counters, but it does put duplicate
1018 <!-------------------------------------------------------------------------- -->
1019 <div class="doc_subsection">
1020 <a name="insert-edge-profiling">Insert instrumentation for edge profiling</a>
1022 <div class="doc_text">
1024 This pass instruments the specified program with counters for edge profiling.
1025 Edge profiling can give a reasonable approximation of the hot paths through a
1026 program, and is used for a wide variety of program transformations.
1030 Note that this implementation is very naïve. It inserts a counter for
1031 <em>every</em> edge in the program, instead of using control flow information
1032 to prune the number of counters inserted.
1036 <!-------------------------------------------------------------------------- -->
1037 <div class="doc_subsection">
1038 <a name="insert-function-profiling">Insert instrumentation for function profiling</a>
1040 <div class="doc_text">
1042 This pass instruments the specified program with counters for function
1043 profiling, which counts the number of times each function is called.
1047 <!-------------------------------------------------------------------------- -->
1048 <div class="doc_subsection">
1049 <a name="insert-null-profiling-rs">Measure profiling framework overhead</a>
1051 <div class="doc_text">
1053 The basic profiler that does nothing. It is the default profiler and thus
1054 terminates <code>RSProfiler</code> chains. It is useful for measuring
1059 <!-------------------------------------------------------------------------- -->
1060 <div class="doc_subsection">
1061 <a name="insert-rs-profiling-framework">Insert random sampling instrumentation framework</a>
1063 <div class="doc_text">
1065 The second stage of the random-sampling instrumentation framework, duplicates
1066 all instructions in a function, ignoring the profiling code, then connects the
1067 two versions together at the entry and at backedges. At each connection point
1068 a choice is made as to whether to jump to the profiled code (take a sample) or
1069 execute the unprofiled code.
1073 After this pass, it is highly recommended to run<a href="#mem2reg">mem2reg</a>
1074 and <a href="#adce">adce</a>. <a href="#instcombine">instcombine</a>,
1075 <a href="#load-vn">load-vn</a>, <a href="#gdce">gdce</a>, and
1076 <a href="#dse">dse</a> also are good to run afterwards.
1080 <!-------------------------------------------------------------------------- -->
1081 <div class="doc_subsection">
1082 <a name="instcombine">Combine redundant instructions</a>
1084 <div class="doc_text">
1086 Combine instructions to form fewer, simple
1087 instructions. This pass does not modify the CFG This pass is where algebraic
1088 simplification happens.
1092 This pass combines things like:
1097 %Z = add i32 %Y, 1</pre></blockquote>
1104 >%Z = add i32 %X, 2</pre></blockquote>
1107 This is a simple worklist driven algorithm.
1111 This pass guarantees that the following canonicalizations are performed on
1116 <li>If a binary operator has a constant operand, it is moved to the right-
1118 <li>Bitwise operators with constant operands are always grouped so that
1119 shifts are performed first, then <code>or</code>s, then
1120 <code>and</code>s, then <code>xor</code>s.</li>
1121 <li>Compare instructions are converted from <code><</code>,
1122 <code>></code>, <code>≤</code>, or <code>≥</code> to
1123 <code>=</code> or <code>≠</code> if possible.</li>
1124 <li>All <code>cmp</code> instructions on boolean values are replaced with
1125 logical operations.</li>
1126 <li><code>add <var>X</var>, <var>X</var></code> is represented as
1127 <code>mul <var>X</var>, 2</code> ⇒ <code>shl <var>X</var>, 1</code></li>
1128 <li>Multiplies with a constant power-of-two argument are transformed into
1134 <!-------------------------------------------------------------------------- -->
1135 <div class="doc_subsection">
1136 <a name="internalize">Internalize Global Symbols</a>
1138 <div class="doc_text">
1140 This pass loops over all of the functions in the input module, looking for a
1141 main function. If a main function is found, all other functions and all
1142 global variables with initializers are marked as internal.
1146 <!-------------------------------------------------------------------------- -->
1147 <div class="doc_subsection">
1148 <a name="ipconstprop">Interprocedural constant propagation</a>
1150 <div class="doc_text">
1152 This pass implements an <em>extremely</em> simple interprocedural constant
1153 propagation pass. It could certainly be improved in many different ways,
1154 like using a worklist. This pass makes arguments dead, but does not remove
1155 them. The existing dead argument elimination pass should be run after this
1156 to clean up the mess.
1160 <!-------------------------------------------------------------------------- -->
1161 <div class="doc_subsection">
1162 <a name="ipsccp">Interprocedural Sparse Conditional Constant Propagation</a>
1164 <div class="doc_text">
1166 An interprocedural variant of <a href="#sccp">Sparse Conditional Constant
1171 <!-------------------------------------------------------------------------- -->
1172 <div class="doc_subsection">
1173 <a name="jump-threading">Thread control through conditional blocks</a>
1175 <div class="doc_text">
1177 Jump threading tries to find distinct threads of control flow running through
1178 a basic block. This pass looks at blocks that have multiple predecessors and
1179 multiple successors. If one or more of the predecessors of the block can be
1180 proven to always cause a jump to one of the successors, we forward the edge
1181 from the predecessor to the successor by duplicating the contents of this
1185 An example of when this can occur is code like this:
1192 if (X < 3) {</pre>
1195 In this case, the unconditional branch at the end of the first if can be
1196 revectored to the false side of the second if.
1200 <!-------------------------------------------------------------------------- -->
1201 <div class="doc_subsection">
1202 <a name="lcssa">Loop-Closed SSA Form Pass</a>
1204 <div class="doc_text">
1206 This pass transforms loops by placing phi nodes at the end of the loops for
1207 all values that are live across the loop boundary. For example, it turns
1208 the left into the right code:
1212 >for (...) for (...)
1217 X3 = phi(X1, X2) X3 = phi(X1, X2)
1218 ... = X3 + 4 X4 = phi(X3)
1222 This is still valid LLVM; the extra phi nodes are purely redundant, and will
1223 be trivially eliminated by <code>InstCombine</code>. The major benefit of
1224 this transformation is that it makes many other loop optimizations, such as
1225 LoopUnswitching, simpler.
1229 <!-------------------------------------------------------------------------- -->
1230 <div class="doc_subsection">
1231 <a name="licm">Loop Invariant Code Motion</a>
1233 <div class="doc_text">
1235 This pass performs loop invariant code motion, attempting to remove as much
1236 code from the body of a loop as possible. It does this by either hoisting
1237 code into the preheader block, or by sinking code to the exit blocks if it is
1238 safe. This pass also promotes must-aliased memory locations in the loop to
1239 live in registers, thus hoisting and sinking "invariant" loads and stores.
1243 This pass uses alias analysis for two purposes:
1247 <li>Moving loop invariant loads and calls out of loops. If we can determine
1248 that a load or call inside of a loop never aliases anything stored to,
1249 we can hoist it or sink it like any other instruction.</li>
1250 <li>Scalar Promotion of Memory - If there is a store instruction inside of
1251 the loop, we try to move the store to happen AFTER the loop instead of
1252 inside of the loop. This can only happen if a few conditions are true:
1254 <li>The pointer stored through is loop invariant.</li>
1255 <li>There are no stores or loads in the loop which <em>may</em> alias
1256 the pointer. There are no calls in the loop which mod/ref the
1259 If these conditions are true, we can promote the loads and stores in the
1260 loop of the pointer to use a temporary alloca'd variable. We then use
1261 the mem2reg functionality to construct the appropriate SSA form for the
1265 <!-------------------------------------------------------------------------- -->
1266 <div class="doc_subsection">
1267 <a name="loop-deletion">Dead Loop Deletion Pass</a>
1269 <div class="doc_text">
1271 This file implements the Dead Loop Deletion Pass. This pass is responsible
1272 for eliminating loops with non-infinite computable trip counts that have no
1273 side effects or volatile instructions, and do not contribute to the
1274 computation of the function's return value.
1278 <!-------------------------------------------------------------------------- -->
1279 <div class="doc_subsection">
1280 <a name="loop-extract">Extract loops into new functions</a>
1282 <div class="doc_text">
1284 A pass wrapper around the <code>ExtractLoop()</code> scalar transformation to
1285 extract each top-level loop into its own new function. If the loop is the
1286 <em>only</em> loop in a given function, it is not touched. This is a pass most
1287 useful for debugging via bugpoint.
1291 <!-------------------------------------------------------------------------- -->
1292 <div class="doc_subsection">
1293 <a name="loop-extract-single">Extract at most one loop into a new function</a>
1295 <div class="doc_text">
1297 Similar to <a href="#loop-extract">Extract loops into new functions</a>,
1298 this pass extracts one natural loop from the program into a function if it
1299 can. This is used by bugpoint.
1303 <!-------------------------------------------------------------------------- -->
1304 <div class="doc_subsection">
1305 <a name="loop-index-split">Index Split Loops</a>
1307 <div class="doc_text">
1309 This pass divides loop's iteration range by spliting loop such that each
1310 individual loop is executed efficiently.
1314 <!-------------------------------------------------------------------------- -->
1315 <div class="doc_subsection">
1316 <a name="loop-reduce">Loop Strength Reduction</a>
1318 <div class="doc_text">
1320 This pass performs a strength reduction on array references inside loops that
1321 have as one or more of their components the loop induction variable. This is
1322 accomplished by creating a new value to hold the initial value of the array
1323 access for the first iteration, and then creating a new GEP instruction in
1324 the loop to increment the value by the appropriate amount.
1328 <!-------------------------------------------------------------------------- -->
1329 <div class="doc_subsection">
1330 <a name="loop-rotate">Rotate Loops</a>
1332 <div class="doc_text">
1333 <p>A simple loop rotation transformation.</p>
1336 <!-------------------------------------------------------------------------- -->
1337 <div class="doc_subsection">
1338 <a name="loop-unroll">Unroll loops</a>
1340 <div class="doc_text">
1342 This pass implements a simple loop unroller. It works best when loops have
1343 been canonicalized by the <a href="#indvars"><tt>-indvars</tt></a> pass,
1344 allowing it to determine the trip counts of loops easily.
1348 <!-------------------------------------------------------------------------- -->
1349 <div class="doc_subsection">
1350 <a name="loop-unswitch">Unswitch loops</a>
1352 <div class="doc_text">
1354 This pass transforms loops that contain branches on loop-invariant conditions
1355 to have multiple loops. For example, it turns the left into the right code:
1367 This can increase the size of the code exponentially (doubling it every time
1368 a loop is unswitched) so we only unswitch if the resultant code will be
1369 smaller than a threshold.
1373 This pass expects LICM to be run before it to hoist invariant conditions out
1374 of the loop, to make the unswitching opportunity obvious.
1378 <!-------------------------------------------------------------------------- -->
1379 <div class="doc_subsection">
1380 <a name="loopsimplify">Canonicalize natural loops</a>
1382 <div class="doc_text">
1384 This pass performs several transformations to transform natural loops into a
1385 simpler form, which makes subsequent analyses and transformations simpler and
1390 Loop pre-header insertion guarantees that there is a single, non-critical
1391 entry edge from outside of the loop to the loop header. This simplifies a
1392 number of analyses and transformations, such as LICM.
1396 Loop exit-block insertion guarantees that all exit blocks from the loop
1397 (blocks which are outside of the loop that have predecessors inside of the
1398 loop) only have predecessors from inside of the loop (and are thus dominated
1399 by the loop header). This simplifies transformations such as store-sinking
1400 that are built into LICM.
1404 This pass also guarantees that loops will have exactly one backedge.
1408 Note that the simplifycfg pass will clean up blocks which are split out but
1409 end up being unnecessary, so usage of this pass should not pessimize
1414 This pass obviously modifies the CFG, but updates loop information and
1415 dominator information.
1419 <!-------------------------------------------------------------------------- -->
1420 <div class="doc_subsection">
1421 <a name="lowerallocs">Lower allocations from instructions to calls</a>
1423 <div class="doc_text">
1425 Turn <tt>malloc</tt> and <tt>free</tt> instructions into <tt>@malloc</tt> and
1426 <tt>@free</tt> calls.
1430 This is a target-dependent tranformation because it depends on the size of
1431 data types and alignment constraints.
1435 <!-------------------------------------------------------------------------- -->
1436 <div class="doc_subsection">
1437 <a name="lowerinvoke">Lower invoke and unwind, for unwindless code generators</a>
1439 <div class="doc_text">
1441 This transformation is designed for use by code generators which do not yet
1442 support stack unwinding. This pass supports two models of exception handling
1443 lowering, the 'cheap' support and the 'expensive' support.
1447 'Cheap' exception handling support gives the program the ability to execute
1448 any program which does not "throw an exception", by turning 'invoke'
1449 instructions into calls and by turning 'unwind' instructions into calls to
1450 abort(). If the program does dynamically use the unwind instruction, the
1451 program will print a message then abort.
1455 'Expensive' exception handling support gives the full exception handling
1456 support to the program at the cost of making the 'invoke' instruction
1457 really expensive. It basically inserts setjmp/longjmp calls to emulate the
1458 exception handling as necessary.
1462 Because the 'expensive' support slows down programs a lot, and EH is only
1463 used for a subset of the programs, it must be specifically enabled by the
1464 <tt>-enable-correct-eh-support</tt> option.
1468 Note that after this pass runs the CFG is not entirely accurate (exceptional
1469 control flow edges are not correct anymore) so only very simple things should
1470 be done after the lowerinvoke pass has run (like generation of native code).
1471 This should not be used as a general purpose "my LLVM-to-LLVM pass doesn't
1472 support the invoke instruction yet" lowering pass.
1476 <!-------------------------------------------------------------------------- -->
1477 <div class="doc_subsection">
1478 <a name="lowersetjmp">Lower Set Jump</a>
1480 <div class="doc_text">
1482 Lowers <tt>setjmp</tt> and <tt>longjmp</tt> to use the LLVM invoke and unwind
1483 instructions as necessary.
1487 Lowering of <tt>longjmp</tt> is fairly trivial. We replace the call with a
1488 call to the LLVM library function <tt>__llvm_sjljeh_throw_longjmp()</tt>.
1489 This unwinds the stack for us calling all of the destructors for
1490 objects allocated on the stack.
1494 At a <tt>setjmp</tt> call, the basic block is split and the <tt>setjmp</tt>
1495 removed. The calls in a function that have a <tt>setjmp</tt> are converted to
1496 invoke where the except part checks to see if it's a <tt>longjmp</tt>
1497 exception and, if so, if it's handled in the function. If it is, then it gets
1498 the value returned by the <tt>longjmp</tt> and goes to where the basic block
1499 was split. <tt>invoke</tt> instructions are handled in a similar fashion with
1500 the original except block being executed if it isn't a <tt>longjmp</tt>
1501 except that is handled by that function.
1505 <!-------------------------------------------------------------------------- -->
1506 <div class="doc_subsection">
1507 <a name="lowerswitch">Lower SwitchInst's to branches</a>
1509 <div class="doc_text">
1511 Rewrites <tt>switch</tt> instructions with a sequence of branches, which
1512 allows targets to get away with not implementing the switch instruction until
1517 <!-------------------------------------------------------------------------- -->
1518 <div class="doc_subsection">
1519 <a name="mem2reg">Promote Memory to Register</a>
1521 <div class="doc_text">
1523 This file promotes memory references to be register references. It promotes
1524 <tt>alloca</tt> instructions which only have <tt>load</tt>s and
1525 <tt>store</tt>s as uses. An <tt>alloca</tt> is transformed by using dominator
1526 frontiers to place <tt>phi</tt> nodes, then traversing the function in
1527 depth-first order to rewrite <tt>load</tt>s and <tt>store</tt>s as
1528 appropriate. This is just the standard SSA construction algorithm to construct
1533 <!-------------------------------------------------------------------------- -->
1534 <div class="doc_subsection">
1535 <a name="memcpyopt">Optimize use of memcpy and friend</a>
1537 <div class="doc_text">
1539 This pass performs various transformations related to eliminating memcpy
1540 calls, or transforming sets of stores into memset's.
1544 <!-------------------------------------------------------------------------- -->
1545 <div class="doc_subsection">
1546 <a name="mergereturn">Unify function exit nodes</a>
1548 <div class="doc_text">
1550 Ensure that functions have at most one <tt>ret</tt> instruction in them.
1551 Additionally, it keeps track of which node is the new exit node of the CFG.
1555 <!-------------------------------------------------------------------------- -->
1556 <div class="doc_subsection">
1557 <a name="predsimplify">Predicate Simplifier</a>
1559 <div class="doc_text">
1561 Path-sensitive optimizer. In a branch where <tt>x == y</tt>, replace uses of
1562 <tt>x</tt> with <tt>y</tt>. Permits further optimization, such as the
1563 elimination of the unreachable call:
1567 >void test(int *p, int *q)
1573 foo(); // unreachable
1574 }</pre></blockquote>
1577 <!-------------------------------------------------------------------------- -->
1578 <div class="doc_subsection">
1579 <a name="prune-eh">Remove unused exception handling info</a>
1581 <div class="doc_text">
1583 This file implements a simple interprocedural pass which walks the call-graph,
1584 turning <tt>invoke</tt> instructions into <tt>call</tt> instructions if and
1585 only if the callee cannot throw an exception. It implements this as a
1586 bottom-up traversal of the call-graph.
1590 <!-------------------------------------------------------------------------- -->
1591 <div class="doc_subsection">
1592 <a name="raiseallocs">Raise allocations from calls to instructions</a>
1594 <div class="doc_text">
1596 Converts <tt>@malloc</tt> and <tt>@free</tt> calls to <tt>malloc</tt> and
1597 <tt>free</tt> instructions.
1601 <!-------------------------------------------------------------------------- -->
1602 <div class="doc_subsection">
1603 <a name="reassociate">Reassociate expressions</a>
1605 <div class="doc_text">
1607 This pass reassociates commutative expressions in an order that is designed
1608 to promote better constant propagation, GCSE, LICM, PRE, etc.
1612 For example: 4 + (<var>x</var> + 5) ⇒ <var>x</var> + (4 + 5)
1616 In the implementation of this algorithm, constants are assigned rank = 0,
1617 function arguments are rank = 1, and other values are assigned ranks
1618 corresponding to the reverse post order traversal of current function
1619 (starting at 2), which effectively gives values in deep loops higher rank
1620 than values not in loops.
1624 <!-------------------------------------------------------------------------- -->
1625 <div class="doc_subsection">
1626 <a name="reg2mem">Demote all values to stack slots</a>
1628 <div class="doc_text">
1630 This file demotes all registers to memory references. It is intented to be
1631 the inverse of <a href="#mem2reg"><tt>-mem2reg</tt></a>. By converting to
1632 <tt>load</tt> instructions, the only values live accross basic blocks are
1633 <tt>alloca</tt> instructions and <tt>load</tt> instructions before
1634 <tt>phi</tt> nodes. It is intended that this should make CFG hacking much
1635 easier. To make later hacking easier, the entry block is split into two, such
1636 that all introduced <tt>alloca</tt> instructions (and nothing else) are in the
1641 <!-------------------------------------------------------------------------- -->
1642 <div class="doc_subsection">
1643 <a name="scalarrepl">Scalar Replacement of Aggregates</a>
1645 <div class="doc_text">
1647 The well-known scalar replacement of aggregates transformation. This
1648 transform breaks up <tt>alloca</tt> instructions of aggregate type (structure
1649 or array) into individual <tt>alloca</tt> instructions for each member if
1650 possible. Then, if possible, it transforms the individual <tt>alloca</tt>
1651 instructions into nice clean scalar SSA form.
1655 This combines a simple scalar replacement of aggregates algorithm with the <a
1656 href="#mem2reg"><tt>mem2reg</tt></a> algorithm because often interact,
1657 especially for C++ programs. As such, iterating between <tt>scalarrepl</tt>,
1658 then <a href="#mem2reg"><tt>mem2reg</tt></a> until we run out of things to
1663 <!-------------------------------------------------------------------------- -->
1664 <div class="doc_subsection">
1665 <a name="sccp">Sparse Conditional Constant Propagation</a>
1667 <div class="doc_text">
1669 Sparse conditional constant propagation and merging, which can be summarized
1674 <li>Assumes values are constant unless proven otherwise</li>
1675 <li>Assumes BasicBlocks are dead unless proven otherwise</li>
1676 <li>Proves values to be constant, and replaces them with constants</li>
1677 <li>Proves conditional branches to be unconditional</li>
1681 Note that this pass has a habit of making definitions be dead. It is a good
1682 idea to to run a DCE pass sometime after running this pass.
1686 <!-------------------------------------------------------------------------- -->
1687 <div class="doc_subsection">
1688 <a name="simplify-libcalls">Simplify well-known library calls</a>
1690 <div class="doc_text">
1692 Applies a variety of small optimizations for calls to specific well-known
1693 function calls (e.g. runtime library functions). For example, a call
1694 <tt>exit(3)</tt> that occurs within the <tt>main()</tt> function can be
1695 transformed into simply <tt>return 3</tt>.
1699 <!-------------------------------------------------------------------------- -->
1700 <div class="doc_subsection">
1701 <a name="simplifycfg">Simplify the CFG</a>
1703 <div class="doc_text">
1705 Performs dead code elimination and basic block merging. Specifically:
1709 <li>Removes basic blocks with no predecessors.</li>
1710 <li>Merges a basic block into its predecessor if there is only one and the
1711 predecessor only has one successor.</li>
1712 <li>Eliminates PHI nodes for basic blocks with a single predecessor.</li>
1713 <li>Eliminates a basic block that only contains an unconditional
1718 <!-------------------------------------------------------------------------- -->
1719 <div class="doc_subsection">
1720 <a name="strip">Strip all symbols from a module</a>
1722 <div class="doc_text">
1724 Performs code stripping. This transformation can delete:
1728 <li>names for virtual registers</li>
1729 <li>symbols for internal globals and functions</li>
1730 <li>debug information</li>
1734 Note that this transformation makes code much less readable, so it should
1735 only be used in situations where the <tt>strip</tt> utility would be used,
1736 such as reducing code size or making it harder to reverse engineer code.
1740 <!-------------------------------------------------------------------------- -->
1741 <div class="doc_subsection">
1742 <a name="strip-dead-prototypes">Remove unused function declarations</a>
1744 <div class="doc_text">
1746 This pass loops over all of the functions in the input module, looking for
1747 dead declarations and removes them. Dead declarations are declarations of
1748 functions for which no implementation is available (i.e., declarations for
1749 unused library functions).
1753 <!-------------------------------------------------------------------------- -->
1754 <div class="doc_subsection">
1755 <a name="sretpromotion">Promote sret arguments</a>
1757 <div class="doc_text">
1759 This pass finds functions that return a struct (using a pointer to the struct
1760 as the first argument of the function, marked with the '<tt>sret</tt>' attribute) and
1761 replaces them with a new function that simply returns each of the elements of
1762 that struct (using multiple return values).
1766 This pass works under a number of conditions:
1770 <li>The returned struct must not contain other structs</li>
1771 <li>The returned struct must only be used to load values from</li>
1772 <li>The placeholder struct passed in is the result of an <tt>alloca</tt></li>
1776 <!-------------------------------------------------------------------------- -->
1777 <div class="doc_subsection">
1778 <a name="tailcallelim">Tail Call Elimination</a>
1780 <div class="doc_text">
1782 This file transforms calls of the current function (self recursion) followed
1783 by a return instruction with a branch to the entry of the function, creating
1784 a loop. This pass also implements the following extensions to the basic
1789 <li>Trivial instructions between the call and return do not prevent the
1790 transformation from taking place, though currently the analysis cannot
1791 support moving any really useful instructions (only dead ones).
1792 <li>This pass transforms functions that are prevented from being tail
1793 recursive by an associative expression to use an accumulator variable,
1794 thus compiling the typical naive factorial or <tt>fib</tt> implementation
1795 into efficient code.
1796 <li>TRE is performed if the function returns void, if the return
1797 returns the result returned by the call, or if the function returns a
1798 run-time constant on all exits from the function. It is possible, though
1799 unlikely, that the return returns something else (like constant 0), and
1800 can still be TRE'd. It can be TRE'd if <em>all other</em> return
1801 instructions in the function return the exact same value.
1802 <li>If it can prove that callees do not access theier caller stack frame,
1803 they are marked as eligible for tail call elimination (by the code
1808 <!-------------------------------------------------------------------------- -->
1809 <div class="doc_subsection">
1810 <a name="tailduplicate">Tail Duplication</a>
1812 <div class="doc_text">
1814 This pass performs a limited form of tail duplication, intended to simplify
1815 CFGs by removing some unconditional branches. This pass is necessary to
1816 straighten out loops created by the C front-end, but also is capable of
1817 making other code nicer. After this pass is run, the CFG simplify pass
1818 should be run to clean up the mess.
1822 <!-- ======================================================================= -->
1823 <div class="doc_section"> <a name="transform">Utility Passes</a></div>
1824 <div class="doc_text">
1825 <p>This section describes the LLVM Utility Passes.</p>
1828 <!-------------------------------------------------------------------------- -->
1829 <div class="doc_subsection">
1830 <a name="deadarghaX0r">Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</a>
1832 <div class="doc_text">
1834 Same as dead argument elimination, but deletes arguments to functions which
1835 are external. This is only for use by <a
1836 href="Bugpoint.html">bugpoint</a>.</p>
1839 <!-------------------------------------------------------------------------- -->
1840 <div class="doc_subsection">
1841 <a name="extract-blocks">Extract Basic Blocks From Module (for bugpoint use)</a>
1843 <div class="doc_text">
1845 This pass is used by bugpoint to extract all blocks from the module into their
1849 <!-------------------------------------------------------------------------- -->
1850 <div class="doc_subsection">
1851 <a name="preverify">Preliminary module verification</a>
1853 <div class="doc_text">
1855 Ensures that the module is in the form required by the <a
1856 href="#verifier">Module Verifier</a> pass.
1860 Running the verifier runs this pass automatically, so there should be no need
1865 <!-------------------------------------------------------------------------- -->
1866 <div class="doc_subsection">
1867 <a name="verify">Module Verifier</a>
1869 <div class="doc_text">
1871 Verifies an LLVM IR code. This is useful to run after an optimization which is
1872 undergoing testing. Note that <tt>llvm-as</tt> verifies its input before
1873 emitting bitcode, and also that malformed bitcode is likely to make LLVM
1874 crash. All language front-ends are therefore encouraged to verify their output
1875 before performing optimizing transformations.
1879 <li>Both of a binary operator's parameters are of the same type.</li>
1880 <li>Verify that the indices of mem access instructions match other
1882 <li>Verify that arithmetic and other things are only performed on
1883 first-class types. Verify that shifts and logicals only happen on
1885 <li>All of the constants in a switch statement are of the correct type.</li>
1886 <li>The code is in valid SSA form.</li>
1887 <li>It should be illegal to put a label into any other type (like a
1888 structure) or to return one. [except constant arrays!]</li>
1889 <li>Only phi nodes can be self referential: <tt>%x = add i32 %x, %x</tt> is
1891 <li>PHI nodes must have an entry for each predecessor, with no extras.</li>
1892 <li>PHI nodes must be the first thing in a basic block, all grouped
1894 <li>PHI nodes must have at least one entry.</li>
1895 <li>All basic blocks should only end with terminator insts, not contain
1897 <li>The entry node to a function must not have predecessors.</li>
1898 <li>All Instructions must be embedded into a basic block.</li>
1899 <li>Functions cannot take a void-typed parameter.</li>
1900 <li>Verify that a function's argument list agrees with its declared
1902 <li>It is illegal to specify a name for a void value.</li>
1903 <li>It is illegal to have a internal global value with no initializer.</li>
1904 <li>It is illegal to have a ret instruction that returns a value that does
1905 not agree with the function return value type.</li>
1906 <li>Function call argument types match the function prototype.</li>
1907 <li>All other things that are tested by asserts spread about the code.</li>
1911 Note that this does not provide full security verification (like Java), but
1912 instead just tries to ensure that code is well-formed.
1916 <!-------------------------------------------------------------------------- -->
1917 <div class="doc_subsection">
1918 <a name="view-cfg">View CFG of function</a>
1920 <div class="doc_text">
1922 Displays the control flow graph using the GraphViz tool.
1926 <!-------------------------------------------------------------------------- -->
1927 <div class="doc_subsection">
1928 <a name="view-cfg-only">View CFG of function (with no function bodies)</a>
1930 <div class="doc_text">
1932 Displays the control flow graph using the GraphViz tool, but omitting function
1937 <!-- *********************************************************************** -->
1941 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1942 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1943 <a href="http://validator.w3.org/check/referer"><img
1944 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
1946 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a><br>
1947 <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
1948 Last modified: $Date$