1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM's Analysis and Transform Passes</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
7 <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
13 If Passes.html is up to date, the following "one-liner" should print
16 egrep -e '^<tr><td><a href="#.*">-.*</a></td><td>.*</td></tr>$' \
17 -e '^ <a name=".*">.*</a>$' < Passes.html >html; \
18 perl >help <<'EOT' && diff -u help html; rm -f help html
19 open HTML, "<Passes.html" or die "open: Passes.html: $!\n";
21 m:^<tr><td><a href="#(.*)">-.*</a></td><td>.*</td></tr>$: or next;
22 $order{$1} = sprintf("%03d", 1 + int %order);
24 open HELP, "../Release/bin/opt -help|" or die "open: opt -help: $!\n";
26 m:^ -([^ ]+) +- (.*)$: or next;
28 $o = "000" unless defined $o;
29 push @x, "$o<tr><td><a href=\"#$1\">-$1</a></td><td>$2</td></tr>\n";
30 push @y, "$o <a name=\"$1\">$2</a>\n";
32 @x = map { s/^\d\d\d//; $_ } sort @x;
33 @y = map { s/^\d\d\d//; $_ } sort @y;
37 This (real) one-liner can also be helpful when converting comments to HTML:
39 perl -e '$/ = undef; for (split(/\n/, <>)) { s:^ *///? ?::; print " <p>\n" if !$on && $_ =~ /\S/; print " </p>\n" if $on && $_ =~ /^\s*$/; print " $_\n"; $on = ($_ =~ /\S/); } print " </p>\n" if $on'
43 <div class="doc_title">LLVM's Analysis and Transform Passes</div>
46 <li><a href="#intro">Introduction</a></li>
47 <li><a href="#analyses">Analysis Passes</a>
48 <li><a href="#transforms">Transform Passes</a></li>
49 <li><a href="#utilities">Utility Passes</a></li>
52 <div class="doc_author">
53 <p>Written by <a href="mailto:rspencer@x10sys.com">Reid Spencer</a>
54 and Gordon Henriksen</p>
57 <!-- ======================================================================= -->
58 <div class="doc_section"> <a name="intro">Introduction</a> </div>
59 <div class="doc_text">
60 <p>This document serves as a high level summary of the optimization features
61 that LLVM provides. Optimizations are implemented as Passes that traverse some
62 portion of a program to either collect information or transform the program.
63 The table below divides the passes that LLVM provides into three categories.
64 Analysis passes compute information that other passes can use or for debugging
65 or program visualization purposes. Transform passes can use (or invalidate)
66 the analysis passes. Transform passes all mutate the program in some way.
67 Utility passes provides some utility but don't otherwise fit categorization.
68 For example passes to extract functions to bitcode or write a module to
69 bitcode are neither analysis nor transform passes.
70 <p>The table below provides a quick summary of each pass and links to the more
71 complete pass description later in the document.</p>
73 <div class="doc_text" >
75 <tr><th colspan="2"><b>ANALYSIS PASSES</b></th></tr>
76 <tr><th>Option</th><th>Name</th></tr>
77 <tr><td><a href="#aa-eval">-aa-eval</a></td><td>Exhaustive Alias Analysis Precision Evaluator</td></tr>
78 <tr><td><a href="#basicaa">-basicaa</a></td><td>Basic Alias Analysis (default AA impl)</td></tr>
79 <tr><td><a href="#basiccg">-basiccg</a></td><td>Basic CallGraph Construction</td></tr>
80 <tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Optimize for code generation</td></tr>
81 <tr><td><a href="#count-aa">-count-aa</a></td><td>Count Alias Analysis Query Responses</td></tr>
82 <tr><td><a href="#debug-aa">-debug-aa</a></td><td>AA use debugger</td></tr>
83 <tr><td><a href="#domfrontier">-domfrontier</a></td><td>Dominance Frontier Construction</td></tr>
84 <tr><td><a href="#domtree">-domtree</a></td><td>Dominator Tree Construction</td></tr>
85 <tr><td><a href="#dot-callgraph">-dot-callgraph</a></td><td>Print Call Graph to 'dot' file</td></tr>
86 <tr><td><a href="#dot-cfg">-dot-cfg</a></td><td>Print CFG of function to 'dot' file</td></tr>
87 <tr><td><a href="#dot-cfg-only">-dot-cfg-only</a></td><td>Print CFG of function to 'dot' file (with no function bodies)</td></tr>
88 <tr><td><a href="#globalsmodref-aa">-globalsmodref-aa</a></td><td>Simple mod/ref analysis for globals</td></tr>
89 <tr><td><a href="#instcount">-instcount</a></td><td>Counts the various types of Instructions</td></tr>
90 <tr><td><a href="#intervals">-intervals</a></td><td>Interval Partition Construction</td></tr>
91 <tr><td><a href="#loops">-loops</a></td><td>Natural Loop Construction</td></tr>
92 <tr><td><a href="#memdep">-memdep</a></td><td>Memory Dependence Analysis</td></tr>
93 <tr><td><a href="#no-aa">-no-aa</a></td><td>No Alias Analysis (always returns 'may' alias)</td></tr>
94 <tr><td><a href="#no-profile">-no-profile</a></td><td>No Profile Information</td></tr>
95 <tr><td><a href="#postdomfrontier">-postdomfrontier</a></td><td>Post-Dominance Frontier Construction</td></tr>
96 <tr><td><a href="#postdomtree">-postdomtree</a></td><td>Post-Dominator Tree Construction</td></tr>
97 <tr><td><a href="#print-alias-sets">-print-alias-sets</a></td><td>Alias Set Printer</td></tr>
98 <tr><td><a href="#print-callgraph">-print-callgraph</a></td><td>Print a call graph</td></tr>
99 <tr><td><a href="#print-callgraph-sccs">-print-callgraph-sccs</a></td><td>Print SCCs of the Call Graph</td></tr>
100 <tr><td><a href="#print-cfg-sccs">-print-cfg-sccs</a></td><td>Print SCCs of each function CFG</td></tr>
101 <tr><td><a href="#print-externalfnconstants">-print-externalfnconstants</a></td><td>Print external fn callsites passed constants</td></tr>
102 <tr><td><a href="#print-function">-print-function</a></td><td>Print function to stderr</td></tr>
103 <tr><td><a href="#print-module">-print-module</a></td><td>Print module to stderr</td></tr>
104 <tr><td><a href="#print-used-types">-print-used-types</a></td><td>Find Used Types</td></tr>
105 <tr><td><a href="#profile-loader">-profile-loader</a></td><td>Load profile information from llvmprof.out</td></tr>
106 <tr><td><a href="#scalar-evolution">-scalar-evolution</a></td><td>Scalar Evolution Analysis</td></tr>
107 <tr><td><a href="#targetdata">-targetdata</a></td><td>Target Data Layout</td></tr>
110 <tr><th colspan="2"><b>TRANSFORM PASSES</b></th></tr>
111 <tr><th>Option</th><th>Name</th></tr>
112 <tr><td><a href="#adce">-adce</a></td><td>Aggressive Dead Code Elimination</td></tr>
113 <tr><td><a href="#argpromotion">-argpromotion</a></td><td>Promote 'by reference' arguments to scalars</td></tr>
114 <tr><td><a href="#block-placement">-block-placement</a></td><td>Profile Guided Basic Block Placement</td></tr>
115 <tr><td><a href="#break-crit-edges">-break-crit-edges</a></td><td>Break critical edges in CFG</td></tr>
116 <tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Prepare a function for code generation </td></tr>
117 <tr><td><a href="#condprop">-condprop</a></td><td>Conditional Propagation</td></tr>
118 <tr><td><a href="#constmerge">-constmerge</a></td><td>Merge Duplicate Global Constants</td></tr>
119 <tr><td><a href="#constprop">-constprop</a></td><td>Simple constant propagation</td></tr>
120 <tr><td><a href="#dce">-dce</a></td><td>Dead Code Elimination</td></tr>
121 <tr><td><a href="#deadargelim">-deadargelim</a></td><td>Dead Argument Elimination</td></tr>
122 <tr><td><a href="#deadtypeelim">-deadtypeelim</a></td><td>Dead Type Elimination</td></tr>
123 <tr><td><a href="#die">-die</a></td><td>Dead Instruction Elimination</td></tr>
124 <tr><td><a href="#dse">-dse</a></td><td>Dead Store Elimination</td></tr>
125 <tr><td><a href="#globaldce">-globaldce</a></td><td>Dead Global Elimination</td></tr>
126 <tr><td><a href="#globalopt">-globalopt</a></td><td>Global Variable Optimizer</td></tr>
127 <tr><td><a href="#gvn">-gvn</a></td><td>Global Value Numbering</td></tr>
128 <tr><td><a href="#indmemrem">-indmemrem</a></td><td>Indirect Malloc and Free Removal</td></tr>
129 <tr><td><a href="#indvars">-indvars</a></td><td>Canonicalize Induction Variables</td></tr>
130 <tr><td><a href="#inline">-inline</a></td><td>Function Integration/Inlining</td></tr>
131 <tr><td><a href="#insert-block-profiling">-insert-block-profiling</a></td><td>Insert instrumentation for block profiling</td></tr>
132 <tr><td><a href="#insert-edge-profiling">-insert-edge-profiling</a></td><td>Insert instrumentation for edge profiling</td></tr>
133 <tr><td><a href="#insert-function-profiling">-insert-function-profiling</a></td><td>Insert instrumentation for function profiling</td></tr>
134 <tr><td><a href="#insert-null-profiling-rs">-insert-null-profiling-rs</a></td><td>Measure profiling framework overhead</td></tr>
135 <tr><td><a href="#insert-rs-profiling-framework">-insert-rs-profiling-framework</a></td><td>Insert random sampling instrumentation framework</td></tr>
136 <tr><td><a href="#instcombine">-instcombine</a></td><td>Combine redundant instructions</td></tr>
137 <tr><td><a href="#internalize">-internalize</a></td><td>Internalize Global Symbols</td></tr>
138 <tr><td><a href="#ipconstprop">-ipconstprop</a></td><td>Interprocedural constant propagation</td></tr>
139 <tr><td><a href="#ipsccp">-ipsccp</a></td><td>Interprocedural Sparse Conditional Constant Propagation</td></tr>
140 <tr><td><a href="#jump-threading">-jump-threading</a></td><td>Thread control through conditional blocks </td></tr>
141 <tr><td><a href="#lcssa">-lcssa</a></td><td>Loop-Closed SSA Form Pass</td></tr>
142 <tr><td><a href="#licm">-licm</a></td><td>Loop Invariant Code Motion</td></tr>
143 <tr><td><a href="#loop-deletion">-loop-deletion</a></td><td>Dead Loop Deletion Pass </td></tr>
144 <tr><td><a href="#loop-extract">-loop-extract</a></td><td>Extract loops into new functions</td></tr>
145 <tr><td><a href="#loop-extract-single">-loop-extract-single</a></td><td>Extract at most one loop into a new function</td></tr>
146 <tr><td><a href="#loop-index-split">-loop-index-split</a></td><td>Index Split Loops</td></tr>
147 <tr><td><a href="#loop-reduce">-loop-reduce</a></td><td>Loop Strength Reduction</td></tr>
148 <tr><td><a href="#loop-rotate">-loop-rotate</a></td><td>Rotate Loops</td></tr>
149 <tr><td><a href="#loop-unroll">-loop-unroll</a></td><td>Unroll loops</td></tr>
150 <tr><td><a href="#loop-unswitch">-loop-unswitch</a></td><td>Unswitch loops</td></tr>
151 <tr><td><a href="#loopsimplify">-loopsimplify</a></td><td>Canonicalize natural loops</td></tr>
152 <tr><td><a href="#lowerallocs">-lowerallocs</a></td><td>Lower allocations from instructions to calls</td></tr>
153 <tr><td><a href="#lowerinvoke">-lowerinvoke</a></td><td>Lower invoke and unwind, for unwindless code generators</td></tr>
154 <tr><td><a href="#lowersetjmp">-lowersetjmp</a></td><td>Lower Set Jump</td></tr>
155 <tr><td><a href="#lowerswitch">-lowerswitch</a></td><td>Lower SwitchInst's to branches</td></tr>
156 <tr><td><a href="#mem2reg">-mem2reg</a></td><td>Promote Memory to Register</td></tr>
157 <tr><td><a href="#memcpyopt">-memcpyopt</a></td><td>Optimize use of memcpy and friends</td></tr>
158 <tr><td><a href="#mergereturn">-mergereturn</a></td><td>Unify function exit nodes</td></tr>
159 <tr><td><a href="#prune-eh">-prune-eh</a></td><td>Remove unused exception handling info</td></tr>
160 <tr><td><a href="#reassociate">-reassociate</a></td><td>Reassociate expressions</td></tr>
161 <tr><td><a href="#reg2mem">-reg2mem</a></td><td>Demote all values to stack slots</td></tr>
162 <tr><td><a href="#scalarrepl">-scalarrepl</a></td><td>Scalar Replacement of Aggregates</td></tr>
163 <tr><td><a href="#sccp">-sccp</a></td><td>Sparse Conditional Constant Propagation</td></tr>
164 <tr><td><a href="#simplify-libcalls">-simplify-libcalls</a></td><td>Simplify well-known library calls</td></tr>
165 <tr><td><a href="#simplifycfg">-simplifycfg</a></td><td>Simplify the CFG</td></tr>
166 <tr><td><a href="#strip">-strip</a></td><td>Strip all symbols from a module</td></tr>
167 <tr><td><a href="#strip-dead-prototypes">-strip-dead-prototypes</a></td><td>Remove unused function declarations</td></tr>
168 <tr><td><a href="#sretpromotion">-sretpromotion</a></td><td>Promote sret arguments</td></tr>
169 <tr><td><a href="#tailcallelim">-tailcallelim</a></td><td>Tail Call Elimination</td></tr>
170 <tr><td><a href="#tailduplicate">-tailduplicate</a></td><td>Tail Duplication</td></tr>
173 <tr><th colspan="2"><b>UTILITY PASSES</b></th></tr>
174 <tr><th>Option</th><th>Name</th></tr>
175 <tr><td><a href="#deadarghaX0r">-deadarghaX0r</a></td><td>Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</td></tr>
176 <tr><td><a href="#extract-blocks">-extract-blocks</a></td><td>Extract Basic Blocks From Module (for bugpoint use)</td></tr>
177 <tr><td><a href="#preverify">-preverify</a></td><td>Preliminary module verification</td></tr>
178 <tr><td><a href="#verify">-verify</a></td><td>Module Verifier</td></tr>
179 <tr><td><a href="#view-cfg">-view-cfg</a></td><td>View CFG of function</td></tr>
180 <tr><td><a href="#view-cfg-only">-view-cfg-only</a></td><td>View CFG of function (with no function bodies)</td></tr>
184 <!-- ======================================================================= -->
185 <div class="doc_section"> <a name="example">Analysis Passes</a></div>
186 <div class="doc_text">
187 <p>This section describes the LLVM Analysis Passes.</p>
190 <!-------------------------------------------------------------------------- -->
191 <div class="doc_subsection">
192 <a name="aa-eval">Exhaustive Alias Analysis Precision Evaluator</a>
194 <div class="doc_text">
195 <p>This is a simple N^2 alias analysis accuracy evaluator.
196 Basically, for each function in the program, it simply queries to see how the
197 alias analysis implementation answers alias queries between each pair of
198 pointers in the function.</p>
200 <p>This is inspired and adapted from code by: Naveen Neelakantam, Francesco
201 Spadini, and Wojciech Stryjewski.</p>
204 <!-------------------------------------------------------------------------- -->
205 <div class="doc_subsection">
206 <a name="basicaa">Basic Alias Analysis (default AA impl)</a>
208 <div class="doc_text">
210 This is the default implementation of the Alias Analysis interface
211 that simply implements a few identities (two different globals cannot alias,
212 etc), but otherwise does no analysis.
216 <!-------------------------------------------------------------------------- -->
217 <div class="doc_subsection">
218 <a name="basiccg">Basic CallGraph Construction</a>
220 <div class="doc_text">
221 <p>Yet to be written.</p>
224 <!-------------------------------------------------------------------------- -->
225 <div class="doc_subsection">
226 <a name="codegenprepare">Optimize for code generation</a>
228 <div class="doc_text">
230 This pass munges the code in the input function to better prepare it for
231 SelectionDAG-based code generation. This works around limitations in it's
232 basic-block-at-a-time approach. It should eventually be removed.
236 <!-------------------------------------------------------------------------- -->
237 <div class="doc_subsection">
238 <a name="count-aa">Count Alias Analysis Query Responses</a>
240 <div class="doc_text">
242 A pass which can be used to count how many alias queries
243 are being made and how the alias analysis implementation being used responds.
247 <!-------------------------------------------------------------------------- -->
248 <div class="doc_subsection">
249 <a name="debug-aa">AA use debugger</a>
251 <div class="doc_text">
253 This simple pass checks alias analysis users to ensure that if they
254 create a new value, they do not query AA without informing it of the value.
255 It acts as a shim over any other AA pass you want.
259 Yes keeping track of every value in the program is expensive, but this is
264 <!-------------------------------------------------------------------------- -->
265 <div class="doc_subsection">
266 <a name="domfrontier">Dominance Frontier Construction</a>
268 <div class="doc_text">
270 This pass is a simple dominator construction algorithm for finding forward
275 <!-------------------------------------------------------------------------- -->
276 <div class="doc_subsection">
277 <a name="domtree">Dominator Tree Construction</a>
279 <div class="doc_text">
281 This pass is a simple dominator construction algorithm for finding forward
286 <!-------------------------------------------------------------------------- -->
287 <div class="doc_subsection">
288 <a name="dot-callgraph">Print Call Graph to 'dot' file</a>
290 <div class="doc_text">
292 This pass, only available in <code>opt</code>, prints the call graph into a
293 <code>.dot</code> graph. This graph can then be processed with the "dot" tool
294 to convert it to postscript or some other suitable format.
298 <!-------------------------------------------------------------------------- -->
299 <div class="doc_subsection">
300 <a name="dot-cfg">Print CFG of function to 'dot' file</a>
302 <div class="doc_text">
304 This pass, only available in <code>opt</code>, prints the control flow graph
305 into a <code>.dot</code> graph. This graph can then be processed with the
306 "dot" tool to convert it to postscript or some other suitable format.
310 <!-------------------------------------------------------------------------- -->
311 <div class="doc_subsection">
312 <a name="dot-cfg-only">Print CFG of function to 'dot' file (with no function bodies)</a>
314 <div class="doc_text">
316 This pass, only available in <code>opt</code>, prints the control flow graph
317 into a <code>.dot</code> graph, omitting the function bodies. This graph can
318 then be processed with the "dot" tool to convert it to postscript or some
319 other suitable format.
323 <!-------------------------------------------------------------------------- -->
324 <div class="doc_subsection">
325 <a name="globalsmodref-aa">Simple mod/ref analysis for globals</a>
327 <div class="doc_text">
329 This simple pass provides alias and mod/ref information for global values
330 that do not have their address taken, and keeps track of whether functions
331 read or write memory (are "pure"). For this simple (but very common) case,
332 we can provide pretty accurate and useful information.
336 <!-------------------------------------------------------------------------- -->
337 <div class="doc_subsection">
338 <a name="instcount">Counts the various types of Instructions</a>
340 <div class="doc_text">
342 This pass collects the count of all instructions and reports them
346 <!-------------------------------------------------------------------------- -->
347 <div class="doc_subsection">
348 <a name="intervals">Interval Partition Construction</a>
350 <div class="doc_text">
352 This analysis calculates and represents the interval partition of a function,
353 or a preexisting interval partition.
357 In this way, the interval partition may be used to reduce a flow graph down
358 to its degenerate single node interval partition (unless it is irreducible).
362 <!-------------------------------------------------------------------------- -->
363 <div class="doc_subsection">
364 <a name="loops">Natural Loop Construction</a>
366 <div class="doc_text">
368 This analysis is used to identify natural loops and determine the loop depth
369 of various nodes of the CFG. Note that the loops identified may actually be
370 several natural loops that share the same header node... not just a single
375 <!-------------------------------------------------------------------------- -->
376 <div class="doc_subsection">
377 <a name="memdep">Memory Dependence Analysis</a>
379 <div class="doc_text">
381 An analysis that determines, for a given memory operation, what preceding
382 memory operations it depends on. It builds on alias analysis information, and
383 tries to provide a lazy, caching interface to a common kind of alias
388 <!-------------------------------------------------------------------------- -->
389 <div class="doc_subsection">
390 <a name="no-aa">No Alias Analysis (always returns 'may' alias)</a>
392 <div class="doc_text">
394 Always returns "I don't know" for alias queries. NoAA is unlike other alias
395 analysis implementations, in that it does not chain to a previous analysis. As
396 such it doesn't follow many of the rules that other alias analyses must.
400 <!-------------------------------------------------------------------------- -->
401 <div class="doc_subsection">
402 <a name="no-profile">No Profile Information</a>
404 <div class="doc_text">
406 The default "no profile" implementation of the abstract
407 <code>ProfileInfo</code> interface.
411 <!-------------------------------------------------------------------------- -->
412 <div class="doc_subsection">
413 <a name="postdomfrontier">Post-Dominance Frontier Construction</a>
415 <div class="doc_text">
417 This pass is a simple post-dominator construction algorithm for finding
418 post-dominator frontiers.
422 <!-------------------------------------------------------------------------- -->
423 <div class="doc_subsection">
424 <a name="postdomtree">Post-Dominator Tree Construction</a>
426 <div class="doc_text">
428 This pass is a simple post-dominator construction algorithm for finding
433 <!-------------------------------------------------------------------------- -->
434 <div class="doc_subsection">
435 <a name="print-alias-sets">Alias Set Printer</a>
437 <div class="doc_text">
438 <p>Yet to be written.</p>
441 <!-------------------------------------------------------------------------- -->
442 <div class="doc_subsection">
443 <a name="print-callgraph">Print a call graph</a>
445 <div class="doc_text">
447 This pass, only available in <code>opt</code>, prints the call graph to
448 standard output in a human-readable form.
452 <!-------------------------------------------------------------------------- -->
453 <div class="doc_subsection">
454 <a name="print-callgraph-sccs">Print SCCs of the Call Graph</a>
456 <div class="doc_text">
458 This pass, only available in <code>opt</code>, prints the SCCs of the call
459 graph to standard output in a human-readable form.
463 <!-------------------------------------------------------------------------- -->
464 <div class="doc_subsection">
465 <a name="print-cfg-sccs">Print SCCs of each function CFG</a>
467 <div class="doc_text">
469 This pass, only available in <code>opt</code>, prints the SCCs of each
470 function CFG to standard output in a human-readable form.
474 <!-------------------------------------------------------------------------- -->
475 <div class="doc_subsection">
476 <a name="print-externalfnconstants">Print external fn callsites passed constants</a>
478 <div class="doc_text">
480 This pass, only available in <code>opt</code>, prints out call sites to
481 external functions that are called with constant arguments. This can be
482 useful when looking for standard library functions we should constant fold
483 or handle in alias analyses.
487 <!-------------------------------------------------------------------------- -->
488 <div class="doc_subsection">
489 <a name="print-function">Print function to stderr</a>
491 <div class="doc_text">
493 The <code>PrintFunctionPass</code> class is designed to be pipelined with
494 other <code>FunctionPass</code>es, and prints out the functions of the module
495 as they are processed.
499 <!-------------------------------------------------------------------------- -->
500 <div class="doc_subsection">
501 <a name="print-module">Print module to stderr</a>
503 <div class="doc_text">
505 This pass simply prints out the entire module when it is executed.
509 <!-------------------------------------------------------------------------- -->
510 <div class="doc_subsection">
511 <a name="print-used-types">Find Used Types</a>
513 <div class="doc_text">
515 This pass is used to seek out all of the types in use by the program. Note
516 that this analysis explicitly does not include types only used by the symbol
520 <!-------------------------------------------------------------------------- -->
521 <div class="doc_subsection">
522 <a name="profile-loader">Load profile information from llvmprof.out</a>
524 <div class="doc_text">
526 A concrete implementation of profiling information that loads the information
527 from a profile dump file.
531 <!-------------------------------------------------------------------------- -->
532 <div class="doc_subsection">
533 <a name="scalar-evolution">Scalar Evolution Analysis</a>
535 <div class="doc_text">
537 The <code>ScalarEvolution</code> analysis can be used to analyze and
538 catagorize scalar expressions in loops. It specializes in recognizing general
539 induction variables, representing them with the abstract and opaque
540 <code>SCEV</code> class. Given this analysis, trip counts of loops and other
541 important properties can be obtained.
545 This analysis is primarily useful for induction variable substitution and
550 <!-------------------------------------------------------------------------- -->
551 <div class="doc_subsection">
552 <a name="targetdata">Target Data Layout</a>
554 <div class="doc_text">
555 <p>Provides other passes access to information on how the size and alignment
556 required by the the target ABI for various data types.</p>
559 <!-- ======================================================================= -->
560 <div class="doc_section"> <a name="transform">Transform Passes</a></div>
561 <div class="doc_text">
562 <p>This section describes the LLVM Transform Passes.</p>
565 <!-------------------------------------------------------------------------- -->
566 <div class="doc_subsection">
567 <a name="adce">Aggressive Dead Code Elimination</a>
569 <div class="doc_text">
570 <p>ADCE aggressively tries to eliminate code. This pass is similar to
571 <a href="#dce">DCE</a> but it assumes that values are dead until proven
572 otherwise. This is similar to <a href="#sccp">SCCP</a>, except applied to
573 the liveness of values.</p>
576 <!-------------------------------------------------------------------------- -->
577 <div class="doc_subsection">
578 <a name="argpromotion">Promote 'by reference' arguments to scalars</a>
580 <div class="doc_text">
582 This pass promotes "by reference" arguments to be "by value" arguments. In
583 practice, this means looking for internal functions that have pointer
584 arguments. If it can prove, through the use of alias analysis, that an
585 argument is *only* loaded, then it can pass the value into the function
586 instead of the address of the value. This can cause recursive simplification
587 of code and lead to the elimination of allocas (especially in C++ template
592 This pass also handles aggregate arguments that are passed into a function,
593 scalarizing them if the elements of the aggregate are only loaded. Note that
594 it refuses to scalarize aggregates which would require passing in more than
595 three operands to the function, because passing thousands of operands for a
596 large array or structure is unprofitable!
600 Note that this transformation could also be done for arguments that are only
601 stored to (returning the value instead), but does not currently. This case
602 would be best handled when and if LLVM starts supporting multiple return
603 values from functions.
607 <!-------------------------------------------------------------------------- -->
608 <div class="doc_subsection">
609 <a name="block-placement">Profile Guided Basic Block Placement</a>
611 <div class="doc_text">
612 <p>This pass is a very simple profile guided basic block placement algorithm.
613 The idea is to put frequently executed blocks together at the start of the
614 function and hopefully increase the number of fall-through conditional
615 branches. If there is no profile information for a particular function, this
616 pass basically orders blocks in depth-first order.</p>
619 <!-------------------------------------------------------------------------- -->
620 <div class="doc_subsection">
621 <a name="break-crit-edges">Break critical edges in CFG</a>
623 <div class="doc_text">
625 Break all of the critical edges in the CFG by inserting a dummy basic block.
626 It may be "required" by passes that cannot deal with critical edges. This
627 transformation obviously invalidates the CFG, but can update forward dominator
628 (set, immediate dominators, tree, and frontier) information.
632 <!-------------------------------------------------------------------------- -->
633 <div class="doc_subsection">
634 <a name="codegenprepare">Prepare a function for code generation</a>
636 <div class="doc_text">
637 This pass munges the code in the input function to better prepare it for
638 SelectionDAG-based code generation. This works around limitations in it's
639 basic-block-at-a-time approach. It should eventually be removed.
642 <!-------------------------------------------------------------------------- -->
643 <div class="doc_subsection">
644 <a name="condprop">Conditional Propagation</a>
646 <div class="doc_text">
647 <p>This pass propagates information about conditional expressions through the
648 program, allowing it to eliminate conditional branches in some cases.</p>
651 <!-------------------------------------------------------------------------- -->
652 <div class="doc_subsection">
653 <a name="constmerge">Merge Duplicate Global Constants</a>
655 <div class="doc_text">
657 Merges duplicate global constants together into a single constant that is
658 shared. This is useful because some passes (ie TraceValues) insert a lot of
659 string constants into the program, regardless of whether or not an existing
664 <!-------------------------------------------------------------------------- -->
665 <div class="doc_subsection">
666 <a name="constprop">Simple constant propagation</a>
668 <div class="doc_text">
669 <p>This file implements constant propagation and merging. It looks for
670 instructions involving only constant operands and replaces them with a
671 constant value instead of an instruction. For example:</p>
672 <blockquote><pre>add i32 1, 2</pre></blockquote>
674 <blockquote><pre>i32 3</pre></blockquote>
675 <p>NOTE: this pass has a habit of making definitions be dead. It is a good
676 idea to to run a <a href="#die">DIE</a> (Dead Instruction Elimination) pass
677 sometime after running this pass.</p>
680 <!-------------------------------------------------------------------------- -->
681 <div class="doc_subsection">
682 <a name="dce">Dead Code Elimination</a>
684 <div class="doc_text">
686 Dead code elimination is similar to <a href="#die">dead instruction
687 elimination</a>, but it rechecks instructions that were used by removed
688 instructions to see if they are newly dead.
692 <!-------------------------------------------------------------------------- -->
693 <div class="doc_subsection">
694 <a name="deadargelim">Dead Argument Elimination</a>
696 <div class="doc_text">
698 This pass deletes dead arguments from internal functions. Dead argument
699 elimination removes arguments which are directly dead, as well as arguments
700 only passed into function calls as dead arguments of other functions. This
701 pass also deletes dead arguments in a similar way.
705 This pass is often useful as a cleanup pass to run after aggressive
706 interprocedural passes, which add possibly-dead arguments.
710 <!-------------------------------------------------------------------------- -->
711 <div class="doc_subsection">
712 <a name="deadtypeelim">Dead Type Elimination</a>
714 <div class="doc_text">
716 This pass is used to cleanup the output of GCC. It eliminate names for types
717 that are unused in the entire translation unit, using the <a
718 href="#findusedtypes">find used types</a> pass.
722 <!-------------------------------------------------------------------------- -->
723 <div class="doc_subsection">
724 <a name="die">Dead Instruction Elimination</a>
726 <div class="doc_text">
728 Dead instruction elimination performs a single pass over the function,
729 removing instructions that are obviously dead.
733 <!-------------------------------------------------------------------------- -->
734 <div class="doc_subsection">
735 <a name="dse">Dead Store Elimination</a>
737 <div class="doc_text">
739 A trivial dead store elimination that only considers basic-block local
744 <!-------------------------------------------------------------------------- -->
745 <div class="doc_subsection">
746 <a name="globaldce">Dead Global Elimination</a>
748 <div class="doc_text">
750 This transform is designed to eliminate unreachable internal globals from the
751 program. It uses an aggressive algorithm, searching out globals that are
752 known to be alive. After it finds all of the globals which are needed, it
753 deletes whatever is left over. This allows it to delete recursive chunks of
754 the program which are unreachable.
758 <!-------------------------------------------------------------------------- -->
759 <div class="doc_subsection">
760 <a name="globalopt">Global Variable Optimizer</a>
762 <div class="doc_text">
764 This pass transforms simple global variables that never have their address
765 taken. If obviously true, it marks read/write globals as constant, deletes
766 variables only stored to, etc.
770 <!-------------------------------------------------------------------------- -->
771 <div class="doc_subsection">
772 <a name="gvn">Global Value Numbering</a>
774 <div class="doc_text">
776 This pass performs global value numbering to eliminate fully and partially
777 redundant instructions. It also performs redundant load elimination.
782 <!-------------------------------------------------------------------------- -->
783 <div class="doc_subsection">
784 <a name="indmemrem">Indirect Malloc and Free Removal</a>
786 <div class="doc_text">
788 This pass finds places where memory allocation functions may escape into
789 indirect land. Some transforms are much easier (aka possible) only if free
790 or malloc are not called indirectly.
794 Thus find places where the address of memory functions are taken and construct
795 bounce functions with direct calls of those functions.
799 <!-------------------------------------------------------------------------- -->
800 <div class="doc_subsection">
801 <a name="indvars">Canonicalize Induction Variables</a>
803 <div class="doc_text">
805 This transformation analyzes and transforms the induction variables (and
806 computations derived from them) into simpler forms suitable for subsequent
807 analysis and transformation.
811 This transformation makes the following changes to each loop with an
812 identifiable induction variable:
816 <li>All loops are transformed to have a <em>single</em> canonical
817 induction variable which starts at zero and steps by one.</li>
818 <li>The canonical induction variable is guaranteed to be the first PHI node
819 in the loop header block.</li>
820 <li>Any pointer arithmetic recurrences are raised to use array
825 If the trip count of a loop is computable, this pass also makes the following
830 <li>The exit condition for the loop is canonicalized to compare the
831 induction value against the exit value. This turns loops like:
832 <blockquote><pre>for (i = 7; i*i < 1000; ++i)</pre></blockquote>
834 <blockquote><pre>for (i = 0; i != 25; ++i)</pre></blockquote></li>
835 <li>Any use outside of the loop of an expression derived from the indvar
836 is changed to compute the derived value outside of the loop, eliminating
837 the dependence on the exit value of the induction variable. If the only
838 purpose of the loop is to compute the exit value of some derived
839 expression, this transformation will make the loop dead.</li>
843 This transformation should be followed by strength reduction after all of the
844 desired loop transformations have been performed. Additionally, on targets
845 where it is profitable, the loop could be transformed to count down to zero
846 (the "do loop" optimization).
850 <!-------------------------------------------------------------------------- -->
851 <div class="doc_subsection">
852 <a name="inline">Function Integration/Inlining</a>
854 <div class="doc_text">
856 Bottom-up inlining of functions into callees.
860 <!-------------------------------------------------------------------------- -->
861 <div class="doc_subsection">
862 <a name="insert-block-profiling">Insert instrumentation for block profiling</a>
864 <div class="doc_text">
866 This pass instruments the specified program with counters for basic block
867 profiling, which counts the number of times each basic block executes. This
868 is the most basic form of profiling, which can tell which blocks are hot, but
869 cannot reliably detect hot paths through the CFG.
873 Note that this implementation is very naïve. Control equivalent regions of
874 the CFG should not require duplicate counters, but it does put duplicate
879 <!-------------------------------------------------------------------------- -->
880 <div class="doc_subsection">
881 <a name="insert-edge-profiling">Insert instrumentation for edge profiling</a>
883 <div class="doc_text">
885 This pass instruments the specified program with counters for edge profiling.
886 Edge profiling can give a reasonable approximation of the hot paths through a
887 program, and is used for a wide variety of program transformations.
891 Note that this implementation is very naïve. It inserts a counter for
892 <em>every</em> edge in the program, instead of using control flow information
893 to prune the number of counters inserted.
897 <!-------------------------------------------------------------------------- -->
898 <div class="doc_subsection">
899 <a name="insert-function-profiling">Insert instrumentation for function profiling</a>
901 <div class="doc_text">
903 This pass instruments the specified program with counters for function
904 profiling, which counts the number of times each function is called.
908 <!-------------------------------------------------------------------------- -->
909 <div class="doc_subsection">
910 <a name="insert-null-profiling-rs">Measure profiling framework overhead</a>
912 <div class="doc_text">
914 The basic profiler that does nothing. It is the default profiler and thus
915 terminates <code>RSProfiler</code> chains. It is useful for measuring
920 <!-------------------------------------------------------------------------- -->
921 <div class="doc_subsection">
922 <a name="insert-rs-profiling-framework">Insert random sampling instrumentation framework</a>
924 <div class="doc_text">
926 The second stage of the random-sampling instrumentation framework, duplicates
927 all instructions in a function, ignoring the profiling code, then connects the
928 two versions together at the entry and at backedges. At each connection point
929 a choice is made as to whether to jump to the profiled code (take a sample) or
930 execute the unprofiled code.
934 After this pass, it is highly recommended to run<a href="#mem2reg">mem2reg</a>
935 and <a href="#adce">adce</a>. <a href="#instcombine">instcombine</a>,
936 <a href="#load-vn">load-vn</a>, <a href="#gdce">gdce</a>, and
937 <a href="#dse">dse</a> also are good to run afterwards.
941 <!-------------------------------------------------------------------------- -->
942 <div class="doc_subsection">
943 <a name="instcombine">Combine redundant instructions</a>
945 <div class="doc_text">
947 Combine instructions to form fewer, simple
948 instructions. This pass does not modify the CFG This pass is where algebraic
949 simplification happens.
953 This pass combines things like:
958 %Z = add i32 %Y, 1</pre></blockquote>
965 >%Z = add i32 %X, 2</pre></blockquote>
968 This is a simple worklist driven algorithm.
972 This pass guarantees that the following canonicalizations are performed on
977 <li>If a binary operator has a constant operand, it is moved to the right-
979 <li>Bitwise operators with constant operands are always grouped so that
980 shifts are performed first, then <code>or</code>s, then
981 <code>and</code>s, then <code>xor</code>s.</li>
982 <li>Compare instructions are converted from <code><</code>,
983 <code>></code>, <code>≤</code>, or <code>≥</code> to
984 <code>=</code> or <code>≠</code> if possible.</li>
985 <li>All <code>cmp</code> instructions on boolean values are replaced with
986 logical operations.</li>
987 <li><code>add <var>X</var>, <var>X</var></code> is represented as
988 <code>mul <var>X</var>, 2</code> ⇒ <code>shl <var>X</var>, 1</code></li>
989 <li>Multiplies with a constant power-of-two argument are transformed into
995 <!-------------------------------------------------------------------------- -->
996 <div class="doc_subsection">
997 <a name="internalize">Internalize Global Symbols</a>
999 <div class="doc_text">
1001 This pass loops over all of the functions in the input module, looking for a
1002 main function. If a main function is found, all other functions and all
1003 global variables with initializers are marked as internal.
1007 <!-------------------------------------------------------------------------- -->
1008 <div class="doc_subsection">
1009 <a name="ipconstprop">Interprocedural constant propagation</a>
1011 <div class="doc_text">
1013 This pass implements an <em>extremely</em> simple interprocedural constant
1014 propagation pass. It could certainly be improved in many different ways,
1015 like using a worklist. This pass makes arguments dead, but does not remove
1016 them. The existing dead argument elimination pass should be run after this
1017 to clean up the mess.
1021 <!-------------------------------------------------------------------------- -->
1022 <div class="doc_subsection">
1023 <a name="ipsccp">Interprocedural Sparse Conditional Constant Propagation</a>
1025 <div class="doc_text">
1027 An interprocedural variant of <a href="#sccp">Sparse Conditional Constant
1032 <!-------------------------------------------------------------------------- -->
1033 <div class="doc_subsection">
1034 <a name="jump-threading">Thread control through conditional blocks</a>
1036 <div class="doc_text">
1038 Jump threading tries to find distinct threads of control flow running through
1039 a basic block. This pass looks at blocks that have multiple predecessors and
1040 multiple successors. If one or more of the predecessors of the block can be
1041 proven to always cause a jump to one of the successors, we forward the edge
1042 from the predecessor to the successor by duplicating the contents of this
1046 An example of when this can occur is code like this:
1053 if (X < 3) {</pre>
1056 In this case, the unconditional branch at the end of the first if can be
1057 revectored to the false side of the second if.
1061 <!-------------------------------------------------------------------------- -->
1062 <div class="doc_subsection">
1063 <a name="lcssa">Loop-Closed SSA Form Pass</a>
1065 <div class="doc_text">
1067 This pass transforms loops by placing phi nodes at the end of the loops for
1068 all values that are live across the loop boundary. For example, it turns
1069 the left into the right code:
1073 >for (...) for (...)
1078 X3 = phi(X1, X2) X3 = phi(X1, X2)
1079 ... = X3 + 4 X4 = phi(X3)
1083 This is still valid LLVM; the extra phi nodes are purely redundant, and will
1084 be trivially eliminated by <code>InstCombine</code>. The major benefit of
1085 this transformation is that it makes many other loop optimizations, such as
1086 LoopUnswitching, simpler.
1090 <!-------------------------------------------------------------------------- -->
1091 <div class="doc_subsection">
1092 <a name="licm">Loop Invariant Code Motion</a>
1094 <div class="doc_text">
1096 This pass performs loop invariant code motion, attempting to remove as much
1097 code from the body of a loop as possible. It does this by either hoisting
1098 code into the preheader block, or by sinking code to the exit blocks if it is
1099 safe. This pass also promotes must-aliased memory locations in the loop to
1100 live in registers, thus hoisting and sinking "invariant" loads and stores.
1104 This pass uses alias analysis for two purposes:
1108 <li>Moving loop invariant loads and calls out of loops. If we can determine
1109 that a load or call inside of a loop never aliases anything stored to,
1110 we can hoist it or sink it like any other instruction.</li>
1111 <li>Scalar Promotion of Memory - If there is a store instruction inside of
1112 the loop, we try to move the store to happen AFTER the loop instead of
1113 inside of the loop. This can only happen if a few conditions are true:
1115 <li>The pointer stored through is loop invariant.</li>
1116 <li>There are no stores or loads in the loop which <em>may</em> alias
1117 the pointer. There are no calls in the loop which mod/ref the
1120 If these conditions are true, we can promote the loads and stores in the
1121 loop of the pointer to use a temporary alloca'd variable. We then use
1122 the mem2reg functionality to construct the appropriate SSA form for the
1126 <!-------------------------------------------------------------------------- -->
1127 <div class="doc_subsection">
1128 <a name="loop-deletion">Dead Loop Deletion Pass</a>
1130 <div class="doc_text">
1132 This file implements the Dead Loop Deletion Pass. This pass is responsible
1133 for eliminating loops with non-infinite computable trip counts that have no
1134 side effects or volatile instructions, and do not contribute to the
1135 computation of the function's return value.
1139 <!-------------------------------------------------------------------------- -->
1140 <div class="doc_subsection">
1141 <a name="loop-extract">Extract loops into new functions</a>
1143 <div class="doc_text">
1145 A pass wrapper around the <code>ExtractLoop()</code> scalar transformation to
1146 extract each top-level loop into its own new function. If the loop is the
1147 <em>only</em> loop in a given function, it is not touched. This is a pass most
1148 useful for debugging via bugpoint.
1152 <!-------------------------------------------------------------------------- -->
1153 <div class="doc_subsection">
1154 <a name="loop-extract-single">Extract at most one loop into a new function</a>
1156 <div class="doc_text">
1158 Similar to <a href="#loop-extract">Extract loops into new functions</a>,
1159 this pass extracts one natural loop from the program into a function if it
1160 can. This is used by bugpoint.
1164 <!-------------------------------------------------------------------------- -->
1165 <div class="doc_subsection">
1166 <a name="loop-index-split">Index Split Loops</a>
1168 <div class="doc_text">
1170 This pass divides loop's iteration range by spliting loop such that each
1171 individual loop is executed efficiently.
1175 <!-------------------------------------------------------------------------- -->
1176 <div class="doc_subsection">
1177 <a name="loop-reduce">Loop Strength Reduction</a>
1179 <div class="doc_text">
1181 This pass performs a strength reduction on array references inside loops that
1182 have as one or more of their components the loop induction variable. This is
1183 accomplished by creating a new value to hold the initial value of the array
1184 access for the first iteration, and then creating a new GEP instruction in
1185 the loop to increment the value by the appropriate amount.
1189 <!-------------------------------------------------------------------------- -->
1190 <div class="doc_subsection">
1191 <a name="loop-rotate">Rotate Loops</a>
1193 <div class="doc_text">
1194 <p>A simple loop rotation transformation.</p>
1197 <!-------------------------------------------------------------------------- -->
1198 <div class="doc_subsection">
1199 <a name="loop-unroll">Unroll loops</a>
1201 <div class="doc_text">
1203 This pass implements a simple loop unroller. It works best when loops have
1204 been canonicalized by the <a href="#indvars"><tt>-indvars</tt></a> pass,
1205 allowing it to determine the trip counts of loops easily.
1209 <!-------------------------------------------------------------------------- -->
1210 <div class="doc_subsection">
1211 <a name="loop-unswitch">Unswitch loops</a>
1213 <div class="doc_text">
1215 This pass transforms loops that contain branches on loop-invariant conditions
1216 to have multiple loops. For example, it turns the left into the right code:
1228 This can increase the size of the code exponentially (doubling it every time
1229 a loop is unswitched) so we only unswitch if the resultant code will be
1230 smaller than a threshold.
1234 This pass expects LICM to be run before it to hoist invariant conditions out
1235 of the loop, to make the unswitching opportunity obvious.
1239 <!-------------------------------------------------------------------------- -->
1240 <div class="doc_subsection">
1241 <a name="loopsimplify">Canonicalize natural loops</a>
1243 <div class="doc_text">
1245 This pass performs several transformations to transform natural loops into a
1246 simpler form, which makes subsequent analyses and transformations simpler and
1251 Loop pre-header insertion guarantees that there is a single, non-critical
1252 entry edge from outside of the loop to the loop header. This simplifies a
1253 number of analyses and transformations, such as LICM.
1257 Loop exit-block insertion guarantees that all exit blocks from the loop
1258 (blocks which are outside of the loop that have predecessors inside of the
1259 loop) only have predecessors from inside of the loop (and are thus dominated
1260 by the loop header). This simplifies transformations such as store-sinking
1261 that are built into LICM.
1265 This pass also guarantees that loops will have exactly one backedge.
1269 Note that the simplifycfg pass will clean up blocks which are split out but
1270 end up being unnecessary, so usage of this pass should not pessimize
1275 This pass obviously modifies the CFG, but updates loop information and
1276 dominator information.
1280 <!-------------------------------------------------------------------------- -->
1281 <div class="doc_subsection">
1282 <a name="lowerallocs">Lower allocations from instructions to calls</a>
1284 <div class="doc_text">
1286 Turn <tt>malloc</tt> and <tt>free</tt> instructions into <tt>@malloc</tt> and
1287 <tt>@free</tt> calls.
1291 This is a target-dependent tranformation because it depends on the size of
1292 data types and alignment constraints.
1296 <!-------------------------------------------------------------------------- -->
1297 <div class="doc_subsection">
1298 <a name="lowerinvoke">Lower invoke and unwind, for unwindless code generators</a>
1300 <div class="doc_text">
1302 This transformation is designed for use by code generators which do not yet
1303 support stack unwinding. This pass supports two models of exception handling
1304 lowering, the 'cheap' support and the 'expensive' support.
1308 'Cheap' exception handling support gives the program the ability to execute
1309 any program which does not "throw an exception", by turning 'invoke'
1310 instructions into calls and by turning 'unwind' instructions into calls to
1311 abort(). If the program does dynamically use the unwind instruction, the
1312 program will print a message then abort.
1316 'Expensive' exception handling support gives the full exception handling
1317 support to the program at the cost of making the 'invoke' instruction
1318 really expensive. It basically inserts setjmp/longjmp calls to emulate the
1319 exception handling as necessary.
1323 Because the 'expensive' support slows down programs a lot, and EH is only
1324 used for a subset of the programs, it must be specifically enabled by the
1325 <tt>-enable-correct-eh-support</tt> option.
1329 Note that after this pass runs the CFG is not entirely accurate (exceptional
1330 control flow edges are not correct anymore) so only very simple things should
1331 be done after the lowerinvoke pass has run (like generation of native code).
1332 This should not be used as a general purpose "my LLVM-to-LLVM pass doesn't
1333 support the invoke instruction yet" lowering pass.
1337 <!-------------------------------------------------------------------------- -->
1338 <div class="doc_subsection">
1339 <a name="lowersetjmp">Lower Set Jump</a>
1341 <div class="doc_text">
1343 Lowers <tt>setjmp</tt> and <tt>longjmp</tt> to use the LLVM invoke and unwind
1344 instructions as necessary.
1348 Lowering of <tt>longjmp</tt> is fairly trivial. We replace the call with a
1349 call to the LLVM library function <tt>__llvm_sjljeh_throw_longjmp()</tt>.
1350 This unwinds the stack for us calling all of the destructors for
1351 objects allocated on the stack.
1355 At a <tt>setjmp</tt> call, the basic block is split and the <tt>setjmp</tt>
1356 removed. The calls in a function that have a <tt>setjmp</tt> are converted to
1357 invoke where the except part checks to see if it's a <tt>longjmp</tt>
1358 exception and, if so, if it's handled in the function. If it is, then it gets
1359 the value returned by the <tt>longjmp</tt> and goes to where the basic block
1360 was split. <tt>invoke</tt> instructions are handled in a similar fashion with
1361 the original except block being executed if it isn't a <tt>longjmp</tt>
1362 except that is handled by that function.
1366 <!-------------------------------------------------------------------------- -->
1367 <div class="doc_subsection">
1368 <a name="lowerswitch">Lower SwitchInst's to branches</a>
1370 <div class="doc_text">
1372 Rewrites <tt>switch</tt> instructions with a sequence of branches, which
1373 allows targets to get away with not implementing the switch instruction until
1378 <!-------------------------------------------------------------------------- -->
1379 <div class="doc_subsection">
1380 <a name="mem2reg">Promote Memory to Register</a>
1382 <div class="doc_text">
1384 This file promotes memory references to be register references. It promotes
1385 <tt>alloca</tt> instructions which only have <tt>load</tt>s and
1386 <tt>store</tt>s as uses. An <tt>alloca</tt> is transformed by using dominator
1387 frontiers to place <tt>phi</tt> nodes, then traversing the function in
1388 depth-first order to rewrite <tt>load</tt>s and <tt>store</tt>s as
1389 appropriate. This is just the standard SSA construction algorithm to construct
1394 <!-------------------------------------------------------------------------- -->
1395 <div class="doc_subsection">
1396 <a name="memcpyopt">Optimize use of memcpy and friend</a>
1398 <div class="doc_text">
1400 This pass performs various transformations related to eliminating memcpy
1401 calls, or transforming sets of stores into memset's.
1405 <!-------------------------------------------------------------------------- -->
1406 <div class="doc_subsection">
1407 <a name="mergereturn">Unify function exit nodes</a>
1409 <div class="doc_text">
1411 Ensure that functions have at most one <tt>ret</tt> instruction in them.
1412 Additionally, it keeps track of which node is the new exit node of the CFG.
1416 <!-------------------------------------------------------------------------- -->
1417 <div class="doc_subsection">
1418 <a name="prune-eh">Remove unused exception handling info</a>
1420 <div class="doc_text">
1422 This file implements a simple interprocedural pass which walks the call-graph,
1423 turning <tt>invoke</tt> instructions into <tt>call</tt> instructions if and
1424 only if the callee cannot throw an exception. It implements this as a
1425 bottom-up traversal of the call-graph.
1429 <!-------------------------------------------------------------------------- -->
1430 <div class="doc_subsection">
1431 <a name="reassociate">Reassociate expressions</a>
1433 <div class="doc_text">
1435 This pass reassociates commutative expressions in an order that is designed
1436 to promote better constant propagation, GCSE, LICM, PRE, etc.
1440 For example: 4 + (<var>x</var> + 5) ⇒ <var>x</var> + (4 + 5)
1444 In the implementation of this algorithm, constants are assigned rank = 0,
1445 function arguments are rank = 1, and other values are assigned ranks
1446 corresponding to the reverse post order traversal of current function
1447 (starting at 2), which effectively gives values in deep loops higher rank
1448 than values not in loops.
1452 <!-------------------------------------------------------------------------- -->
1453 <div class="doc_subsection">
1454 <a name="reg2mem">Demote all values to stack slots</a>
1456 <div class="doc_text">
1458 This file demotes all registers to memory references. It is intented to be
1459 the inverse of <a href="#mem2reg"><tt>-mem2reg</tt></a>. By converting to
1460 <tt>load</tt> instructions, the only values live across basic blocks are
1461 <tt>alloca</tt> instructions and <tt>load</tt> instructions before
1462 <tt>phi</tt> nodes. It is intended that this should make CFG hacking much
1463 easier. To make later hacking easier, the entry block is split into two, such
1464 that all introduced <tt>alloca</tt> instructions (and nothing else) are in the
1469 <!-------------------------------------------------------------------------- -->
1470 <div class="doc_subsection">
1471 <a name="scalarrepl">Scalar Replacement of Aggregates</a>
1473 <div class="doc_text">
1475 The well-known scalar replacement of aggregates transformation. This
1476 transform breaks up <tt>alloca</tt> instructions of aggregate type (structure
1477 or array) into individual <tt>alloca</tt> instructions for each member if
1478 possible. Then, if possible, it transforms the individual <tt>alloca</tt>
1479 instructions into nice clean scalar SSA form.
1483 This combines a simple scalar replacement of aggregates algorithm with the <a
1484 href="#mem2reg"><tt>mem2reg</tt></a> algorithm because often interact,
1485 especially for C++ programs. As such, iterating between <tt>scalarrepl</tt>,
1486 then <a href="#mem2reg"><tt>mem2reg</tt></a> until we run out of things to
1491 <!-------------------------------------------------------------------------- -->
1492 <div class="doc_subsection">
1493 <a name="sccp">Sparse Conditional Constant Propagation</a>
1495 <div class="doc_text">
1497 Sparse conditional constant propagation and merging, which can be summarized
1502 <li>Assumes values are constant unless proven otherwise</li>
1503 <li>Assumes BasicBlocks are dead unless proven otherwise</li>
1504 <li>Proves values to be constant, and replaces them with constants</li>
1505 <li>Proves conditional branches to be unconditional</li>
1509 Note that this pass has a habit of making definitions be dead. It is a good
1510 idea to to run a DCE pass sometime after running this pass.
1514 <!-------------------------------------------------------------------------- -->
1515 <div class="doc_subsection">
1516 <a name="simplify-libcalls">Simplify well-known library calls</a>
1518 <div class="doc_text">
1520 Applies a variety of small optimizations for calls to specific well-known
1521 function calls (e.g. runtime library functions). For example, a call
1522 <tt>exit(3)</tt> that occurs within the <tt>main()</tt> function can be
1523 transformed into simply <tt>return 3</tt>.
1527 <!-------------------------------------------------------------------------- -->
1528 <div class="doc_subsection">
1529 <a name="simplifycfg">Simplify the CFG</a>
1531 <div class="doc_text">
1533 Performs dead code elimination and basic block merging. Specifically:
1537 <li>Removes basic blocks with no predecessors.</li>
1538 <li>Merges a basic block into its predecessor if there is only one and the
1539 predecessor only has one successor.</li>
1540 <li>Eliminates PHI nodes for basic blocks with a single predecessor.</li>
1541 <li>Eliminates a basic block that only contains an unconditional
1546 <!-------------------------------------------------------------------------- -->
1547 <div class="doc_subsection">
1548 <a name="strip">Strip all symbols from a module</a>
1550 <div class="doc_text">
1552 Performs code stripping. This transformation can delete:
1556 <li>names for virtual registers</li>
1557 <li>symbols for internal globals and functions</li>
1558 <li>debug information</li>
1562 Note that this transformation makes code much less readable, so it should
1563 only be used in situations where the <tt>strip</tt> utility would be used,
1564 such as reducing code size or making it harder to reverse engineer code.
1568 <!-------------------------------------------------------------------------- -->
1569 <div class="doc_subsection">
1570 <a name="strip-dead-prototypes">Remove unused function declarations</a>
1572 <div class="doc_text">
1574 This pass loops over all of the functions in the input module, looking for
1575 dead declarations and removes them. Dead declarations are declarations of
1576 functions for which no implementation is available (i.e., declarations for
1577 unused library functions).
1581 <!-------------------------------------------------------------------------- -->
1582 <div class="doc_subsection">
1583 <a name="sretpromotion">Promote sret arguments</a>
1585 <div class="doc_text">
1587 This pass finds functions that return a struct (using a pointer to the struct
1588 as the first argument of the function, marked with the '<tt>sret</tt>' attribute) and
1589 replaces them with a new function that simply returns each of the elements of
1590 that struct (using multiple return values).
1594 This pass works under a number of conditions:
1598 <li>The returned struct must not contain other structs</li>
1599 <li>The returned struct must only be used to load values from</li>
1600 <li>The placeholder struct passed in is the result of an <tt>alloca</tt></li>
1604 <!-------------------------------------------------------------------------- -->
1605 <div class="doc_subsection">
1606 <a name="tailcallelim">Tail Call Elimination</a>
1608 <div class="doc_text">
1610 This file transforms calls of the current function (self recursion) followed
1611 by a return instruction with a branch to the entry of the function, creating
1612 a loop. This pass also implements the following extensions to the basic
1617 <li>Trivial instructions between the call and return do not prevent the
1618 transformation from taking place, though currently the analysis cannot
1619 support moving any really useful instructions (only dead ones).
1620 <li>This pass transforms functions that are prevented from being tail
1621 recursive by an associative expression to use an accumulator variable,
1622 thus compiling the typical naive factorial or <tt>fib</tt> implementation
1623 into efficient code.
1624 <li>TRE is performed if the function returns void, if the return
1625 returns the result returned by the call, or if the function returns a
1626 run-time constant on all exits from the function. It is possible, though
1627 unlikely, that the return returns something else (like constant 0), and
1628 can still be TRE'd. It can be TRE'd if <em>all other</em> return
1629 instructions in the function return the exact same value.
1630 <li>If it can prove that callees do not access theier caller stack frame,
1631 they are marked as eligible for tail call elimination (by the code
1636 <!-------------------------------------------------------------------------- -->
1637 <div class="doc_subsection">
1638 <a name="tailduplicate">Tail Duplication</a>
1640 <div class="doc_text">
1642 This pass performs a limited form of tail duplication, intended to simplify
1643 CFGs by removing some unconditional branches. This pass is necessary to
1644 straighten out loops created by the C front-end, but also is capable of
1645 making other code nicer. After this pass is run, the CFG simplify pass
1646 should be run to clean up the mess.
1650 <!-- ======================================================================= -->
1651 <div class="doc_section"> <a name="transform">Utility Passes</a></div>
1652 <div class="doc_text">
1653 <p>This section describes the LLVM Utility Passes.</p>
1656 <!-------------------------------------------------------------------------- -->
1657 <div class="doc_subsection">
1658 <a name="deadarghaX0r">Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</a>
1660 <div class="doc_text">
1662 Same as dead argument elimination, but deletes arguments to functions which
1663 are external. This is only for use by <a
1664 href="Bugpoint.html">bugpoint</a>.</p>
1667 <!-------------------------------------------------------------------------- -->
1668 <div class="doc_subsection">
1669 <a name="extract-blocks">Extract Basic Blocks From Module (for bugpoint use)</a>
1671 <div class="doc_text">
1673 This pass is used by bugpoint to extract all blocks from the module into their
1677 <!-------------------------------------------------------------------------- -->
1678 <div class="doc_subsection">
1679 <a name="preverify">Preliminary module verification</a>
1681 <div class="doc_text">
1683 Ensures that the module is in the form required by the <a
1684 href="#verifier">Module Verifier</a> pass.
1688 Running the verifier runs this pass automatically, so there should be no need
1693 <!-------------------------------------------------------------------------- -->
1694 <div class="doc_subsection">
1695 <a name="verify">Module Verifier</a>
1697 <div class="doc_text">
1699 Verifies an LLVM IR code. This is useful to run after an optimization which is
1700 undergoing testing. Note that <tt>llvm-as</tt> verifies its input before
1701 emitting bitcode, and also that malformed bitcode is likely to make LLVM
1702 crash. All language front-ends are therefore encouraged to verify their output
1703 before performing optimizing transformations.
1707 <li>Both of a binary operator's parameters are of the same type.</li>
1708 <li>Verify that the indices of mem access instructions match other
1710 <li>Verify that arithmetic and other things are only performed on
1711 first-class types. Verify that shifts and logicals only happen on
1713 <li>All of the constants in a switch statement are of the correct type.</li>
1714 <li>The code is in valid SSA form.</li>
1715 <li>It is illegal to put a label into any other type (like a structure) or
1717 <li>Only phi nodes can be self referential: <tt>%x = add i32 %x, %x</tt> is
1719 <li>PHI nodes must have an entry for each predecessor, with no extras.</li>
1720 <li>PHI nodes must be the first thing in a basic block, all grouped
1722 <li>PHI nodes must have at least one entry.</li>
1723 <li>All basic blocks should only end with terminator insts, not contain
1725 <li>The entry node to a function must not have predecessors.</li>
1726 <li>All Instructions must be embedded into a basic block.</li>
1727 <li>Functions cannot take a void-typed parameter.</li>
1728 <li>Verify that a function's argument list agrees with its declared
1730 <li>It is illegal to specify a name for a void value.</li>
1731 <li>It is illegal to have a internal global value with no initializer.</li>
1732 <li>It is illegal to have a ret instruction that returns a value that does
1733 not agree with the function return value type.</li>
1734 <li>Function call argument types match the function prototype.</li>
1735 <li>All other things that are tested by asserts spread about the code.</li>
1739 Note that this does not provide full security verification (like Java), but
1740 instead just tries to ensure that code is well-formed.
1744 <!-------------------------------------------------------------------------- -->
1745 <div class="doc_subsection">
1746 <a name="view-cfg">View CFG of function</a>
1748 <div class="doc_text">
1750 Displays the control flow graph using the GraphViz tool.
1754 <!-------------------------------------------------------------------------- -->
1755 <div class="doc_subsection">
1756 <a name="view-cfg-only">View CFG of function (with no function bodies)</a>
1758 <div class="doc_text">
1760 Displays the control flow graph using the GraphViz tool, but omitting function
1765 <!-- *********************************************************************** -->
1769 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1770 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1771 <a href="http://validator.w3.org/check/referer"><img
1772 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1774 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a><br>
1775 <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
1776 Last modified: $Date$