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="#codegenprepare">-codegenprepare</a></td><td>Optimize for code generation</td></tr>
82 <tr><td><a href="#count-aa">-count-aa</a></td><td>Count Alias Analysis Query Responses</td></tr>
83 <tr><td><a href="#debug-aa">-debug-aa</a></td><td>AA use debugger</td></tr>
84 <tr><td><a href="#domfrontier">-domfrontier</a></td><td>Dominance Frontier Construction</td></tr>
85 <tr><td><a href="#domtree">-domtree</a></td><td>Dominator Tree Construction</td></tr>
86 <tr><td><a href="#dot-callgraph">-dot-callgraph</a></td><td>Print Call Graph to 'dot' file</td></tr>
87 <tr><td><a href="#dot-cfg">-dot-cfg</a></td><td>Print CFG of function to 'dot' file</td></tr>
88 <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>
89 <tr><td><a href="#globalsmodref-aa">-globalsmodref-aa</a></td><td>Simple mod/ref analysis for globals</td></tr>
90 <tr><td><a href="#instcount">-instcount</a></td><td>Counts the various types of Instructions</td></tr>
91 <tr><td><a href="#intervals">-intervals</a></td><td>Interval Partition Construction</td></tr>
92 <tr><td><a href="#loops">-loops</a></td><td>Natural Loop Construction</td></tr>
93 <tr><td><a href="#memdep">-memdep</a></td><td>Memory Dependence Analysis</td></tr>
94 <tr><td><a href="#no-aa">-no-aa</a></td><td>No Alias Analysis (always returns 'may' alias)</td></tr>
95 <tr><td><a href="#no-profile">-no-profile</a></td><td>No Profile Information</td></tr>
96 <tr><td><a href="#postdomfrontier">-postdomfrontier</a></td><td>Post-Dominance Frontier Construction</td></tr>
97 <tr><td><a href="#postdomtree">-postdomtree</a></td><td>Post-Dominator Tree Construction</td></tr>
98 <tr><td><a href="#print-alias-sets">-print-alias-sets</a></td><td>Alias Set Printer</td></tr>
99 <tr><td><a href="#print-callgraph">-print-callgraph</a></td><td>Print a call graph</td></tr>
100 <tr><td><a href="#print-callgraph-sccs">-print-callgraph-sccs</a></td><td>Print SCCs of the Call Graph</td></tr>
101 <tr><td><a href="#print-cfg-sccs">-print-cfg-sccs</a></td><td>Print SCCs of each function CFG</td></tr>
102 <tr><td><a href="#print-externalfnconstants">-print-externalfnconstants</a></td><td>Print external fn callsites passed constants</td></tr>
103 <tr><td><a href="#print-function">-print-function</a></td><td>Print function to stderr</td></tr>
104 <tr><td><a href="#print-module">-print-module</a></td><td>Print module to stderr</td></tr>
105 <tr><td><a href="#print-used-types">-print-used-types</a></td><td>Find Used Types</td></tr>
106 <tr><td><a href="#profile-loader">-profile-loader</a></td><td>Load profile information from llvmprof.out</td></tr>
107 <tr><td><a href="#scalar-evolution">-scalar-evolution</a></td><td>Scalar Evolution Analysis</td></tr>
108 <tr><td><a href="#targetdata">-targetdata</a></td><td>Target Data Layout</td></tr>
111 <tr><th colspan="2"><b>TRANSFORM PASSES</b></th></tr>
112 <tr><th>Option</th><th>Name</th></tr>
113 <tr><td><a href="#adce">-adce</a></td><td>Aggressive Dead Code Elimination</td></tr>
114 <tr><td><a href="#argpromotion">-argpromotion</a></td><td>Promote 'by reference' arguments to scalars</td></tr>
115 <tr><td><a href="#block-placement">-block-placement</a></td><td>Profile Guided Basic Block Placement</td></tr>
116 <tr><td><a href="#break-crit-edges">-break-crit-edges</a></td><td>Break critical edges in CFG</td></tr>
117 <tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Prepare a function for code generation </td></tr>
118 <tr><td><a href="#condprop">-condprop</a></td><td>Conditional Propagation</td></tr>
119 <tr><td><a href="#constmerge">-constmerge</a></td><td>Merge Duplicate Global Constants</td></tr>
120 <tr><td><a href="#constprop">-constprop</a></td><td>Simple constant propagation</td></tr>
121 <tr><td><a href="#dce">-dce</a></td><td>Dead Code Elimination</td></tr>
122 <tr><td><a href="#deadargelim">-deadargelim</a></td><td>Dead Argument Elimination</td></tr>
123 <tr><td><a href="#deadtypeelim">-deadtypeelim</a></td><td>Dead Type Elimination</td></tr>
124 <tr><td><a href="#die">-die</a></td><td>Dead Instruction Elimination</td></tr>
125 <tr><td><a href="#dse">-dse</a></td><td>Dead Store Elimination</td></tr>
126 <tr><td><a href="#globaldce">-globaldce</a></td><td>Dead Global Elimination</td></tr>
127 <tr><td><a href="#globalopt">-globalopt</a></td><td>Global Variable Optimizer</td></tr>
128 <tr><td><a href="#gvn">-gvn</a></td><td>Global Value Numbering</td></tr>
129 <tr><td><a href="#indmemrem">-indmemrem</a></td><td>Indirect Malloc and Free Removal</td></tr>
130 <tr><td><a href="#indvars">-indvars</a></td><td>Canonicalize Induction Variables</td></tr>
131 <tr><td><a href="#inline">-inline</a></td><td>Function Integration/Inlining</td></tr>
132 <tr><td><a href="#insert-block-profiling">-insert-block-profiling</a></td><td>Insert instrumentation for block profiling</td></tr>
133 <tr><td><a href="#insert-edge-profiling">-insert-edge-profiling</a></td><td>Insert instrumentation for edge profiling</td></tr>
134 <tr><td><a href="#insert-function-profiling">-insert-function-profiling</a></td><td>Insert instrumentation for function profiling</td></tr>
135 <tr><td><a href="#insert-null-profiling-rs">-insert-null-profiling-rs</a></td><td>Measure profiling framework overhead</td></tr>
136 <tr><td><a href="#insert-rs-profiling-framework">-insert-rs-profiling-framework</a></td><td>Insert random sampling instrumentation framework</td></tr>
137 <tr><td><a href="#instcombine">-instcombine</a></td><td>Combine redundant instructions</td></tr>
138 <tr><td><a href="#internalize">-internalize</a></td><td>Internalize Global Symbols</td></tr>
139 <tr><td><a href="#ipconstprop">-ipconstprop</a></td><td>Interprocedural constant propagation</td></tr>
140 <tr><td><a href="#ipsccp">-ipsccp</a></td><td>Interprocedural Sparse Conditional Constant Propagation</td></tr>
141 <tr><td><a href="#jump-threading">-jump-threading</a></td><td>Thread control through conditional blocks </td></tr>
142 <tr><td><a href="#lcssa">-lcssa</a></td><td>Loop-Closed SSA Form Pass</td></tr>
143 <tr><td><a href="#licm">-licm</a></td><td>Loop Invariant Code Motion</td></tr>
144 <tr><td><a href="#loop-deletion">-loop-deletion</a></td><td>Dead Loop Deletion Pass </td></tr>
145 <tr><td><a href="#loop-extract">-loop-extract</a></td><td>Extract loops into new functions</td></tr>
146 <tr><td><a href="#loop-extract-single">-loop-extract-single</a></td><td>Extract at most one loop into a new function</td></tr>
147 <tr><td><a href="#loop-index-split">-loop-index-split</a></td><td>Index Split Loops</td></tr>
148 <tr><td><a href="#loop-reduce">-loop-reduce</a></td><td>Loop Strength Reduction</td></tr>
149 <tr><td><a href="#loop-rotate">-loop-rotate</a></td><td>Rotate Loops</td></tr>
150 <tr><td><a href="#loop-unroll">-loop-unroll</a></td><td>Unroll loops</td></tr>
151 <tr><td><a href="#loop-unswitch">-loop-unswitch</a></td><td>Unswitch loops</td></tr>
152 <tr><td><a href="#loopsimplify">-loopsimplify</a></td><td>Canonicalize natural loops</td></tr>
153 <tr><td><a href="#lowerallocs">-lowerallocs</a></td><td>Lower allocations from instructions to calls</td></tr>
154 <tr><td><a href="#lowerinvoke">-lowerinvoke</a></td><td>Lower invoke and unwind, for unwindless code generators</td></tr>
155 <tr><td><a href="#lowersetjmp">-lowersetjmp</a></td><td>Lower Set Jump</td></tr>
156 <tr><td><a href="#lowerswitch">-lowerswitch</a></td><td>Lower SwitchInst's to branches</td></tr>
157 <tr><td><a href="#mem2reg">-mem2reg</a></td><td>Promote Memory to Register</td></tr>
158 <tr><td><a href="#memcpyopt">-memcpyopt</a></td><td>Optimize use of memcpy and friends</td></tr>
159 <tr><td><a href="#mergereturn">-mergereturn</a></td><td>Unify function exit nodes</td></tr>
160 <tr><td><a href="#prune-eh">-prune-eh</a></td><td>Remove unused exception handling info</td></tr>
161 <tr><td><a href="#raiseallocs">-raiseallocs</a></td><td>Raise allocations from calls to instructions</td></tr>
162 <tr><td><a href="#reassociate">-reassociate</a></td><td>Reassociate expressions</td></tr>
163 <tr><td><a href="#reg2mem">-reg2mem</a></td><td>Demote all values to stack slots</td></tr>
164 <tr><td><a href="#scalarrepl">-scalarrepl</a></td><td>Scalar Replacement of Aggregates</td></tr>
165 <tr><td><a href="#sccp">-sccp</a></td><td>Sparse Conditional Constant Propagation</td></tr>
166 <tr><td><a href="#simplify-libcalls">-simplify-libcalls</a></td><td>Simplify well-known library calls</td></tr>
167 <tr><td><a href="#simplifycfg">-simplifycfg</a></td><td>Simplify the CFG</td></tr>
168 <tr><td><a href="#strip">-strip</a></td><td>Strip all symbols from a module</td></tr>
169 <tr><td><a href="#strip-dead-prototypes">-strip-dead-prototypes</a></td><td>Remove unused function declarations</td></tr>
170 <tr><td><a href="#sretpromotion">-sretpromotion</a></td><td>Promote sret arguments</td></tr>
171 <tr><td><a href="#tailcallelim">-tailcallelim</a></td><td>Tail Call Elimination</td></tr>
172 <tr><td><a href="#tailduplicate">-tailduplicate</a></td><td>Tail Duplication</td></tr>
175 <tr><th colspan="2"><b>UTILITY PASSES</b></th></tr>
176 <tr><th>Option</th><th>Name</th></tr>
177 <tr><td><a href="#deadarghaX0r">-deadarghaX0r</a></td><td>Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</td></tr>
178 <tr><td><a href="#extract-blocks">-extract-blocks</a></td><td>Extract Basic Blocks From Module (for bugpoint use)</td></tr>
179 <tr><td><a href="#preverify">-preverify</a></td><td>Preliminary module verification</td></tr>
180 <tr><td><a href="#verify">-verify</a></td><td>Module Verifier</td></tr>
181 <tr><td><a href="#view-cfg">-view-cfg</a></td><td>View CFG of function</td></tr>
182 <tr><td><a href="#view-cfg-only">-view-cfg-only</a></td><td>View CFG of function (with no function bodies)</td></tr>
186 <!-- ======================================================================= -->
187 <div class="doc_section"> <a name="example">Analysis Passes</a></div>
188 <div class="doc_text">
189 <p>This section describes the LLVM Analysis Passes.</p>
192 <!-------------------------------------------------------------------------- -->
193 <div class="doc_subsection">
194 <a name="aa-eval">Exhaustive Alias Analysis Precision Evaluator</a>
196 <div class="doc_text">
197 <p>This is a simple N^2 alias analysis accuracy evaluator.
198 Basically, for each function in the program, it simply queries to see how the
199 alias analysis implementation answers alias queries between each pair of
200 pointers in the function.</p>
202 <p>This is inspired and adapted from code by: Naveen Neelakantam, Francesco
203 Spadini, and Wojciech Stryjewski.</p>
206 <!-------------------------------------------------------------------------- -->
207 <div class="doc_subsection">
208 <a name="anders-aa">Andersen's Interprocedural Alias Analysis</a>
210 <div class="doc_text">
212 This is an implementation of Andersen's interprocedural alias
217 In pointer analysis terms, this is a subset-based, flow-insensitive,
218 field-sensitive, and context-insensitive algorithm pointer algorithm.
222 This algorithm is implemented as three stages:
226 <li>Object identification.</li>
227 <li>Inclusion constraint identification.</li>
228 <li>Offline constraint graph optimization.</li>
229 <li>Inclusion constraint solving.</li>
233 The object identification stage identifies all of the memory objects in the
234 program, which includes globals, heap allocated objects, and stack allocated
239 The inclusion constraint identification stage finds all inclusion constraints
240 in the program by scanning the program, looking for pointer assignments and
241 other statements that effect the points-to graph. For a statement like
242 <code><var>A</var> = <var>B</var></code>, this statement is processed to
243 indicate that <var>A</var> can point to anything that <var>B</var> can point
244 to. Constraints can handle copies, loads, and stores, and address taking.
248 The offline constraint graph optimization portion includes offline variable
249 substitution algorithms intended to computer pointer and location
250 equivalences. Pointer equivalences are those pointers that will have the
251 same points-to sets, and location equivalences are those variables that
252 always appear together in points-to sets.
256 The inclusion constraint solving phase iteratively propagates the inclusion
257 constraints until a fixed point is reached. This is an O(<var>n</var>³)
262 Function constraints are handled as if they were structs with <var>X</var>
263 fields. Thus, an access to argument <var>X</var> of function <var>Y</var> is
264 an access to node index <code>getNode(<var>Y</var>) + <var>X</var></code>.
265 This representation allows handling of indirect calls without any issues. To
266 wit, an indirect call <code><var>Y</var>(<var>a</var>,<var>b</var>)</code> is
267 equivalent to <code>*(<var>Y</var> + 1) = <var>a</var>, *(<var>Y</var> + 2) =
268 <var>b</var></code>. The return node for a function <var>F</var> is always
269 located at <code>getNode(<var>F</var>) + CallReturnPos</code>. The arguments
270 start at <code>getNode(<var>F</var>) + CallArgPos</code>.
274 Please keep in mind that the current andersen's pass has many known
275 problems and bugs. It should be considered "research quality".
280 <!-------------------------------------------------------------------------- -->
281 <div class="doc_subsection">
282 <a name="basicaa">Basic Alias Analysis (default AA impl)</a>
284 <div class="doc_text">
286 This is the default implementation of the Alias Analysis interface
287 that simply implements a few identities (two different globals cannot alias,
288 etc), but otherwise does no analysis.
292 <!-------------------------------------------------------------------------- -->
293 <div class="doc_subsection">
294 <a name="basiccg">Basic CallGraph Construction</a>
296 <div class="doc_text">
297 <p>Yet to be written.</p>
300 <!-------------------------------------------------------------------------- -->
301 <div class="doc_subsection">
302 <a name="codegenprepare">Optimize for code generation</a>
304 <div class="doc_text">
306 This pass munges the code in the input function to better prepare it for
307 SelectionDAG-based code generation. This works around limitations in it's
308 basic-block-at-a-time approach. It should eventually be removed.
312 <!-------------------------------------------------------------------------- -->
313 <div class="doc_subsection">
314 <a name="count-aa">Count Alias Analysis Query Responses</a>
316 <div class="doc_text">
318 A pass which can be used to count how many alias queries
319 are being made and how the alias analysis implementation being used responds.
323 <!-------------------------------------------------------------------------- -->
324 <div class="doc_subsection">
325 <a name="debug-aa">AA use debugger</a>
327 <div class="doc_text">
329 This simple pass checks alias analysis users to ensure that if they
330 create a new value, they do not query AA without informing it of the value.
331 It acts as a shim over any other AA pass you want.
335 Yes keeping track of every value in the program is expensive, but this is
340 <!-------------------------------------------------------------------------- -->
341 <div class="doc_subsection">
342 <a name="domfrontier">Dominance Frontier Construction</a>
344 <div class="doc_text">
346 This pass is a simple dominator construction algorithm for finding forward
351 <!-------------------------------------------------------------------------- -->
352 <div class="doc_subsection">
353 <a name="domtree">Dominator Tree Construction</a>
355 <div class="doc_text">
357 This pass is a simple dominator construction algorithm for finding forward
362 <!-------------------------------------------------------------------------- -->
363 <div class="doc_subsection">
364 <a name="dot-callgraph">Print Call Graph to 'dot' file</a>
366 <div class="doc_text">
368 This pass, only available in <code>opt</code>, prints the call graph into a
369 <code>.dot</code> graph. This graph can then be processed with the "dot" tool
370 to convert it to postscript or some other suitable format.
374 <!-------------------------------------------------------------------------- -->
375 <div class="doc_subsection">
376 <a name="dot-cfg">Print CFG of function to 'dot' file</a>
378 <div class="doc_text">
380 This pass, only available in <code>opt</code>, prints the control flow graph
381 into a <code>.dot</code> graph. This graph can then be processed with the
382 "dot" tool to convert it to postscript or some other suitable format.
386 <!-------------------------------------------------------------------------- -->
387 <div class="doc_subsection">
388 <a name="dot-cfg-only">Print CFG of function to 'dot' file (with no function bodies)</a>
390 <div class="doc_text">
392 This pass, only available in <code>opt</code>, prints the control flow graph
393 into a <code>.dot</code> graph, omitting the function bodies. This graph can
394 then be processed with the "dot" tool to convert it to postscript or some
395 other suitable format.
399 <!-------------------------------------------------------------------------- -->
400 <div class="doc_subsection">
401 <a name="globalsmodref-aa">Simple mod/ref analysis for globals</a>
403 <div class="doc_text">
405 This simple pass provides alias and mod/ref information for global values
406 that do not have their address taken, and keeps track of whether functions
407 read or write memory (are "pure"). For this simple (but very common) case,
408 we can provide pretty accurate and useful information.
412 <!-------------------------------------------------------------------------- -->
413 <div class="doc_subsection">
414 <a name="instcount">Counts the various types of Instructions</a>
416 <div class="doc_text">
418 This pass collects the count of all instructions and reports them
422 <!-------------------------------------------------------------------------- -->
423 <div class="doc_subsection">
424 <a name="intervals">Interval Partition Construction</a>
426 <div class="doc_text">
428 This analysis calculates and represents the interval partition of a function,
429 or a preexisting interval partition.
433 In this way, the interval partition may be used to reduce a flow graph down
434 to its degenerate single node interval partition (unless it is irreducible).
438 <!-------------------------------------------------------------------------- -->
439 <div class="doc_subsection">
440 <a name="loops">Natural Loop Construction</a>
442 <div class="doc_text">
444 This analysis is used to identify natural loops and determine the loop depth
445 of various nodes of the CFG. Note that the loops identified may actually be
446 several natural loops that share the same header node... not just a single
451 <!-------------------------------------------------------------------------- -->
452 <div class="doc_subsection">
453 <a name="memdep">Memory Dependence Analysis</a>
455 <div class="doc_text">
457 An analysis that determines, for a given memory operation, what preceding
458 memory operations it depends on. It builds on alias analysis information, and
459 tries to provide a lazy, caching interface to a common kind of alias
464 <!-------------------------------------------------------------------------- -->
465 <div class="doc_subsection">
466 <a name="no-aa">No Alias Analysis (always returns 'may' alias)</a>
468 <div class="doc_text">
470 Always returns "I don't know" for alias queries. NoAA is unlike other alias
471 analysis implementations, in that it does not chain to a previous analysis. As
472 such it doesn't follow many of the rules that other alias analyses must.
476 <!-------------------------------------------------------------------------- -->
477 <div class="doc_subsection">
478 <a name="no-profile">No Profile Information</a>
480 <div class="doc_text">
482 The default "no profile" implementation of the abstract
483 <code>ProfileInfo</code> interface.
487 <!-------------------------------------------------------------------------- -->
488 <div class="doc_subsection">
489 <a name="postdomfrontier">Post-Dominance Frontier Construction</a>
491 <div class="doc_text">
493 This pass is a simple post-dominator construction algorithm for finding
494 post-dominator frontiers.
498 <!-------------------------------------------------------------------------- -->
499 <div class="doc_subsection">
500 <a name="postdomtree">Post-Dominator Tree Construction</a>
502 <div class="doc_text">
504 This pass is a simple post-dominator construction algorithm for finding
509 <!-------------------------------------------------------------------------- -->
510 <div class="doc_subsection">
511 <a name="print-alias-sets">Alias Set Printer</a>
513 <div class="doc_text">
514 <p>Yet to be written.</p>
517 <!-------------------------------------------------------------------------- -->
518 <div class="doc_subsection">
519 <a name="print-callgraph">Print a call graph</a>
521 <div class="doc_text">
523 This pass, only available in <code>opt</code>, prints the call graph to
524 standard output in a human-readable form.
528 <!-------------------------------------------------------------------------- -->
529 <div class="doc_subsection">
530 <a name="print-callgraph-sccs">Print SCCs of the Call Graph</a>
532 <div class="doc_text">
534 This pass, only available in <code>opt</code>, prints the SCCs of the call
535 graph to standard output in a human-readable form.
539 <!-------------------------------------------------------------------------- -->
540 <div class="doc_subsection">
541 <a name="print-cfg-sccs">Print SCCs of each function CFG</a>
543 <div class="doc_text">
545 This pass, only available in <code>opt</code>, prints the SCCs of each
546 function CFG to standard output in a human-readable form.
550 <!-------------------------------------------------------------------------- -->
551 <div class="doc_subsection">
552 <a name="print-externalfnconstants">Print external fn callsites passed constants</a>
554 <div class="doc_text">
556 This pass, only available in <code>opt</code>, prints out call sites to
557 external functions that are called with constant arguments. This can be
558 useful when looking for standard library functions we should constant fold
559 or handle in alias analyses.
563 <!-------------------------------------------------------------------------- -->
564 <div class="doc_subsection">
565 <a name="print-function">Print function to stderr</a>
567 <div class="doc_text">
569 The <code>PrintFunctionPass</code> class is designed to be pipelined with
570 other <code>FunctionPass</code>es, and prints out the functions of the module
571 as they are processed.
575 <!-------------------------------------------------------------------------- -->
576 <div class="doc_subsection">
577 <a name="print-module">Print module to stderr</a>
579 <div class="doc_text">
581 This pass simply prints out the entire module when it is executed.
585 <!-------------------------------------------------------------------------- -->
586 <div class="doc_subsection">
587 <a name="print-used-types">Find Used Types</a>
589 <div class="doc_text">
591 This pass is used to seek out all of the types in use by the program. Note
592 that this analysis explicitly does not include types only used by the symbol
596 <!-------------------------------------------------------------------------- -->
597 <div class="doc_subsection">
598 <a name="profile-loader">Load profile information from llvmprof.out</a>
600 <div class="doc_text">
602 A concrete implementation of profiling information that loads the information
603 from a profile dump file.
607 <!-------------------------------------------------------------------------- -->
608 <div class="doc_subsection">
609 <a name="scalar-evolution">Scalar Evolution Analysis</a>
611 <div class="doc_text">
613 The <code>ScalarEvolution</code> analysis can be used to analyze and
614 catagorize scalar expressions in loops. It specializes in recognizing general
615 induction variables, representing them with the abstract and opaque
616 <code>SCEV</code> class. Given this analysis, trip counts of loops and other
617 important properties can be obtained.
621 This analysis is primarily useful for induction variable substitution and
626 <!-------------------------------------------------------------------------- -->
627 <div class="doc_subsection">
628 <a name="targetdata">Target Data Layout</a>
630 <div class="doc_text">
631 <p>Provides other passes access to information on how the size and alignment
632 required by the the target ABI for various data types.</p>
635 <!-- ======================================================================= -->
636 <div class="doc_section"> <a name="transform">Transform Passes</a></div>
637 <div class="doc_text">
638 <p>This section describes the LLVM Transform Passes.</p>
641 <!-------------------------------------------------------------------------- -->
642 <div class="doc_subsection">
643 <a name="adce">Aggressive Dead Code Elimination</a>
645 <div class="doc_text">
646 <p>ADCE aggressively tries to eliminate code. This pass is similar to
647 <a href="#dce">DCE</a> but it assumes that values are dead until proven
648 otherwise. This is similar to <a href="#sccp">SCCP</a>, except applied to
649 the liveness of values.</p>
652 <!-------------------------------------------------------------------------- -->
653 <div class="doc_subsection">
654 <a name="argpromotion">Promote 'by reference' arguments to scalars</a>
656 <div class="doc_text">
658 This pass promotes "by reference" arguments to be "by value" arguments. In
659 practice, this means looking for internal functions that have pointer
660 arguments. If it can prove, through the use of alias analysis, that an
661 argument is *only* loaded, then it can pass the value into the function
662 instead of the address of the value. This can cause recursive simplification
663 of code and lead to the elimination of allocas (especially in C++ template
668 This pass also handles aggregate arguments that are passed into a function,
669 scalarizing them if the elements of the aggregate are only loaded. Note that
670 it refuses to scalarize aggregates which would require passing in more than
671 three operands to the function, because passing thousands of operands for a
672 large array or structure is unprofitable!
676 Note that this transformation could also be done for arguments that are only
677 stored to (returning the value instead), but does not currently. This case
678 would be best handled when and if LLVM starts supporting multiple return
679 values from functions.
683 <!-------------------------------------------------------------------------- -->
684 <div class="doc_subsection">
685 <a name="block-placement">Profile Guided Basic Block Placement</a>
687 <div class="doc_text">
688 <p>This pass is a very simple profile guided basic block placement algorithm.
689 The idea is to put frequently executed blocks together at the start of the
690 function and hopefully increase the number of fall-through conditional
691 branches. If there is no profile information for a particular function, this
692 pass basically orders blocks in depth-first order.</p>
695 <!-------------------------------------------------------------------------- -->
696 <div class="doc_subsection">
697 <a name="break-crit-edges">Break critical edges in CFG</a>
699 <div class="doc_text">
701 Break all of the critical edges in the CFG by inserting a dummy basic block.
702 It may be "required" by passes that cannot deal with critical edges. This
703 transformation obviously invalidates the CFG, but can update forward dominator
704 (set, immediate dominators, tree, and frontier) information.
708 <!-------------------------------------------------------------------------- -->
709 <div class="doc_subsection">
710 <a name="codegenprepare">Prepare a function for code generation</a>
712 <div class="doc_text">
713 This pass munges the code in the input function to better prepare it for
714 SelectionDAG-based code generation. This works around limitations in it's
715 basic-block-at-a-time approach. It should eventually be removed.
718 <!-------------------------------------------------------------------------- -->
719 <div class="doc_subsection">
720 <a name="condprop">Conditional Propagation</a>
722 <div class="doc_text">
723 <p>This pass propagates information about conditional expressions through the
724 program, allowing it to eliminate conditional branches in some cases.</p>
727 <!-------------------------------------------------------------------------- -->
728 <div class="doc_subsection">
729 <a name="constmerge">Merge Duplicate Global Constants</a>
731 <div class="doc_text">
733 Merges duplicate global constants together into a single constant that is
734 shared. This is useful because some passes (ie TraceValues) insert a lot of
735 string constants into the program, regardless of whether or not an existing
740 <!-------------------------------------------------------------------------- -->
741 <div class="doc_subsection">
742 <a name="constprop">Simple constant propagation</a>
744 <div class="doc_text">
745 <p>This file implements constant propagation and merging. It looks for
746 instructions involving only constant operands and replaces them with a
747 constant value instead of an instruction. For example:</p>
748 <blockquote><pre>add i32 1, 2</pre></blockquote>
750 <blockquote><pre>i32 3</pre></blockquote>
751 <p>NOTE: this pass has a habit of making definitions be dead. It is a good
752 idea to to run a <a href="#die">DIE</a> (Dead Instruction Elimination) pass
753 sometime after running this pass.</p>
756 <!-------------------------------------------------------------------------- -->
757 <div class="doc_subsection">
758 <a name="dce">Dead Code Elimination</a>
760 <div class="doc_text">
762 Dead code elimination is similar to <a href="#die">dead instruction
763 elimination</a>, but it rechecks instructions that were used by removed
764 instructions to see if they are newly dead.
768 <!-------------------------------------------------------------------------- -->
769 <div class="doc_subsection">
770 <a name="deadargelim">Dead Argument Elimination</a>
772 <div class="doc_text">
774 This pass deletes dead arguments from internal functions. Dead argument
775 elimination removes arguments which are directly dead, as well as arguments
776 only passed into function calls as dead arguments of other functions. This
777 pass also deletes dead arguments in a similar way.
781 This pass is often useful as a cleanup pass to run after aggressive
782 interprocedural passes, which add possibly-dead arguments.
786 <!-------------------------------------------------------------------------- -->
787 <div class="doc_subsection">
788 <a name="deadtypeelim">Dead Type Elimination</a>
790 <div class="doc_text">
792 This pass is used to cleanup the output of GCC. It eliminate names for types
793 that are unused in the entire translation unit, using the <a
794 href="#findusedtypes">find used types</a> pass.
798 <!-------------------------------------------------------------------------- -->
799 <div class="doc_subsection">
800 <a name="die">Dead Instruction Elimination</a>
802 <div class="doc_text">
804 Dead instruction elimination performs a single pass over the function,
805 removing instructions that are obviously dead.
809 <!-------------------------------------------------------------------------- -->
810 <div class="doc_subsection">
811 <a name="dse">Dead Store Elimination</a>
813 <div class="doc_text">
815 A trivial dead store elimination that only considers basic-block local
820 <!-------------------------------------------------------------------------- -->
821 <div class="doc_subsection">
822 <a name="globaldce">Dead Global Elimination</a>
824 <div class="doc_text">
826 This transform is designed to eliminate unreachable internal globals from the
827 program. It uses an aggressive algorithm, searching out globals that are
828 known to be alive. After it finds all of the globals which are needed, it
829 deletes whatever is left over. This allows it to delete recursive chunks of
830 the program which are unreachable.
834 <!-------------------------------------------------------------------------- -->
835 <div class="doc_subsection">
836 <a name="globalopt">Global Variable Optimizer</a>
838 <div class="doc_text">
840 This pass transforms simple global variables that never have their address
841 taken. If obviously true, it marks read/write globals as constant, deletes
842 variables only stored to, etc.
846 <!-------------------------------------------------------------------------- -->
847 <div class="doc_subsection">
848 <a name="gvn">Global Value Numbering</a>
850 <div class="doc_text">
852 This pass performs global value numbering to eliminate fully and partially
853 redundant instructions. It also performs redundant load elimination.
858 <!-------------------------------------------------------------------------- -->
859 <div class="doc_subsection">
860 <a name="indmemrem">Indirect Malloc and Free Removal</a>
862 <div class="doc_text">
864 This pass finds places where memory allocation functions may escape into
865 indirect land. Some transforms are much easier (aka possible) only if free
866 or malloc are not called indirectly.
870 Thus find places where the address of memory functions are taken and construct
871 bounce functions with direct calls of those functions.
875 <!-------------------------------------------------------------------------- -->
876 <div class="doc_subsection">
877 <a name="indvars">Canonicalize Induction Variables</a>
879 <div class="doc_text">
881 This transformation analyzes and transforms the induction variables (and
882 computations derived from them) into simpler forms suitable for subsequent
883 analysis and transformation.
887 This transformation makes the following changes to each loop with an
888 identifiable induction variable:
892 <li>All loops are transformed to have a <em>single</em> canonical
893 induction variable which starts at zero and steps by one.</li>
894 <li>The canonical induction variable is guaranteed to be the first PHI node
895 in the loop header block.</li>
896 <li>Any pointer arithmetic recurrences are raised to use array
901 If the trip count of a loop is computable, this pass also makes the following
906 <li>The exit condition for the loop is canonicalized to compare the
907 induction value against the exit value. This turns loops like:
908 <blockquote><pre>for (i = 7; i*i < 1000; ++i)</pre></blockquote>
910 <blockquote><pre>for (i = 0; i != 25; ++i)</pre></blockquote></li>
911 <li>Any use outside of the loop of an expression derived from the indvar
912 is changed to compute the derived value outside of the loop, eliminating
913 the dependence on the exit value of the induction variable. If the only
914 purpose of the loop is to compute the exit value of some derived
915 expression, this transformation will make the loop dead.</li>
919 This transformation should be followed by strength reduction after all of the
920 desired loop transformations have been performed. Additionally, on targets
921 where it is profitable, the loop could be transformed to count down to zero
922 (the "do loop" optimization).
926 <!-------------------------------------------------------------------------- -->
927 <div class="doc_subsection">
928 <a name="inline">Function Integration/Inlining</a>
930 <div class="doc_text">
932 Bottom-up inlining of functions into callees.
936 <!-------------------------------------------------------------------------- -->
937 <div class="doc_subsection">
938 <a name="insert-block-profiling">Insert instrumentation for block profiling</a>
940 <div class="doc_text">
942 This pass instruments the specified program with counters for basic block
943 profiling, which counts the number of times each basic block executes. This
944 is the most basic form of profiling, which can tell which blocks are hot, but
945 cannot reliably detect hot paths through the CFG.
949 Note that this implementation is very naïve. Control equivalent regions of
950 the CFG should not require duplicate counters, but it does put duplicate
955 <!-------------------------------------------------------------------------- -->
956 <div class="doc_subsection">
957 <a name="insert-edge-profiling">Insert instrumentation for edge profiling</a>
959 <div class="doc_text">
961 This pass instruments the specified program with counters for edge profiling.
962 Edge profiling can give a reasonable approximation of the hot paths through a
963 program, and is used for a wide variety of program transformations.
967 Note that this implementation is very naïve. It inserts a counter for
968 <em>every</em> edge in the program, instead of using control flow information
969 to prune the number of counters inserted.
973 <!-------------------------------------------------------------------------- -->
974 <div class="doc_subsection">
975 <a name="insert-function-profiling">Insert instrumentation for function profiling</a>
977 <div class="doc_text">
979 This pass instruments the specified program with counters for function
980 profiling, which counts the number of times each function is called.
984 <!-------------------------------------------------------------------------- -->
985 <div class="doc_subsection">
986 <a name="insert-null-profiling-rs">Measure profiling framework overhead</a>
988 <div class="doc_text">
990 The basic profiler that does nothing. It is the default profiler and thus
991 terminates <code>RSProfiler</code> chains. It is useful for measuring
996 <!-------------------------------------------------------------------------- -->
997 <div class="doc_subsection">
998 <a name="insert-rs-profiling-framework">Insert random sampling instrumentation framework</a>
1000 <div class="doc_text">
1002 The second stage of the random-sampling instrumentation framework, duplicates
1003 all instructions in a function, ignoring the profiling code, then connects the
1004 two versions together at the entry and at backedges. At each connection point
1005 a choice is made as to whether to jump to the profiled code (take a sample) or
1006 execute the unprofiled code.
1010 After this pass, it is highly recommended to run<a href="#mem2reg">mem2reg</a>
1011 and <a href="#adce">adce</a>. <a href="#instcombine">instcombine</a>,
1012 <a href="#load-vn">load-vn</a>, <a href="#gdce">gdce</a>, and
1013 <a href="#dse">dse</a> also are good to run afterwards.
1017 <!-------------------------------------------------------------------------- -->
1018 <div class="doc_subsection">
1019 <a name="instcombine">Combine redundant instructions</a>
1021 <div class="doc_text">
1023 Combine instructions to form fewer, simple
1024 instructions. This pass does not modify the CFG This pass is where algebraic
1025 simplification happens.
1029 This pass combines things like:
1034 %Z = add i32 %Y, 1</pre></blockquote>
1041 >%Z = add i32 %X, 2</pre></blockquote>
1044 This is a simple worklist driven algorithm.
1048 This pass guarantees that the following canonicalizations are performed on
1053 <li>If a binary operator has a constant operand, it is moved to the right-
1055 <li>Bitwise operators with constant operands are always grouped so that
1056 shifts are performed first, then <code>or</code>s, then
1057 <code>and</code>s, then <code>xor</code>s.</li>
1058 <li>Compare instructions are converted from <code><</code>,
1059 <code>></code>, <code>≤</code>, or <code>≥</code> to
1060 <code>=</code> or <code>≠</code> if possible.</li>
1061 <li>All <code>cmp</code> instructions on boolean values are replaced with
1062 logical operations.</li>
1063 <li><code>add <var>X</var>, <var>X</var></code> is represented as
1064 <code>mul <var>X</var>, 2</code> ⇒ <code>shl <var>X</var>, 1</code></li>
1065 <li>Multiplies with a constant power-of-two argument are transformed into
1071 <!-------------------------------------------------------------------------- -->
1072 <div class="doc_subsection">
1073 <a name="internalize">Internalize Global Symbols</a>
1075 <div class="doc_text">
1077 This pass loops over all of the functions in the input module, looking for a
1078 main function. If a main function is found, all other functions and all
1079 global variables with initializers are marked as internal.
1083 <!-------------------------------------------------------------------------- -->
1084 <div class="doc_subsection">
1085 <a name="ipconstprop">Interprocedural constant propagation</a>
1087 <div class="doc_text">
1089 This pass implements an <em>extremely</em> simple interprocedural constant
1090 propagation pass. It could certainly be improved in many different ways,
1091 like using a worklist. This pass makes arguments dead, but does not remove
1092 them. The existing dead argument elimination pass should be run after this
1093 to clean up the mess.
1097 <!-------------------------------------------------------------------------- -->
1098 <div class="doc_subsection">
1099 <a name="ipsccp">Interprocedural Sparse Conditional Constant Propagation</a>
1101 <div class="doc_text">
1103 An interprocedural variant of <a href="#sccp">Sparse Conditional Constant
1108 <!-------------------------------------------------------------------------- -->
1109 <div class="doc_subsection">
1110 <a name="jump-threading">Thread control through conditional blocks</a>
1112 <div class="doc_text">
1114 Jump threading tries to find distinct threads of control flow running through
1115 a basic block. This pass looks at blocks that have multiple predecessors and
1116 multiple successors. If one or more of the predecessors of the block can be
1117 proven to always cause a jump to one of the successors, we forward the edge
1118 from the predecessor to the successor by duplicating the contents of this
1122 An example of when this can occur is code like this:
1129 if (X < 3) {</pre>
1132 In this case, the unconditional branch at the end of the first if can be
1133 revectored to the false side of the second if.
1137 <!-------------------------------------------------------------------------- -->
1138 <div class="doc_subsection">
1139 <a name="lcssa">Loop-Closed SSA Form Pass</a>
1141 <div class="doc_text">
1143 This pass transforms loops by placing phi nodes at the end of the loops for
1144 all values that are live across the loop boundary. For example, it turns
1145 the left into the right code:
1149 >for (...) for (...)
1154 X3 = phi(X1, X2) X3 = phi(X1, X2)
1155 ... = X3 + 4 X4 = phi(X3)
1159 This is still valid LLVM; the extra phi nodes are purely redundant, and will
1160 be trivially eliminated by <code>InstCombine</code>. The major benefit of
1161 this transformation is that it makes many other loop optimizations, such as
1162 LoopUnswitching, simpler.
1166 <!-------------------------------------------------------------------------- -->
1167 <div class="doc_subsection">
1168 <a name="licm">Loop Invariant Code Motion</a>
1170 <div class="doc_text">
1172 This pass performs loop invariant code motion, attempting to remove as much
1173 code from the body of a loop as possible. It does this by either hoisting
1174 code into the preheader block, or by sinking code to the exit blocks if it is
1175 safe. This pass also promotes must-aliased memory locations in the loop to
1176 live in registers, thus hoisting and sinking "invariant" loads and stores.
1180 This pass uses alias analysis for two purposes:
1184 <li>Moving loop invariant loads and calls out of loops. If we can determine
1185 that a load or call inside of a loop never aliases anything stored to,
1186 we can hoist it or sink it like any other instruction.</li>
1187 <li>Scalar Promotion of Memory - If there is a store instruction inside of
1188 the loop, we try to move the store to happen AFTER the loop instead of
1189 inside of the loop. This can only happen if a few conditions are true:
1191 <li>The pointer stored through is loop invariant.</li>
1192 <li>There are no stores or loads in the loop which <em>may</em> alias
1193 the pointer. There are no calls in the loop which mod/ref the
1196 If these conditions are true, we can promote the loads and stores in the
1197 loop of the pointer to use a temporary alloca'd variable. We then use
1198 the mem2reg functionality to construct the appropriate SSA form for the
1202 <!-------------------------------------------------------------------------- -->
1203 <div class="doc_subsection">
1204 <a name="loop-deletion">Dead Loop Deletion Pass</a>
1206 <div class="doc_text">
1208 This file implements the Dead Loop Deletion Pass. This pass is responsible
1209 for eliminating loops with non-infinite computable trip counts that have no
1210 side effects or volatile instructions, and do not contribute to the
1211 computation of the function's return value.
1215 <!-------------------------------------------------------------------------- -->
1216 <div class="doc_subsection">
1217 <a name="loop-extract">Extract loops into new functions</a>
1219 <div class="doc_text">
1221 A pass wrapper around the <code>ExtractLoop()</code> scalar transformation to
1222 extract each top-level loop into its own new function. If the loop is the
1223 <em>only</em> loop in a given function, it is not touched. This is a pass most
1224 useful for debugging via bugpoint.
1228 <!-------------------------------------------------------------------------- -->
1229 <div class="doc_subsection">
1230 <a name="loop-extract-single">Extract at most one loop into a new function</a>
1232 <div class="doc_text">
1234 Similar to <a href="#loop-extract">Extract loops into new functions</a>,
1235 this pass extracts one natural loop from the program into a function if it
1236 can. This is used by bugpoint.
1240 <!-------------------------------------------------------------------------- -->
1241 <div class="doc_subsection">
1242 <a name="loop-index-split">Index Split Loops</a>
1244 <div class="doc_text">
1246 This pass divides loop's iteration range by spliting loop such that each
1247 individual loop is executed efficiently.
1251 <!-------------------------------------------------------------------------- -->
1252 <div class="doc_subsection">
1253 <a name="loop-reduce">Loop Strength Reduction</a>
1255 <div class="doc_text">
1257 This pass performs a strength reduction on array references inside loops that
1258 have as one or more of their components the loop induction variable. This is
1259 accomplished by creating a new value to hold the initial value of the array
1260 access for the first iteration, and then creating a new GEP instruction in
1261 the loop to increment the value by the appropriate amount.
1265 <!-------------------------------------------------------------------------- -->
1266 <div class="doc_subsection">
1267 <a name="loop-rotate">Rotate Loops</a>
1269 <div class="doc_text">
1270 <p>A simple loop rotation transformation.</p>
1273 <!-------------------------------------------------------------------------- -->
1274 <div class="doc_subsection">
1275 <a name="loop-unroll">Unroll loops</a>
1277 <div class="doc_text">
1279 This pass implements a simple loop unroller. It works best when loops have
1280 been canonicalized by the <a href="#indvars"><tt>-indvars</tt></a> pass,
1281 allowing it to determine the trip counts of loops easily.
1285 <!-------------------------------------------------------------------------- -->
1286 <div class="doc_subsection">
1287 <a name="loop-unswitch">Unswitch loops</a>
1289 <div class="doc_text">
1291 This pass transforms loops that contain branches on loop-invariant conditions
1292 to have multiple loops. For example, it turns the left into the right code:
1304 This can increase the size of the code exponentially (doubling it every time
1305 a loop is unswitched) so we only unswitch if the resultant code will be
1306 smaller than a threshold.
1310 This pass expects LICM to be run before it to hoist invariant conditions out
1311 of the loop, to make the unswitching opportunity obvious.
1315 <!-------------------------------------------------------------------------- -->
1316 <div class="doc_subsection">
1317 <a name="loopsimplify">Canonicalize natural loops</a>
1319 <div class="doc_text">
1321 This pass performs several transformations to transform natural loops into a
1322 simpler form, which makes subsequent analyses and transformations simpler and
1327 Loop pre-header insertion guarantees that there is a single, non-critical
1328 entry edge from outside of the loop to the loop header. This simplifies a
1329 number of analyses and transformations, such as LICM.
1333 Loop exit-block insertion guarantees that all exit blocks from the loop
1334 (blocks which are outside of the loop that have predecessors inside of the
1335 loop) only have predecessors from inside of the loop (and are thus dominated
1336 by the loop header). This simplifies transformations such as store-sinking
1337 that are built into LICM.
1341 This pass also guarantees that loops will have exactly one backedge.
1345 Note that the simplifycfg pass will clean up blocks which are split out but
1346 end up being unnecessary, so usage of this pass should not pessimize
1351 This pass obviously modifies the CFG, but updates loop information and
1352 dominator information.
1356 <!-------------------------------------------------------------------------- -->
1357 <div class="doc_subsection">
1358 <a name="lowerallocs">Lower allocations from instructions to calls</a>
1360 <div class="doc_text">
1362 Turn <tt>malloc</tt> and <tt>free</tt> instructions into <tt>@malloc</tt> and
1363 <tt>@free</tt> calls.
1367 This is a target-dependent tranformation because it depends on the size of
1368 data types and alignment constraints.
1372 <!-------------------------------------------------------------------------- -->
1373 <div class="doc_subsection">
1374 <a name="lowerinvoke">Lower invoke and unwind, for unwindless code generators</a>
1376 <div class="doc_text">
1378 This transformation is designed for use by code generators which do not yet
1379 support stack unwinding. This pass supports two models of exception handling
1380 lowering, the 'cheap' support and the 'expensive' support.
1384 'Cheap' exception handling support gives the program the ability to execute
1385 any program which does not "throw an exception", by turning 'invoke'
1386 instructions into calls and by turning 'unwind' instructions into calls to
1387 abort(). If the program does dynamically use the unwind instruction, the
1388 program will print a message then abort.
1392 'Expensive' exception handling support gives the full exception handling
1393 support to the program at the cost of making the 'invoke' instruction
1394 really expensive. It basically inserts setjmp/longjmp calls to emulate the
1395 exception handling as necessary.
1399 Because the 'expensive' support slows down programs a lot, and EH is only
1400 used for a subset of the programs, it must be specifically enabled by the
1401 <tt>-enable-correct-eh-support</tt> option.
1405 Note that after this pass runs the CFG is not entirely accurate (exceptional
1406 control flow edges are not correct anymore) so only very simple things should
1407 be done after the lowerinvoke pass has run (like generation of native code).
1408 This should not be used as a general purpose "my LLVM-to-LLVM pass doesn't
1409 support the invoke instruction yet" lowering pass.
1413 <!-------------------------------------------------------------------------- -->
1414 <div class="doc_subsection">
1415 <a name="lowersetjmp">Lower Set Jump</a>
1417 <div class="doc_text">
1419 Lowers <tt>setjmp</tt> and <tt>longjmp</tt> to use the LLVM invoke and unwind
1420 instructions as necessary.
1424 Lowering of <tt>longjmp</tt> is fairly trivial. We replace the call with a
1425 call to the LLVM library function <tt>__llvm_sjljeh_throw_longjmp()</tt>.
1426 This unwinds the stack for us calling all of the destructors for
1427 objects allocated on the stack.
1431 At a <tt>setjmp</tt> call, the basic block is split and the <tt>setjmp</tt>
1432 removed. The calls in a function that have a <tt>setjmp</tt> are converted to
1433 invoke where the except part checks to see if it's a <tt>longjmp</tt>
1434 exception and, if so, if it's handled in the function. If it is, then it gets
1435 the value returned by the <tt>longjmp</tt> and goes to where the basic block
1436 was split. <tt>invoke</tt> instructions are handled in a similar fashion with
1437 the original except block being executed if it isn't a <tt>longjmp</tt>
1438 except that is handled by that function.
1442 <!-------------------------------------------------------------------------- -->
1443 <div class="doc_subsection">
1444 <a name="lowerswitch">Lower SwitchInst's to branches</a>
1446 <div class="doc_text">
1448 Rewrites <tt>switch</tt> instructions with a sequence of branches, which
1449 allows targets to get away with not implementing the switch instruction until
1454 <!-------------------------------------------------------------------------- -->
1455 <div class="doc_subsection">
1456 <a name="mem2reg">Promote Memory to Register</a>
1458 <div class="doc_text">
1460 This file promotes memory references to be register references. It promotes
1461 <tt>alloca</tt> instructions which only have <tt>load</tt>s and
1462 <tt>store</tt>s as uses. An <tt>alloca</tt> is transformed by using dominator
1463 frontiers to place <tt>phi</tt> nodes, then traversing the function in
1464 depth-first order to rewrite <tt>load</tt>s and <tt>store</tt>s as
1465 appropriate. This is just the standard SSA construction algorithm to construct
1470 <!-------------------------------------------------------------------------- -->
1471 <div class="doc_subsection">
1472 <a name="memcpyopt">Optimize use of memcpy and friend</a>
1474 <div class="doc_text">
1476 This pass performs various transformations related to eliminating memcpy
1477 calls, or transforming sets of stores into memset's.
1481 <!-------------------------------------------------------------------------- -->
1482 <div class="doc_subsection">
1483 <a name="mergereturn">Unify function exit nodes</a>
1485 <div class="doc_text">
1487 Ensure that functions have at most one <tt>ret</tt> instruction in them.
1488 Additionally, it keeps track of which node is the new exit node of the CFG.
1492 <!-------------------------------------------------------------------------- -->
1493 <div class="doc_subsection">
1494 <a name="prune-eh">Remove unused exception handling info</a>
1496 <div class="doc_text">
1498 This file implements a simple interprocedural pass which walks the call-graph,
1499 turning <tt>invoke</tt> instructions into <tt>call</tt> instructions if and
1500 only if the callee cannot throw an exception. It implements this as a
1501 bottom-up traversal of the call-graph.
1505 <!-------------------------------------------------------------------------- -->
1506 <div class="doc_subsection">
1507 <a name="raiseallocs">Raise allocations from calls to instructions</a>
1509 <div class="doc_text">
1511 Converts <tt>@malloc</tt> and <tt>@free</tt> calls to <tt>malloc</tt> and
1512 <tt>free</tt> instructions.
1516 <!-------------------------------------------------------------------------- -->
1517 <div class="doc_subsection">
1518 <a name="reassociate">Reassociate expressions</a>
1520 <div class="doc_text">
1522 This pass reassociates commutative expressions in an order that is designed
1523 to promote better constant propagation, GCSE, LICM, PRE, etc.
1527 For example: 4 + (<var>x</var> + 5) ⇒ <var>x</var> + (4 + 5)
1531 In the implementation of this algorithm, constants are assigned rank = 0,
1532 function arguments are rank = 1, and other values are assigned ranks
1533 corresponding to the reverse post order traversal of current function
1534 (starting at 2), which effectively gives values in deep loops higher rank
1535 than values not in loops.
1539 <!-------------------------------------------------------------------------- -->
1540 <div class="doc_subsection">
1541 <a name="reg2mem">Demote all values to stack slots</a>
1543 <div class="doc_text">
1545 This file demotes all registers to memory references. It is intented to be
1546 the inverse of <a href="#mem2reg"><tt>-mem2reg</tt></a>. By converting to
1547 <tt>load</tt> instructions, the only values live across basic blocks are
1548 <tt>alloca</tt> instructions and <tt>load</tt> instructions before
1549 <tt>phi</tt> nodes. It is intended that this should make CFG hacking much
1550 easier. To make later hacking easier, the entry block is split into two, such
1551 that all introduced <tt>alloca</tt> instructions (and nothing else) are in the
1556 <!-------------------------------------------------------------------------- -->
1557 <div class="doc_subsection">
1558 <a name="scalarrepl">Scalar Replacement of Aggregates</a>
1560 <div class="doc_text">
1562 The well-known scalar replacement of aggregates transformation. This
1563 transform breaks up <tt>alloca</tt> instructions of aggregate type (structure
1564 or array) into individual <tt>alloca</tt> instructions for each member if
1565 possible. Then, if possible, it transforms the individual <tt>alloca</tt>
1566 instructions into nice clean scalar SSA form.
1570 This combines a simple scalar replacement of aggregates algorithm with the <a
1571 href="#mem2reg"><tt>mem2reg</tt></a> algorithm because often interact,
1572 especially for C++ programs. As such, iterating between <tt>scalarrepl</tt>,
1573 then <a href="#mem2reg"><tt>mem2reg</tt></a> until we run out of things to
1578 <!-------------------------------------------------------------------------- -->
1579 <div class="doc_subsection">
1580 <a name="sccp">Sparse Conditional Constant Propagation</a>
1582 <div class="doc_text">
1584 Sparse conditional constant propagation and merging, which can be summarized
1589 <li>Assumes values are constant unless proven otherwise</li>
1590 <li>Assumes BasicBlocks are dead unless proven otherwise</li>
1591 <li>Proves values to be constant, and replaces them with constants</li>
1592 <li>Proves conditional branches to be unconditional</li>
1596 Note that this pass has a habit of making definitions be dead. It is a good
1597 idea to to run a DCE pass sometime after running this pass.
1601 <!-------------------------------------------------------------------------- -->
1602 <div class="doc_subsection">
1603 <a name="simplify-libcalls">Simplify well-known library calls</a>
1605 <div class="doc_text">
1607 Applies a variety of small optimizations for calls to specific well-known
1608 function calls (e.g. runtime library functions). For example, a call
1609 <tt>exit(3)</tt> that occurs within the <tt>main()</tt> function can be
1610 transformed into simply <tt>return 3</tt>.
1614 <!-------------------------------------------------------------------------- -->
1615 <div class="doc_subsection">
1616 <a name="simplifycfg">Simplify the CFG</a>
1618 <div class="doc_text">
1620 Performs dead code elimination and basic block merging. Specifically:
1624 <li>Removes basic blocks with no predecessors.</li>
1625 <li>Merges a basic block into its predecessor if there is only one and the
1626 predecessor only has one successor.</li>
1627 <li>Eliminates PHI nodes for basic blocks with a single predecessor.</li>
1628 <li>Eliminates a basic block that only contains an unconditional
1633 <!-------------------------------------------------------------------------- -->
1634 <div class="doc_subsection">
1635 <a name="strip">Strip all symbols from a module</a>
1637 <div class="doc_text">
1639 Performs code stripping. This transformation can delete:
1643 <li>names for virtual registers</li>
1644 <li>symbols for internal globals and functions</li>
1645 <li>debug information</li>
1649 Note that this transformation makes code much less readable, so it should
1650 only be used in situations where the <tt>strip</tt> utility would be used,
1651 such as reducing code size or making it harder to reverse engineer code.
1655 <!-------------------------------------------------------------------------- -->
1656 <div class="doc_subsection">
1657 <a name="strip-dead-prototypes">Remove unused function declarations</a>
1659 <div class="doc_text">
1661 This pass loops over all of the functions in the input module, looking for
1662 dead declarations and removes them. Dead declarations are declarations of
1663 functions for which no implementation is available (i.e., declarations for
1664 unused library functions).
1668 <!-------------------------------------------------------------------------- -->
1669 <div class="doc_subsection">
1670 <a name="sretpromotion">Promote sret arguments</a>
1672 <div class="doc_text">
1674 This pass finds functions that return a struct (using a pointer to the struct
1675 as the first argument of the function, marked with the '<tt>sret</tt>' attribute) and
1676 replaces them with a new function that simply returns each of the elements of
1677 that struct (using multiple return values).
1681 This pass works under a number of conditions:
1685 <li>The returned struct must not contain other structs</li>
1686 <li>The returned struct must only be used to load values from</li>
1687 <li>The placeholder struct passed in is the result of an <tt>alloca</tt></li>
1691 <!-------------------------------------------------------------------------- -->
1692 <div class="doc_subsection">
1693 <a name="tailcallelim">Tail Call Elimination</a>
1695 <div class="doc_text">
1697 This file transforms calls of the current function (self recursion) followed
1698 by a return instruction with a branch to the entry of the function, creating
1699 a loop. This pass also implements the following extensions to the basic
1704 <li>Trivial instructions between the call and return do not prevent the
1705 transformation from taking place, though currently the analysis cannot
1706 support moving any really useful instructions (only dead ones).
1707 <li>This pass transforms functions that are prevented from being tail
1708 recursive by an associative expression to use an accumulator variable,
1709 thus compiling the typical naive factorial or <tt>fib</tt> implementation
1710 into efficient code.
1711 <li>TRE is performed if the function returns void, if the return
1712 returns the result returned by the call, or if the function returns a
1713 run-time constant on all exits from the function. It is possible, though
1714 unlikely, that the return returns something else (like constant 0), and
1715 can still be TRE'd. It can be TRE'd if <em>all other</em> return
1716 instructions in the function return the exact same value.
1717 <li>If it can prove that callees do not access theier caller stack frame,
1718 they are marked as eligible for tail call elimination (by the code
1723 <!-------------------------------------------------------------------------- -->
1724 <div class="doc_subsection">
1725 <a name="tailduplicate">Tail Duplication</a>
1727 <div class="doc_text">
1729 This pass performs a limited form of tail duplication, intended to simplify
1730 CFGs by removing some unconditional branches. This pass is necessary to
1731 straighten out loops created by the C front-end, but also is capable of
1732 making other code nicer. After this pass is run, the CFG simplify pass
1733 should be run to clean up the mess.
1737 <!-- ======================================================================= -->
1738 <div class="doc_section"> <a name="transform">Utility Passes</a></div>
1739 <div class="doc_text">
1740 <p>This section describes the LLVM Utility Passes.</p>
1743 <!-------------------------------------------------------------------------- -->
1744 <div class="doc_subsection">
1745 <a name="deadarghaX0r">Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</a>
1747 <div class="doc_text">
1749 Same as dead argument elimination, but deletes arguments to functions which
1750 are external. This is only for use by <a
1751 href="Bugpoint.html">bugpoint</a>.</p>
1754 <!-------------------------------------------------------------------------- -->
1755 <div class="doc_subsection">
1756 <a name="extract-blocks">Extract Basic Blocks From Module (for bugpoint use)</a>
1758 <div class="doc_text">
1760 This pass is used by bugpoint to extract all blocks from the module into their
1764 <!-------------------------------------------------------------------------- -->
1765 <div class="doc_subsection">
1766 <a name="preverify">Preliminary module verification</a>
1768 <div class="doc_text">
1770 Ensures that the module is in the form required by the <a
1771 href="#verifier">Module Verifier</a> pass.
1775 Running the verifier runs this pass automatically, so there should be no need
1780 <!-------------------------------------------------------------------------- -->
1781 <div class="doc_subsection">
1782 <a name="verify">Module Verifier</a>
1784 <div class="doc_text">
1786 Verifies an LLVM IR code. This is useful to run after an optimization which is
1787 undergoing testing. Note that <tt>llvm-as</tt> verifies its input before
1788 emitting bitcode, and also that malformed bitcode is likely to make LLVM
1789 crash. All language front-ends are therefore encouraged to verify their output
1790 before performing optimizing transformations.
1794 <li>Both of a binary operator's parameters are of the same type.</li>
1795 <li>Verify that the indices of mem access instructions match other
1797 <li>Verify that arithmetic and other things are only performed on
1798 first-class types. Verify that shifts and logicals only happen on
1800 <li>All of the constants in a switch statement are of the correct type.</li>
1801 <li>The code is in valid SSA form.</li>
1802 <li>It should be illegal to put a label into any other type (like a
1803 structure) or to return one. [except constant arrays!]</li>
1804 <li>Only phi nodes can be self referential: <tt>%x = add i32 %x, %x</tt> is
1806 <li>PHI nodes must have an entry for each predecessor, with no extras.</li>
1807 <li>PHI nodes must be the first thing in a basic block, all grouped
1809 <li>PHI nodes must have at least one entry.</li>
1810 <li>All basic blocks should only end with terminator insts, not contain
1812 <li>The entry node to a function must not have predecessors.</li>
1813 <li>All Instructions must be embedded into a basic block.</li>
1814 <li>Functions cannot take a void-typed parameter.</li>
1815 <li>Verify that a function's argument list agrees with its declared
1817 <li>It is illegal to specify a name for a void value.</li>
1818 <li>It is illegal to have a internal global value with no initializer.</li>
1819 <li>It is illegal to have a ret instruction that returns a value that does
1820 not agree with the function return value type.</li>
1821 <li>Function call argument types match the function prototype.</li>
1822 <li>All other things that are tested by asserts spread about the code.</li>
1826 Note that this does not provide full security verification (like Java), but
1827 instead just tries to ensure that code is well-formed.
1831 <!-------------------------------------------------------------------------- -->
1832 <div class="doc_subsection">
1833 <a name="view-cfg">View CFG of function</a>
1835 <div class="doc_text">
1837 Displays the control flow graph using the GraphViz tool.
1841 <!-------------------------------------------------------------------------- -->
1842 <div class="doc_subsection">
1843 <a name="view-cfg-only">View CFG of function (with no function bodies)</a>
1845 <div class="doc_text">
1847 Displays the control flow graph using the GraphViz tool, but omitting function
1852 <!-- *********************************************************************** -->
1856 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1857 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1858 <a href="http://validator.w3.org/check/referer"><img
1859 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1861 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a><br>
1862 <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
1863 Last modified: $Date$