docs/Passes.html

   1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
   2                       "http://www.w3.org/TR/html4/strict.dtd">
   3 <html>
   4 <head>
   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">
   8 </head>
   9 <body>
  10
  11 <!--
  12
  13 If Passes.html is up to date, the following "one-liner" should print
  14 an empty diff.
  15
  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: $!\n";
  20 while (<HTML>) {
  21   m:^<tr><td><a href="#(.*)">-.*</a></td><td>.*</td></tr>$: or next;
  22   $order{$1} = sprintf("%03d", 1 + int %order);
  23 }
  24 open HELP, "../Release/bin/opt -help|" or die "open: opt -help: $!\n";
  25 while (<HELP>) {
  26   m:^    -([^ ]+) +- (.*)$: or next;
  27   my $o = $order{$1};
  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";
  31 }
  32 @x = map { s/^\d\d\d//; $_ } sort @x;
  33 @y = map { s/^\d\d\d//; $_ } sort @y;
  34 print @x, @y;
  35 EOT
  36
  37 This (real) one-liner can also be helpful when converting comments to HTML:
  38
  39 perl -e '$/ = 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'
  40
  41   -->
  42
  43 <div class="doc_title">LLVM's Analysis and Transform Passes</div>
  44
  45 <ol>
  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>
  50 </ol>
  51
  52 <div class="doc_author">
  53   <p>Written by <a href="mailto:rspencer@x10sys.com">Reid Spencer</a>
  54             and Gordon Henriksen</p>
  55 </div>
  56
  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>
  72 </div>
  73 <div class="doc_text" >
  74 <table>
  75 <tr><th colspan="2"><b>ANALYSIS PASSES</b></th></tr>
  76 <tr><th>Option</th><th>Name</th></tr>
  77 <tr><td><a href="#aa-eval">-aa-eval</a></td><td>Exhaustive Alias Analysis Precision Evaluator</td></tr>
  78 <tr><td><a href="#anders-aa">-anders-aa</a></td><td>Andersen's Interprocedural Alias Analysis</td></tr>
  79 <tr><td><a href="#basicaa">-basicaa</a></td><td>Basic Alias Analysis (default AA impl)</td></tr>
  80 <tr><td><a href="#basiccg">-basiccg</a></td><td>Basic CallGraph Construction</td></tr>
  81 <tr><td><a href="#basicvn">-basicvn</a></td><td>Basic Value Numbering (default GVN impl)</td></tr>
  82 <tr><td><a href="#callgraph">-callgraph</a></td><td>Print a call graph</td></tr>
  83 <tr><td><a href="#callscc">-callscc</a></td><td>Print SCCs of the Call Graph</td></tr>
  84 <tr><td><a href="#cfgscc">-cfgscc</a></td><td>Print SCCs of each function CFG</td></tr>
  85 <tr><td><a href="#codegenprepare">-codegenprepare</a></td><td>Optimize for code generation</td></tr>
  86 <tr><td><a href="#count-aa">-count-aa</a></td><td>Count Alias Analysis Query Responses</td></tr>
  87 <tr><td><a href="#debug-aa">-debug-aa</a></td><td>AA use debugger</td></tr>
  88 <tr><td><a href="#domfrontier">-domfrontier</a></td><td>Dominance Frontier Construction</td></tr>
  89 <tr><td><a href="#domtree">-domtree</a></td><td>Dominator Tree Construction</td></tr>
  90 <tr><td><a href="#externalfnconstants">-externalfnconstants</a></td><td>Print external fn callsites passed constants</td></tr>
  91 <tr><td><a href="#globalsmodref-aa">-globalsmodref-aa</a></td><td>Simple mod/ref analysis for globals</td></tr>
  92 <tr><td><a href="#instcount">-instcount</a></td><td>Counts the various types of Instructions</td></tr>
  93 <tr><td><a href="#intervals">-intervals</a></td><td>Interval Partition Construction</td></tr>
  94 <tr><td><a href="#load-vn">-load-vn</a></td><td>Load Value Numbering</td></tr>
  95 <tr><td><a href="#loops">-loops</a></td><td>Natural Loop Construction</td></tr>
  96 <tr><td><a href="#memdep">-memdep</a></td><td>Memory Dependence Analysis</td></tr>
  97 <tr><td><a href="#no-aa">-no-aa</a></td><td>No Alias Analysis (always returns 'may' alias)</td></tr>
  98 <tr><td><a href="#no-profile">-no-profile</a></td><td>No Profile Information</td></tr>
  99 <tr><td><a href="#postdomfrontier">-postdomfrontier</a></td><td>Post-Dominance Frontier Construction</td></tr>
 100 <tr><td><a href="#postdomtree">-postdomtree</a></td><td>Post-Dominator Tree Construction</td></tr>
 101 <tr><td><a href="#print">-print</a></td><td>Print function to stderr</td></tr>
 102 <tr><td><a href="#print-alias-sets">-print-alias-sets</a></td><td>Alias Set Printer</td></tr>
 103 <tr><td><a href="#print-callgraph">-print-callgraph</a></td><td>Print Call Graph to 'dot' file</td></tr>
 104 <tr><td><a href="#print-cfg">-print-cfg</a></td><td>Print CFG of function to 'dot' file</td></tr>
 105 <tr><td><a href="#print-cfg-only">-print-cfg-only</a></td><td>Print CFG of function to 'dot' file (with no function bodies)</td></tr>
 106 <tr><td><a href="#printm">-printm</a></td><td>Print module to stderr</td></tr>
 107 <tr><td><a href="#printusedtypes">-printusedtypes</a></td><td>Find Used Types</td></tr>
 108 <tr><td><a href="#profile-loader">-profile-loader</a></td><td>Load profile information from llvmprof.out</td></tr>
 109 <tr><td><a href="#scalar-evolution">-scalar-evolution</a></td><td>Scalar Evolution Analysis</td></tr>
 110 <tr><td><a href="#targetdata">-targetdata</a></td><td>Target Data Layout</td></tr>
 111
 112
 113 <tr><th colspan="2"><b>TRANSFORM PASSES</b></th></tr>
 114 <tr><th>Option</th><th>Name</th></tr>
 115 <tr><td><a href="#adce">-adce</a></td><td>Aggressive Dead Code Elimination</td></tr>
 116 <tr><td><a href="#argpromotion">-argpromotion</a></td><td>Promote 'by reference' arguments to scalars</td></tr>
 117 <tr><td><a href="#block-placement">-block-placement</a></td><td>Profile Guided Basic Block Placement</td></tr>
 118 <tr><td><a href="#break-crit-edges">-break-crit-edges</a></td><td>Break critical edges in CFG</td></tr>
 119 <tr><td><a href="#cee">-cee</a></td><td>Correlated Expression Elimination</td></tr>
 120 <tr><td><a href="#condprop">-condprop</a></td><td>Conditional Propagation</td></tr>
 121 <tr><td><a href="#constmerge">-constmerge</a></td><td>Merge Duplicate Global Constants</td></tr>
 122 <tr><td><a href="#constprop">-constprop</a></td><td>Simple constant propagation</td></tr>
 123 <tr><td><a href="#dce">-dce</a></td><td>Dead Code Elimination</td></tr>
 124 <tr><td><a href="#deadargelim">-deadargelim</a></td><td>Dead Argument Elimination</td></tr>
 125 <tr><td><a href="#deadtypeelim">-deadtypeelim</a></td><td>Dead Type Elimination</td></tr>
 126 <tr><td><a href="#die">-die</a></td><td>Dead Instruction Elimination</td></tr>
 127 <tr><td><a href="#dse">-dse</a></td><td>Dead Store Elimination</td></tr>
 128 <tr><td><a href="#gcse">-gcse</a></td><td>Global Common Subexpression Elimination</td></tr>
 129 <tr><td><a href="#globaldce">-globaldce</a></td><td>Dead Global Elimination</td></tr>
 130 <tr><td><a href="#globalopt">-globalopt</a></td><td>Global Variable Optimizer</td></tr>
 131 <tr><td><a href="#gvn">-gvn</a></td><td>Global Value Numbering</td></tr>
 132 <tr><td><a href="#gvnpre">-gvnpre</a></td><td>Global Value Numbering/Partial Redundancy Elimination</td></tr>
 133 <tr><td><a href="#indmemrem">-indmemrem</a></td><td>Indirect Malloc and Free Removal</td></tr>
 134 <tr><td><a href="#indvars">-indvars</a></td><td>Canonicalize Induction Variables</td></tr>
 135 <tr><td><a href="#inline">-inline</a></td><td>Function Integration/Inlining</td></tr>
 136 <tr><td><a href="#insert-block-profiling">-insert-block-profiling</a></td><td>Insert instrumentation for block profiling</td></tr>
 137 <tr><td><a href="#insert-edge-profiling">-insert-edge-profiling</a></td><td>Insert instrumentation for edge profiling</td></tr>
 138 <tr><td><a href="#insert-function-profiling">-insert-function-profiling</a></td><td>Insert instrumentation for function profiling</td></tr>
 139 <tr><td><a href="#insert-null-profiling-rs">-insert-null-profiling-rs</a></td><td>Measure profiling framework overhead</td></tr>
 140 <tr><td><a href="#insert-rs-profiling-framework">-insert-rs-profiling-framework</a></td><td>Insert random sampling instrumentation framework</td></tr>
 141 <tr><td><a href="#instcombine">-instcombine</a></td><td>Combine redundant instructions</td></tr>
 142 <tr><td><a href="#internalize">-internalize</a></td><td>Internalize Global Symbols</td></tr>
 143 <tr><td><a href="#ipconstprop">-ipconstprop</a></td><td>Interprocedural constant propagation</td></tr>
 144 <tr><td><a href="#ipsccp">-ipsccp</a></td><td>Interprocedural Sparse Conditional Constant Propagation</td></tr>
 145 <tr><td><a href="#lcssa">-lcssa</a></td><td>Loop-Closed SSA Form Pass</td></tr>
 146 <tr><td><a href="#licm">-licm</a></td><td>Loop Invariant Code Motion</td></tr>
 147 <tr><td><a href="#loop-extract">-loop-extract</a></td><td>Extract loops into new functions</td></tr>
 148 <tr><td><a href="#loop-extract-single">-loop-extract-single</a></td><td>Extract at most one loop into a new function</td></tr>
 149 <tr><td><a href="#loop-index-split">-loop-index-split</a></td><td>Index Split Loops</td></tr>
 150 <tr><td><a href="#loop-reduce">-loop-reduce</a></td><td>Loop Strength Reduction</td></tr>
 151 <tr><td><a href="#loop-rotate">-loop-rotate</a></td><td>Rotate Loops</td></tr>
 152 <tr><td><a href="#loop-unroll">-loop-unroll</a></td><td>Unroll loops</td></tr>
 153 <tr><td><a href="#loop-unswitch">-loop-unswitch</a></td><td>Unswitch loops</td></tr>
 154 <tr><td><a href="#loopsimplify">-loopsimplify</a></td><td>Canonicalize natural loops</td></tr>
 155 <tr><td><a href="#lower-packed">-lower-packed</a></td><td>lowers packed operations to operations on smaller packed datatypes</td></tr>
 156 <tr><td><a href="#lowerallocs">-lowerallocs</a></td><td>Lower allocations from instructions to calls</td></tr>
 157 <tr><td><a href="#lowergc">-lowergc</a></td><td>Lower GC intrinsics, for GCless code generators</td></tr>
 158 <tr><td><a href="#lowerinvoke">-lowerinvoke</a></td><td>Lower invoke and unwind, for unwindless code generators</td></tr>
 159 <tr><td><a href="#lowerselect">-lowerselect</a></td><td>Lower select instructions to branches</td></tr>
 160 <tr><td><a href="#lowersetjmp">-lowersetjmp</a></td><td>Lower Set Jump</td></tr>
 161 <tr><td><a href="#lowerswitch">-lowerswitch</a></td><td>Lower SwitchInst's to branches</td></tr>
 162 <tr><td><a href="#mem2reg">-mem2reg</a></td><td>Promote Memory to Register</td></tr>
 163 <tr><td><a href="#mergereturn">-mergereturn</a></td><td>Unify function exit nodes</td></tr>
 164 <tr><td><a href="#predsimplify">-predsimplify</a></td><td>Predicate Simplifier</td></tr>
 165 <tr><td><a href="#prune-eh">-prune-eh</a></td><td>Remove unused exception handling info</td></tr>
 166 <tr><td><a href="#raiseallocs">-raiseallocs</a></td><td>Raise allocations from calls to instructions</td></tr>
 167 <tr><td><a href="#reassociate">-reassociate</a></td><td>Reassociate expressions</td></tr>
 168 <tr><td><a href="#reg2mem">-reg2mem</a></td><td>Demote all values to stack slots</td></tr>
 169 <tr><td><a href="#scalarrepl">-scalarrepl</a></td><td>Scalar Replacement of Aggregates</td></tr>
 170 <tr><td><a href="#sccp">-sccp</a></td><td>Sparse Conditional Constant Propagation</td></tr>
 171 <tr><td><a href="#simplify-libcalls">-simplify-libcalls</a></td><td>Simplify well-known library calls</td></tr>
 172 <tr><td><a href="#simplifycfg">-simplifycfg</a></td><td>Simplify the CFG</td></tr>
 173 <tr><td><a href="#strip">-strip</a></td><td>Strip all symbols from a module</td></tr>
 174 <tr><td><a href="#tailcallelim">-tailcallelim</a></td><td>Tail Call Elimination</td></tr>
 175 <tr><td><a href="#tailduplicate">-tailduplicate</a></td><td>Tail Duplication</td></tr>
 176
 177
 178 <tr><th colspan="2"><b>UTILITY PASSES</b></th></tr>
 179 <tr><th>Option</th><th>Name</th></tr>
 180 <tr><td><a href="#deadarghaX0r">-deadarghaX0r</a></td><td>Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</td></tr>
 181 <tr><td><a href="#extract-blocks">-extract-blocks</a></td><td>Extract Basic Blocks From Module (for bugpoint use)</td></tr>
 182 <tr><td><a href="#preverify">-preverify</a></td><td>Preliminary module verification</td></tr>
 183 <tr><td><a href="#verify">-verify</a></td><td>Module Verifier</td></tr>
 184 <tr><td><a href="#view-cfg">-view-cfg</a></td><td>View CFG of function</td></tr>
 185 <tr><td><a href="#view-cfg-only">-view-cfg-only</a></td><td>View CFG of function (with no function bodies)</td></tr>
 186 </table>
 187 </div>
 188
 189 <!-- ======================================================================= -->
 190 <div class="doc_section"> <a name="example">Analysis Passes</a></div>
 191 <div class="doc_text">
 192   <p>This section describes the LLVM Analysis Passes.</p>
 193 </div>
 194
 195 <!-------------------------------------------------------------------------- -->
 196 <div class="doc_subsection">
 197   <a name="aa-eval">Exhaustive Alias Analysis Precision Evaluator</a>
 198 </div>
 199 <div class="doc_text">
 200   <p>This is a simple N^2 alias analysis accuracy evaluator.
 201   Basically, for each function in the program, it simply queries to see how the
 202   alias analysis implementation answers alias queries between each pair of
 203   pointers in the function.</p>
 204
 205   <p>This is inspired and adapted from code by: Naveen Neelakantam, Francesco
 206   Spadini, and Wojciech Stryjewski.</p>
 207 </div>
 208
 209 <!-------------------------------------------------------------------------- -->
 210 <div class="doc_subsection">
 211   <a name="anders-aa">Andersen's Interprocedural Alias Analysis</a>
 212 </div>
 213 <div class="doc_text">
 214   <p>
 215   This is an implementation of Andersen's interprocedural alias
 216   analysis
 217   </p>
 218
 219   <p>
 220   In pointer analysis terms, this is a subset-based, flow-insensitive,
 221   field-sensitive, and context-insensitive algorithm pointer algorithm.
 222   </p>
 223
 224   <p>
 225   This algorithm is implemented as three stages:
 226   </p>
 227
 228   <ol>
 229     <li>Object identification.</li>
 230     <li>Inclusion constraint identification.</li>
 231     <li>Offline constraint graph optimization.</li>
 232     <li>Inclusion constraint solving.</li>
 233   </ol>
 234
 235   <p>
 236   The object identification stage identifies all of the memory objects in the
 237   program, which includes globals, heap allocated objects, and stack allocated
 238   objects.
 239   </p>
 240
 241   <p>
 242   The inclusion constraint identification stage finds all inclusion constraints
 243   in the program by scanning the program, looking for pointer assignments and
 244   other statements that effect the points-to graph.  For a statement like
 245   <code><var>A</var> = <var>B</var></code>, this statement is processed to
 246   indicate that <var>A</var> can point to anything that <var>B</var> can point
 247   to.  Constraints can handle copies, loads, and stores, and address taking.
 248   </p>
 249
 250   <p>
 251   The offline constraint graph optimization portion includes offline variable
 252   substitution algorithms intended to computer pointer and location
 253   equivalences.  Pointer equivalences are those pointers that will have the
 254   same points-to sets, and location equivalences are those variables that
 255   always appear together in points-to sets.
 256   </p>
 257
 258   <p>
 259   The inclusion constraint solving phase iteratively propagates the inclusion
 260   constraints until a fixed point is reached.  This is an O(<var>n</var>³)
 261   algorithm.
 262   </p>
 263
 264   <p>
 265   Function constraints are handled as if they were structs with <var>X</var>
 266   fields. Thus, an access to argument <var>X</var> of function <var>Y</var> is
 267   an access to node index <code>getNode(<var>Y</var>) + <var>X</var></code>.
 268   This representation allows handling of indirect calls without any issues.  To
 269   wit, an indirect call <code><var>Y</var>(<var>a</var>,<var>b</var>)</code> is
 270   equivalent to <code>*(<var>Y</var> + 1) = <var>a</var>, *(<var>Y</var> + 2) =
 271   <var>b</var></code>. The return node for a function <var>F</var> is always
 272   located at <code>getNode(<var>F</var>) + CallReturnPos</code>. The arguments
 273   start at <code>getNode(<var>F</var>) + CallArgPos</code>.
 274   </p>
 275 </div>
 276
 277 <!-------------------------------------------------------------------------- -->
 278 <div class="doc_subsection">
 279   <a name="basicaa">Basic Alias Analysis (default AA impl)</a>
 280 </div>
 281 <div class="doc_text">
 282   <p>
 283   This is the default implementation of the Alias Analysis interface
 284   that simply implements a few identities (two different globals cannot alias,
 285   etc), but otherwise does no analysis.
 286   </p>
 287 </div>
 288
 289 <!-------------------------------------------------------------------------- -->
 290 <div class="doc_subsection">
 291   <a name="basiccg">Basic CallGraph Construction</a>
 292 </div>
 293 <div class="doc_text">
 294   <p>Yet to be written.</p>
 295 </div>
 296
 297 <!-------------------------------------------------------------------------- -->
 298 <div class="doc_subsection">
 299   <a name="basicvn">Basic Value Numbering (default GVN impl)</a>
 300 </div>
 301 <div class="doc_text">
 302   <p>
 303   This is the default implementation of the <code>ValueNumbering</code>
 304   interface.  It walks the SSA def-use chains to trivially identify
 305   lexically identical expressions.  This does not require any ahead of time
 306   analysis, so it is a very fast default implementation.
 307   </p>
 308 </div>
 309
 310 <!-------------------------------------------------------------------------- -->
 311 <div class="doc_subsection">
 312   <a name="callgraph">Print a call graph</a>
 313 </div>
 314 <div class="doc_text">
 315   <p>
 316   This pass, only available in <code>opt</code>, prints the call graph to
 317   standard output in a human-readable form.
 318   </p>
 319 </div>
 320
 321 <!-------------------------------------------------------------------------- -->
 322 <div class="doc_subsection">
 323   <a name="callscc">Print SCCs of the Call Graph</a>
 324 </div>
 325 <div class="doc_text">
 326   <p>
 327   This pass, only available in <code>opt</code>, prints the SCCs of the call
 328   graph to standard output in a human-readable form.
 329   </p>
 330 </div>
 331
 332 <!-------------------------------------------------------------------------- -->
 333 <div class="doc_subsection">
 334   <a name="cfgscc">Print SCCs of each function CFG</a>
 335 </div>
 336 <div class="doc_text">
 337   <p>
 338   This pass, only available in <code>opt</code>, prints the SCCs of each
 339   function CFG to standard output in a human-readable form.
 340   </p>
 341 </div>
 342
 343 <!-------------------------------------------------------------------------- -->
 344 <div class="doc_subsection">
 345   <a name="codegenprepare">Optimize for code generation</a>
 346 </div>
 347 <div class="doc_text">
 348   <p>
 349   This pass munges the code in the input function to better prepare it for
 350   SelectionDAG-based code generation.  This works around limitations in it's
 351   basic-block-at-a-time approach.  It should eventually be removed.
 352   </p>
 353 </div>
 354
 355 <!-------------------------------------------------------------------------- -->
 356 <div class="doc_subsection">
 357   <a name="count-aa">Count Alias Analysis Query Responses</a>
 358 </div>
 359 <div class="doc_text">
 360   <p>
 361   A pass which can be used to count how many alias queries
 362   are being made and how the alias analysis implementation being used responds.
 363   </p>
 364 </div>
 365
 366 <!-------------------------------------------------------------------------- -->
 367 <div class="doc_subsection">
 368   <a name="debug-aa">AA use debugger</a>
 369 </div>
 370 <div class="doc_text">
 371   <p>
 372   This simple pass checks alias analysis users to ensure that if they
 373   create a new value, they do not query AA without informing it of the value.
 374   It acts as a shim over any other AA pass you want.
 375   </p>
 376
 377   <p>
 378   Yes keeping track of every value in the program is expensive, but this is
 379   a debugging pass.
 380   </p>
 381 </div>
 382
 383 <!-------------------------------------------------------------------------- -->
 384 <div class="doc_subsection">
 385   <a name="domfrontier">Dominance Frontier Construction</a>
 386 </div>
 387 <div class="doc_text">
 388   <p>
 389   This pass is a simple dominator construction algorithm for finding forward
 390   dominator frontiers.
 391   </p>
 392 </div>
 393
 394 <!-------------------------------------------------------------------------- -->
 395 <div class="doc_subsection">
 396   <a name="domtree">Dominator Tree Construction</a>
 397 </div>
 398 <div class="doc_text">
 399   <p>
 400   This pass is a simple dominator construction algorithm for finding forward
 401   dominators.
 402   </p>
 403 </div>
 404
 405 <!-------------------------------------------------------------------------- -->
 406 <div class="doc_subsection">
 407   <a name="externalfnconstants">Print external fn callsites passed constants</a>
 408 </div>
 409 <div class="doc_text">
 410   <p>
 411   This pass, only available in <code>opt</code>, prints out call sites to
 412   external functions that are called with constant arguments.  This can be
 413   useful when looking for standard library functions we should constant fold
 414   or handle in alias analyses.
 415   </p>
 416 </div>
 417
 418 <!-------------------------------------------------------------------------- -->
 419 <div class="doc_subsection">
 420   <a name="globalsmodref-aa">Simple mod/ref analysis for globals</a>
 421 </div>
 422 <div class="doc_text">
 423   <p>
 424   This simple pass provides alias and mod/ref information for global values
 425   that do not have their address taken, and keeps track of whether functions
 426   read or write memory (are "pure").  For this simple (but very common) case,
 427   we can provide pretty accurate and useful information.
 428   </p>
 429 </div>
 430
 431 <!-------------------------------------------------------------------------- -->
 432 <div class="doc_subsection">
 433   <a name="instcount">Counts the various types of Instructions</a>
 434 </div>
 435 <div class="doc_text">
 436   <p>
 437   This pass collects the count of all instructions and reports them
 438   </p>
 439 </div>
 440
 441 <!-------------------------------------------------------------------------- -->
 442 <div class="doc_subsection">
 443   <a name="intervals">Interval Partition Construction</a>
 444 </div>
 445 <div class="doc_text">
 446   <p>
 447   This analysis calculates and represents the interval partition of a function,
 448   or a preexisting interval partition.
 449   </p>
 450
 451   <p>
 452   In this way, the interval partition may be used to reduce a flow graph down
 453   to its degenerate single node interval partition (unless it is irreducible).
 454   </p>
 455 </div>
 456
 457 <!-------------------------------------------------------------------------- -->
 458 <div class="doc_subsection">
 459   <a name="load-vn">Load Value Numbering</a>
 460 </div>
 461 <div class="doc_text">
 462   <p>
 463   This pass value numbers load and call instructions.  To do this, it finds
 464   lexically identical load instructions, and uses alias analysis to determine
 465   which loads are guaranteed to produce the same value.  To value number call
 466   instructions, it looks for calls to functions that do not write to memory
 467   which do not have intervening instructions that clobber the memory that is
 468   read from.
 469   </p>
 470
 471   <p>
 472   This pass builds off of another value numbering pass to implement value
 473   numbering for non-load and non-call instructions.  It uses Alias Analysis so
 474   that it can disambiguate the load instructions.  The more powerful these base
 475   analyses are, the more powerful the resultant value numbering will be.
 476   </p>
 477 </div>
 478
 479 <!-------------------------------------------------------------------------- -->
 480 <div class="doc_subsection">
 481   <a name="loops">Natural Loop Construction</a>
 482 </div>
 483 <div class="doc_text">
 484   <p>
 485   This analysis is used to identify natural loops and determine the loop depth
 486   of various nodes of the CFG.  Note that the loops identified may actually be
 487   several natural loops that share the same header node... not just a single
 488   natural loop.
 489   </p>
 490 </div>
 491
 492 <!-------------------------------------------------------------------------- -->
 493 <div class="doc_subsection">
 494   <a name="memdep">Memory Dependence Analysis</a>
 495 </div>
 496 <div class="doc_text">
 497   <p>
 498   An analysis that determines, for a given memory operation, what preceding
 499   memory operations it depends on.  It builds on alias analysis information, and
 500   tries to provide a lazy, caching interface to a common kind of alias
 501   information query.
 502   </p>
 503 </div>
 504
 505 <!-------------------------------------------------------------------------- -->
 506 <div class="doc_subsection">
 507   <a name="no-aa">No Alias Analysis (always returns 'may' alias)</a>
 508 </div>
 509 <div class="doc_text">
 510   <p>
 511   Always returns "I don't know" for alias queries.  NoAA is unlike other alias
 512   analysis implementations, in that it does not chain to a previous analysis. As
 513   such it doesn't follow many of the rules that other alias analyses must.
 514   </p>
 515 </div>
 516
 517 <!-------------------------------------------------------------------------- -->
 518 <div class="doc_subsection">
 519   <a name="no-profile">No Profile Information</a>
 520 </div>
 521 <div class="doc_text">
 522   <p>
 523   The default "no profile" implementation of the abstract
 524   <code>ProfileInfo</code> interface.
 525   </p>
 526 </div>
 527
 528 <!-------------------------------------------------------------------------- -->
 529 <div class="doc_subsection">
 530   <a name="postdomfrontier">Post-Dominance Frontier Construction</a>
 531 </div>
 532 <div class="doc_text">
 533   <p>
 534   This pass is a simple post-dominator construction algorithm for finding
 535   post-dominator frontiers.
 536   </p>
 537 </div>
 538
 539 <!-------------------------------------------------------------------------- -->
 540 <div class="doc_subsection">
 541   <a name="postdomtree">Post-Dominator Tree Construction</a>
 542 </div>
 543 <div class="doc_text">
 544   <p>
 545   This pass is a simple post-dominator construction algorithm for finding
 546   post-dominators.
 547   </p>
 548 </div>
 549
 550 <!-------------------------------------------------------------------------- -->
 551 <div class="doc_subsection">
 552   <a name="print">Print function to stderr</a>
 553 </div>
 554 <div class="doc_text">
 555   <p>
 556   The <code>PrintFunctionPass</code> class is designed to be pipelined with
 557   other <code>FunctionPass</code>es, and prints out the functions of the module
 558   as they are processed.
 559   </p>
 560 </div>
 561
 562 <!-------------------------------------------------------------------------- -->
 563 <div class="doc_subsection">
 564   <a name="print-alias-sets">Alias Set Printer</a>
 565 </div>
 566 <div class="doc_text">
 567   <p>Yet to be written.</p>
 568 </div>
 569
 570 <!-------------------------------------------------------------------------- -->
 571 <div class="doc_subsection">
 572   <a name="print-callgraph">Print Call Graph to 'dot' file</a>
 573 </div>
 574 <div class="doc_text">
 575   <p>
 576   This pass, only available in <code>opt</code>, prints the call graph into a
 577   <code>.dot</code> graph.  This graph can then be processed with the "dot" tool
 578   to convert it to postscript or some other suitable format.
 579   </p>
 580 </div>
 581
 582 <!-------------------------------------------------------------------------- -->
 583 <div class="doc_subsection">
 584   <a name="print-cfg">Print CFG of function to 'dot' file</a>
 585 </div>
 586 <div class="doc_text">
 587   <p>
 588   This pass, only available in <code>opt</code>, prints the control flow graph
 589   into a <code>.dot</code> graph.  This graph can then be processed with the
 590   "dot" tool to convert it to postscript or some other suitable format.
 591   </p>
 592 </div>
 593
 594 <!-------------------------------------------------------------------------- -->
 595 <div class="doc_subsection">
 596   <a name="print-cfg-only">Print CFG of function to 'dot' file (with no function bodies)</a>
 597 </div>
 598 <div class="doc_text">
 599   <p>
 600   This pass, only available in <code>opt</code>, prints the control flow graph
 601   into a <code>.dot</code> graph, omitting the function bodies.  This graph can
 602   then be processed with the "dot" tool to convert it to postscript or some
 603   other suitable format.
 604   </p>
 605 </div>
 606
 607 <!-------------------------------------------------------------------------- -->
 608 <div class="doc_subsection">
 609   <a name="printm">Print module to stderr</a>
 610 </div>
 611 <div class="doc_text">
 612   <p>
 613   This pass simply prints out the entire module when it is executed.
 614   </p>
 615 </div>
 616
 617 <!-------------------------------------------------------------------------- -->
 618 <div class="doc_subsection">
 619   <a name="printusedtypes">Find Used Types</a>
 620 </div>
 621 <div class="doc_text">
 622   <p>
 623   This pass is used to seek out all of the types in use by the program.  Note
 624   that this analysis explicitly does not include types only used by the symbol
 625   table.
 626 </div>
 627
 628 <!-------------------------------------------------------------------------- -->
 629 <div class="doc_subsection">
 630   <a name="profile-loader">Load profile information from llvmprof.out</a>
 631 </div>
 632 <div class="doc_text">
 633   <p>
 634   A concrete implementation of profiling information that loads the information
 635   from a profile dump file.
 636   </p>
 637 </div>
 638
 639 <!-------------------------------------------------------------------------- -->
 640 <div class="doc_subsection">
 641   <a name="scalar-evolution">Scalar Evolution Analysis</a>
 642 </div>
 643 <div class="doc_text">
 644   <p>
 645   The <code>ScalarEvolution</code> analysis can be used to analyze and
 646   catagorize scalar expressions in loops.  It specializes in recognizing general
 647   induction variables, representing them with the abstract and opaque
 648   <code>SCEV</code> class.  Given this analysis, trip counts of loops and other
 649   important properties can be obtained.
 650   </p>
 651
 652   <p>
 653   This analysis is primarily useful for induction variable substitution and
 654   strength reduction.
 655   </p>
 656 </div>
 657
 658 <!-------------------------------------------------------------------------- -->
 659 <div class="doc_subsection">
 660   <a name="targetdata">Target Data Layout</a>
 661 </div>
 662 <div class="doc_text">
 663   <p>Provides other passes access to information on how the size and alignment
 664   required by the the target ABI for various data types.</p>
 665 </div>
 666
 667 <!-- ======================================================================= -->
 668 <div class="doc_section"> <a name="transform">Transform Passes</a></div>
 669 <div class="doc_text">
 670   <p>This section describes the LLVM Transform Passes.</p>
 671 </div>
 672
 673 <!-------------------------------------------------------------------------- -->
 674 <div class="doc_subsection">
 675   <a name="adce">Aggressive Dead Code Elimination</a>
 676 </div>
 677 <div class="doc_text">
 678   <p>ADCE aggressively tries to eliminate code. This pass is similar to
 679   <a href="#dce">DCE</a> but it assumes that values are dead until proven
 680   otherwise. This is similar to <a href="#sccp">SCCP</a>, except applied to
 681   the liveness of values.</p>
 682 </div>
 683
 684 <!-------------------------------------------------------------------------- -->
 685 <div class="doc_subsection">
 686   <a name="argpromotion">Promote 'by reference' arguments to scalars</a>
 687 </div>
 688 <div class="doc_text">
 689   <p>
 690   This pass promotes "by reference" arguments to be "by value" arguments.  In
 691   practice, this means looking for internal functions that have pointer
 692   arguments.  If it can prove, through the use of alias analysis, that an
 693   argument is *only* loaded, then it can pass the value into the function
 694   instead of the address of the value.  This can cause recursive simplification
 695   of code and lead to the elimination of allocas (especially in C++ template
 696   code like the STL).
 697   </p>
 698
 699   <p>
 700   This pass also handles aggregate arguments that are passed into a function,
 701   scalarizing them if the elements of the aggregate are only loaded.  Note that
 702   it refuses to scalarize aggregates which would require passing in more than
 703   three operands to the function, because passing thousands of operands for a
 704   large array or structure is unprofitable!
 705   </p>
 706
 707   <p>
 708   Note that this transformation could also be done for arguments that are only
 709   stored to (returning the value instead), but does not currently.  This case
 710   would be best handled when and if LLVM starts supporting multiple return
 711   values from functions.
 712   </p>
 713 </div>
 714
 715 <!-------------------------------------------------------------------------- -->
 716 <div class="doc_subsection">
 717   <a name="block-placement">Profile Guided Basic Block Placement</a>
 718 </div>
 719 <div class="doc_text">
 720   <p>This pass is a very simple profile guided basic block placement algorithm.
 721   The idea is to put frequently executed blocks together at the start of the
 722   function and hopefully increase the number of fall-through conditional
 723   branches.  If there is no profile information for a particular function, this
 724   pass basically orders blocks in depth-first order.</p>
 725 </div>
 726
 727 <!-------------------------------------------------------------------------- -->
 728 <div class="doc_subsection">
 729   <a name="break-crit-edges">Break critical edges in CFG</a>
 730 </div>
 731 <div class="doc_text">
 732   <p>
 733   Break all of the critical edges in the CFG by inserting a dummy basic block.
 734   It may be "required" by passes that cannot deal with critical edges. This
 735   transformation obviously invalidates the CFG, but can update forward dominator
 736   (set, immediate dominators, tree, and frontier) information.
 737   </p>
 738 </div>
 739
 740 <!-------------------------------------------------------------------------- -->
 741 <div class="doc_subsection">
 742   <a name="cee">Correlated Expression Elimination</a>
 743 </div>
 744 <div class="doc_text">
 745   <p>Correlated Expression Elimination propagates information from conditional
 746   branches to blocks dominated by destinations of the branch.  It propagates
 747   information from the condition check itself into the body of the branch,
 748   allowing transformations like these for example:</p>
 749
 750 <blockquote><pre>
 751 if (i == 7)
 752   ... 4*i;  // constant propagation
 753
 754 M = i+1; N = j+1;
 755 if (i == j)
 756   X = M-N;  // = M-M == 0;
 757 </pre></blockquote>
 758
 759    <p>This is called Correlated Expression Elimination because we eliminate or
 760    simplify expressions that are correlated with the direction of a branch. In
 761    this way we use static information to give us some information about the
 762    dynamic value of a variable.</p>
 763 </div>
 764
 765 <!-------------------------------------------------------------------------- -->
 766 <div class="doc_subsection">
 767   <a name="condprop">Conditional Propagation</a>
 768 </div>
 769 <div class="doc_text">
 770   <p>This pass propagates information about conditional expressions through the
 771   program, allowing it to eliminate conditional branches in some cases.</p>
 772 </div>
 773
 774 <!-------------------------------------------------------------------------- -->
 775 <div class="doc_subsection">
 776   <a name="constmerge">Merge Duplicate Global Constants</a>
 777 </div>
 778 <div class="doc_text">
 779   <p>
 780   Merges duplicate global constants together into a single constant that is
 781   shared.  This is useful because some passes (ie TraceValues) insert a lot of
 782   string constants into the program, regardless of whether or not an existing
 783   string is available.
 784   </p>
 785 </div>
 786
 787 <!-------------------------------------------------------------------------- -->
 788 <div class="doc_subsection">
 789   <a name="constprop">Simple constant propagation</a>
 790 </div>
 791 <div class="doc_text">
 792   <p>This file implements constant propagation and merging. It looks for
 793   instructions involving only constant operands and replaces them with a
 794   constant value instead of an instruction. For example:</p>
 795   <blockquote><pre>add i32 1, 2</pre></blockquote>
 796   <p>becomes</p>
 797   <blockquote><pre>i32 3</pre></blockquote>
 798   <p>NOTE: this pass has a habit of making definitions be dead.  It is a good
 799   idea to to run a <a href="#die">DIE</a> (Dead Instruction Elimination) pass
 800   sometime after running this pass.</p>
 801 </div>
 802
 803 <!-------------------------------------------------------------------------- -->
 804 <div class="doc_subsection">
 805   <a name="dce">Dead Code Elimination</a>
 806 </div>
 807 <div class="doc_text">
 808   <p>
 809   Dead code elimination is similar to <a href="#die">dead instruction
 810   elimination</a>, but it rechecks instructions that were used by removed
 811   instructions to see if they are newly dead.
 812   </p>
 813 </div>
 814
 815 <!-------------------------------------------------------------------------- -->
 816 <div class="doc_subsection">
 817   <a name="deadargelim">Dead Argument Elimination</a>
 818 </div>
 819 <div class="doc_text">
 820   <p>
 821   This pass deletes dead arguments from internal functions.  Dead argument
 822   elimination removes arguments which are directly dead, as well as arguments
 823   only passed into function calls as dead arguments of other functions.  This
 824   pass also deletes dead arguments in a similar way.
 825   </p>
 826
 827   <p>
 828   This pass is often useful as a cleanup pass to run after aggressive
 829   interprocedural passes, which add possibly-dead arguments.
 830   </p>
 831 </div>
 832
 833 <!-------------------------------------------------------------------------- -->
 834 <div class="doc_subsection">
 835   <a name="deadtypeelim">Dead Type Elimination</a>
 836 </div>
 837 <div class="doc_text">
 838   <p>
 839   This pass is used to cleanup the output of GCC.  It eliminate names for types
 840   that are unused in the entire translation unit, using the <a
 841   href="#findusedtypes">find used types</a> pass.
 842   </p>
 843 </div>
 844
 845 <!-------------------------------------------------------------------------- -->
 846 <div class="doc_subsection">
 847   <a name="die">Dead Instruction Elimination</a>
 848 </div>
 849 <div class="doc_text">
 850   <p>
 851   Dead instruction elimination performs a single pass over the function,
 852   removing instructions that are obviously dead.
 853   </p>
 854 </div>
 855
 856 <!-------------------------------------------------------------------------- -->
 857 <div class="doc_subsection">
 858   <a name="dse">Dead Store Elimination</a>
 859 </div>
 860 <div class="doc_text">
 861   <p>
 862   A trivial dead store elimination that only considers basic-block local
 863   redundant stores.
 864   </p>
 865 </div>
 866
 867 <!-------------------------------------------------------------------------- -->
 868 <div class="doc_subsection">
 869   <a name="gcse">Global Common Subexpression Elimination</a>
 870 </div>
 871 <div class="doc_text">
 872   <p>
 873   This pass is designed to be a very quick global transformation that
 874   eliminates global common subexpressions from a function.  It does this by
 875   using an existing value numbering implementation to identify the common
 876   subexpressions, eliminating them when possible.
 877   </p>
 878 </div>
 879
 880 <!-------------------------------------------------------------------------- -->
 881 <div class="doc_subsection">
 882   <a name="globaldce">Dead Global Elimination</a>
 883 </div>
 884 <div class="doc_text">
 885   <p>
 886   This transform is designed to eliminate unreachable internal globals from the
 887   program.  It uses an aggressive algorithm, searching out globals that are
 888   known to be alive.  After it finds all of the globals which are needed, it
 889   deletes whatever is left over.  This allows it to delete recursive chunks of
 890   the program which are unreachable.
 891   </p>
 892 </div>
 893
 894 <!-------------------------------------------------------------------------- -->
 895 <div class="doc_subsection">
 896   <a name="globalopt">Global Variable Optimizer</a>
 897 </div>
 898 <div class="doc_text">
 899   <p>
 900   This pass transforms simple global variables that never have their address
 901   taken.  If obviously true, it marks read/write globals as constant, deletes
 902   variables only stored to, etc.
 903   </p>
 904 </div>
 905
 906 <!-------------------------------------------------------------------------- -->
 907 <div class="doc_subsection">
 908   <a name="gvn">Global Value Numbering</a>
 909 </div>
 910 <div class="doc_text">
 911   <p>
 912   This pass performs global value numbering to eliminate fully redundant
 913   instructions.  It also performs simple dead load elimination.
 914   </p>
 915 </div>
 916
 917 <!-------------------------------------------------------------------------- -->
 918 <div class="doc_subsection">
 919   <a name="gvnpre">Global Value Numbering/Partial Redundancy Elimination</a>
 920 </div>
 921 <div class="doc_text">
 922   <p>
 923   This pass performs a hybrid of global value numbering and partial redundancy
 924   elimination, known as GVN-PRE.  It performs partial redundancy elimination on
 925   values, rather than lexical expressions, allowing a more comprehensive view
 926   the optimization.  It replaces redundant values with uses of earlier
 927   occurences of the same value.  While this is beneficial in that it eliminates
 928   unneeded computation, it also increases register pressure by creating large
 929   live ranges, and should be used with caution on platforms that are very
 930   sensitive to register pressure.
 931   </p>
 932 </div>
 933
 934 <!-------------------------------------------------------------------------- -->
 935 <div class="doc_subsection">
 936   <a name="indmemrem">Indirect Malloc and Free Removal</a>
 937 </div>
 938 <div class="doc_text">
 939   <p>
 940   This pass finds places where memory allocation functions may escape into
 941   indirect land.  Some transforms are much easier (aka possible) only if free
 942   or malloc are not called indirectly.
 943   </p>
 944
 945   <p>
 946   Thus find places where the address of memory functions are taken and construct
 947   bounce functions with direct calls of those functions.
 948   </p>
 949 </div>
 950
 951 <!-------------------------------------------------------------------------- -->
 952 <div class="doc_subsection">
 953   <a name="indvars">Canonicalize Induction Variables</a>
 954 </div>
 955 <div class="doc_text">
 956   <p>
 957   This transformation analyzes and transforms the induction variables (and
 958   computations derived from them) into simpler forms suitable for subsequent
 959   analysis and transformation.
 960   </p>
 961
 962   <p>
 963   This transformation makes the following changes to each loop with an
 964   identifiable induction variable:
 965   </p>
 966
 967   <ol>
 968     <li>All loops are transformed to have a <em>single</em> canonical
 969         induction variable which starts at zero and steps by one.</li>
 970     <li>The canonical induction variable is guaranteed to be the first PHI node
 971         in the loop header block.</li>
 972     <li>Any pointer arithmetic recurrences are raised to use array
 973         subscripts.</li>
 974   </ol>
 975
 976   <p>
 977   If the trip count of a loop is computable, this pass also makes the following
 978   changes:
 979   </p>
 980
 981   <ol>
 982     <li>The exit condition for the loop is canonicalized to compare the
 983         induction value against the exit value.  This turns loops like:
 984         <blockquote><pre>for (i = 7; i*i < 1000; ++i)</pre></blockquote>
 985         into
 986         <blockquote><pre>for (i = 0; i != 25; ++i)</pre></blockquote></li>
 987     <li>Any use outside of the loop of an expression derived from the indvar
 988         is changed to compute the derived value outside of the loop, eliminating
 989         the dependence on the exit value of the induction variable.  If the only
 990         purpose of the loop is to compute the exit value of some derived
 991         expression, this transformation will make the loop dead.</li>
 992   </ol>
 993
 994   <p>
 995   This transformation should be followed by strength reduction after all of the
 996   desired loop transformations have been performed.  Additionally, on targets
 997   where it is profitable, the loop could be transformed to count down to zero
 998   (the "do loop" optimization).
 999   </p>
1000 </div>
1001
1002 <!-------------------------------------------------------------------------- -->
1003 <div class="doc_subsection">
1004   <a name="inline">Function Integration/Inlining</a>
1005 </div>
1006 <div class="doc_text">
1007   <p>
1008   Bottom-up inlining of functions into callees.
1009   </p>
1010 </div>
1011
1012 <!-------------------------------------------------------------------------- -->
1013 <div class="doc_subsection">
1014   <a name="insert-block-profiling">Insert instrumentation for block profiling</a>
1015 </div>
1016 <div class="doc_text">
1017   <p>
1018   This pass instruments the specified program with counters for basic block
1019   profiling, which counts the number of times each basic block executes.  This
1020   is the most basic form of profiling, which can tell which blocks are hot, but
1021   cannot reliably detect hot paths through the CFG.
1022   </p>
1023
1024   <p>
1025   Note that this implementation is very naïve.  Control equivalent regions of
1026   the CFG should not require duplicate counters, but it does put duplicate
1027   counters in.
1028   </p>
1029 </div>
1030
1031 <!-------------------------------------------------------------------------- -->
1032 <div class="doc_subsection">
1033   <a name="insert-edge-profiling">Insert instrumentation for edge profiling</a>
1034 </div>
1035 <div class="doc_text">
1036   <p>
1037   This pass instruments the specified program with counters for edge profiling.
1038   Edge profiling can give a reasonable approximation of the hot paths through a
1039   program, and is used for a wide variety of program transformations.
1040   </p>
1041
1042   <p>
1043   Note that this implementation is very naïve.  It inserts a counter for
1044   <em>every</em> edge in the program, instead of using control flow information
1045   to prune the number of counters inserted.
1046   </p>
1047 </div>
1048
1049 <!-------------------------------------------------------------------------- -->
1050 <div class="doc_subsection">
1051   <a name="insert-function-profiling">Insert instrumentation for function profiling</a>
1052 </div>
1053 <div class="doc_text">
1054   <p>
1055   This pass instruments the specified program with counters for function
1056   profiling, which counts the number of times each function is called.
1057   </p>
1058 </div>
1059
1060 <!-------------------------------------------------------------------------- -->
1061 <div class="doc_subsection">
1062   <a name="insert-null-profiling-rs">Measure profiling framework overhead</a>
1063 </div>
1064 <div class="doc_text">
1065   <p>
1066   The basic profiler that does nothing.  It is the default profiler and thus
1067   terminates <code>RSProfiler</code> chains.  It is useful for  measuring
1068   framework overhead.
1069   </p>
1070 </div>
1071
1072 <!-------------------------------------------------------------------------- -->
1073 <div class="doc_subsection">
1074   <a name="insert-rs-profiling-framework">Insert random sampling instrumentation framework</a>
1075 </div>
1076 <div class="doc_text">
1077   <p>
1078   The second stage of the random-sampling instrumentation framework, duplicates
1079   all instructions in a function, ignoring the profiling code, then connects the
1080   two versions together at the entry and at backedges.  At each connection point
1081   a choice is made as to whether to jump to the profiled code (take a sample) or
1082   execute the unprofiled code.
1083   </p>
1084
1085   <p>
1086   After this pass, it is highly recommended to run<a href="#mem2reg">mem2reg</a>
1087   and <a href="#adce">adce</a>. <a href="#instcombine">instcombine</a>,
1088   <a href="#load-vn">load-vn</a>, <a href="#gdce">gdce</a>, and
1089   <a href="#dse">dse</a> also are good to run afterwards.
1090   </p>
1091 </div>
1092
1093 <!-------------------------------------------------------------------------- -->
1094 <div class="doc_subsection">
1095   <a name="instcombine">Combine redundant instructions</a>
1096 </div>
1097 <div class="doc_text">
1098   <p>
1099   Combine instructions to form fewer, simple
1100   instructions.  This pass does not modify the CFG This pass is where algebraic
1101   simplification happens.
1102   </p>
1103
1104   <p>
1105   This pass combines things like:
1106   </p>
1107
1108 <blockquote><pre
1109 >%Y = add i32 %X, 1
1110 %Z = add i32 %Y, 1</pre></blockquote>
1111
1112   <p>
1113   into:
1114   </p>
1115
1116 <blockquote><pre
1117 >%Z = add i32 %X, 2</pre></blockquote>
1118
1119   <p>
1120   This is a simple worklist driven algorithm.
1121   </p>
1122
1123   <p>
1124   This pass guarantees that the following canonicalizations are performed on
1125   the program:
1126   </p>
1127
1128   <ul>
1129     <li>If a binary operator has a constant operand, it is moved to the right-
1130         hand side.</li>
1131     <li>Bitwise operators with constant operands are always grouped so that
1132         shifts are performed first, then <code>or</code>s, then
1133         <code>and</code>s, then <code>xor</code>s.</li>
1134     <li>Compare instructions are converted from <code>&lt;</code>,
1135         <code>&gt;</code>, <code>≤</code>, or <code>≥</code> to
1136         <code>=</code> or <code>≠</code> if possible.</li>
1137     <li>All <code>cmp</code> instructions on boolean values are replaced with
1138         logical operations.</li>
1139     <li><code>add <var>X</var>, <var>X</var></code> is represented as
1140         <code>mul <var>X</var>, 2</code> ⇒ <code>shl <var>X</var>, 1</code></li>
1141     <li>Multiplies with a constant power-of-two argument are transformed into
1142         shifts.</li>
1143     <li>… etc.</li>
1144   </ul>
1145 </div>
1146
1147 <!-------------------------------------------------------------------------- -->
1148 <div class="doc_subsection">
1149   <a name="internalize">Internalize Global Symbols</a>
1150 </div>
1151 <div class="doc_text">
1152   <p>
1153   This pass loops over all of the functions in the input module, looking for a
1154   main function.  If a main function is found, all other functions and all
1155   global variables with initializers are marked as internal.
1156   </p>
1157 </div>
1158
1159 <!-------------------------------------------------------------------------- -->
1160 <div class="doc_subsection">
1161   <a name="ipconstprop">Interprocedural constant propagation</a>
1162 </div>
1163 <div class="doc_text">
1164   <p>
1165   This pass implements an <em>extremely</em> simple interprocedural constant
1166   propagation pass.  It could certainly be improved in many different ways,
1167   like using a worklist.  This pass makes arguments dead, but does not remove
1168   them.  The existing dead argument elimination pass should be run after this
1169   to clean up the mess.
1170   </p>
1171 </div>
1172
1173 <!-------------------------------------------------------------------------- -->
1174 <div class="doc_subsection">
1175   <a name="ipsccp">Interprocedural Sparse Conditional Constant Propagation</a>
1176 </div>
1177 <div class="doc_text">
1178   <p>
1179   An interprocedural variant of <a href="#sccp">Sparse Conditional Constant
1180   Propagation</a>.
1181   </p>
1182 </div>
1183
1184 <!-------------------------------------------------------------------------- -->
1185 <div class="doc_subsection">
1186   <a name="lcssa">Loop-Closed SSA Form Pass</a>
1187 </div>
1188 <div class="doc_text">
1189   <p>
1190   This pass transforms loops by placing phi nodes at the end of the loops for
1191   all values that are live across the loop boundary.  For example, it turns
1192   the left into the right code:
1193   </p>
1194
1195   <pre
1196 >for (...)                for (...)
1197   if (c)                   if (c)
1198     X1 = ...                 X1 = ...
1199   else                     else
1200     X2 = ...                 X2 = ...
1201   X3 = phi(X1, X2)         X3 = phi(X1, X2)
1202 ... = X3 + 4              X4 = phi(X3)
1203                           ... = X4 + 4</pre>
1204
1205   <p>
1206   This is still valid LLVM; the extra phi nodes are purely redundant, and will
1207   be trivially eliminated by <code>InstCombine</code>.  The major benefit of
1208   this transformation is that it makes many other loop optimizations, such as
1209   LoopUnswitching, simpler.
1210   </p>
1211 </div>
1212
1213 <!-------------------------------------------------------------------------- -->
1214 <div class="doc_subsection">
1215   <a name="licm">Loop Invariant Code Motion</a>
1216 </div>
1217 <div class="doc_text">
1218   <p>
1219   This pass performs loop invariant code motion, attempting to remove as much
1220   code from the body of a loop as possible.  It does this by either hoisting
1221   code into the preheader block, or by sinking code to the exit blocks if it is
1222   safe.  This pass also promotes must-aliased memory locations in the loop to
1223   live in registers, thus hoisting and sinking "invariant" loads and stores.
1224   </p>
1225
1226   <p>
1227   This pass uses alias analysis for two purposes:
1228   </p>
1229
1230   <ul>
1231     <li>Moving loop invariant loads and calls out of loops.  If we can determine
1232         that a load or call inside of a loop never aliases anything stored to,
1233         we can hoist it or sink it like any other instruction.</li>
1234     <li>Scalar Promotion of Memory - If there is a store instruction inside of
1235         the loop, we try to move the store to happen AFTER the loop instead of
1236         inside of the loop.  This can only happen if a few conditions are true:
1237         <ul>
1238           <li>The pointer stored through is loop invariant.</li>
1239           <li>There are no stores or loads in the loop which <em>may</em> alias
1240               the pointer.  There are no calls in the loop which mod/ref the
1241               pointer.</li>
1242         </ul>
1243         If these conditions are true, we can promote the loads and stores in the
1244         loop of the pointer to use a temporary alloca'd variable.  We then use
1245         the mem2reg functionality to construct the appropriate SSA form for the
1246         variable.</li>
1247   </ul>
1248 </div>
1249
1250 <!-------------------------------------------------------------------------- -->
1251 <div class="doc_subsection">
1252   <a name="loop-extract">Extract loops into new functions</a>
1253 </div>
1254 <div class="doc_text">
1255   <p>
1256   A pass wrapper around the <code>ExtractLoop()</code> scalar transformation to
1257   extract each top-level loop into its own new function. If the loop is the
1258   <em>only</em> loop in a given function, it is not touched. This is a pass most
1259   useful for debugging via bugpoint.
1260   </p>
1261 </div>
1262
1263 <!-------------------------------------------------------------------------- -->
1264 <div class="doc_subsection">
1265   <a name="loop-extract-single">Extract at most one loop into a new function</a>
1266 </div>
1267 <div class="doc_text">
1268   <p>
1269   Similar to <a href="#loop-extract">Extract loops into new functions</a>,
1270   this pass extracts one natural loop from the program into a function if it
1271   can. This is used by bugpoint.
1272   </p>
1273 </div>
1274
1275 <!-------------------------------------------------------------------------- -->
1276 <div class="doc_subsection">
1277   <a name="loop-index-split">Index Split Loops</a>
1278 </div>
1279 <div class="doc_text">
1280   <p>
1281   This pass divides loop's iteration range by spliting loop such that each
1282   individual loop is executed efficiently.
1283   </p>
1284 </div>
1285
1286 <!-------------------------------------------------------------------------- -->
1287 <div class="doc_subsection">
1288   <a name="loop-reduce">Loop Strength Reduction</a>
1289 </div>
1290 <div class="doc_text">
1291   <p>
1292   This pass performs a strength reduction on array references inside loops that
1293   have as one or more of their components the loop induction variable.  This is
1294   accomplished by creating a new value to hold the initial value of the array
1295   access for the first iteration, and then creating a new GEP instruction in
1296   the loop to increment the value by the appropriate amount.
1297   </p>
1298 </div>
1299
1300 <!-------------------------------------------------------------------------- -->
1301 <div class="doc_subsection">
1302   <a name="loop-rotate">Rotate Loops</a>
1303 </div>
1304 <div class="doc_text">
1305   <p>A simple loop rotation transformation.</p>
1306 </div>
1307
1308 <!-------------------------------------------------------------------------- -->
1309 <div class="doc_subsection">
1310   <a name="loop-unroll">Unroll loops</a>
1311 </div>
1312 <div class="doc_text">
1313   <p>
1314   This pass implements a simple loop unroller.  It works best when loops have
1315   been canonicalized by the <a href="#indvars"><tt>-indvars</tt></a> pass,
1316   allowing it to determine the trip counts of loops easily.
1317   </p>
1318 </div>
1319
1320 <!-------------------------------------------------------------------------- -->
1321 <div class="doc_subsection">
1322   <a name="loop-unswitch">Unswitch loops</a>
1323 </div>
1324 <div class="doc_text">
1325   <p>
1326   This pass transforms loops that contain branches on loop-invariant conditions
1327   to have multiple loops.  For example, it turns the left into the right code:
1328   </p>
1329
1330   <pre
1331 >for (...)                  if (lic)
1332   A                          for (...)
1333   if (lic)                     A; B; C
1334     B                      else
1335   C                          for (...)
1336                                A; C</pre>
1337
1338   <p>
1339   This can increase the size of the code exponentially (doubling it every time
1340   a loop is unswitched) so we only unswitch if the resultant code will be
1341   smaller than a threshold.
1342   </p>
1343
1344   <p>
1345   This pass expects LICM to be run before it to hoist invariant conditions out
1346   of the loop, to make the unswitching opportunity obvious.
1347   </p>
1348 </div>
1349
1350 <!-------------------------------------------------------------------------- -->
1351 <div class="doc_subsection">
1352   <a name="loopsimplify">Canonicalize natural loops</a>
1353 </div>
1354 <div class="doc_text">
1355   <p>
1356   This pass performs several transformations to transform natural loops into a
1357   simpler form, which makes subsequent analyses and transformations simpler and
1358   more effective.
1359   </p>
1360
1361   <p>
1362   Loop pre-header insertion guarantees that there is a single, non-critical
1363   entry edge from outside of the loop to the loop header.  This simplifies a
1364   number of analyses and transformations, such as LICM.
1365   </p>
1366
1367   <p>
1368   Loop exit-block insertion guarantees that all exit blocks from the loop
1369   (blocks which are outside of the loop that have predecessors inside of the
1370   loop) only have predecessors from inside of the loop (and are thus dominated
1371   by the loop header).  This simplifies transformations such as store-sinking
1372   that are built into LICM.
1373   </p>
1374
1375   <p>
1376   This pass also guarantees that loops will have exactly one backedge.
1377   </p>
1378
1379   <p>
1380   Note that the simplifycfg pass will clean up blocks which are split out but
1381   end up being unnecessary, so usage of this pass should not pessimize
1382   generated code.
1383   </p>
1384
1385   <p>
1386   This pass obviously modifies the CFG, but updates loop information and
1387   dominator information.
1388   </p>
1389 </div>
1390
1391 <!-------------------------------------------------------------------------- -->
1392 <div class="doc_subsection">
1393   <a name="lower-packed">lowers packed operations to operations on smaller packed datatypes</a>
1394 </div>
1395 <div class="doc_text">
1396   <p>
1397   Lowers operations on vector datatypes into operations on more primitive vector
1398   datatypes, and finally to scalar operations.
1399   </p>
1400 </div>
1401
1402 <!-------------------------------------------------------------------------- -->
1403 <div class="doc_subsection">
1404   <a name="lowerallocs">Lower allocations from instructions to calls</a>
1405 </div>
1406 <div class="doc_text">
1407   <p>
1408   Turn <tt>malloc</tt> and <tt>free</tt> instructions into <tt>@malloc</tt> and
1409   <tt>@free</tt> calls.
1410   </p>
1411
1412   <p>
1413   This is a target-dependent tranformation because it depends on the size of
1414   data types and alignment constraints.
1415   </p>
1416 </div>
1417
1418 <!-------------------------------------------------------------------------- -->
1419 <div class="doc_subsection">
1420   <a name="lowergc">Lower GC intrinsics, for GCless code generators</a>
1421 </div>
1422 <div class="doc_text">
1423   <p>
1424   This file implements lowering for the <tt>llvm.gc*</tt> intrinsics for targets
1425   that do not natively support them (which includes the C backend).  Note that
1426   the code generated is not as efficient as it would be for targets that
1427   natively support the GC intrinsics, but it is useful for getting new targets
1428   up-and-running quickly.
1429   </p>
1430
1431   <p>
1432   This pass implements the code transformation described in this paper:
1433   </p>
1434
1435   <blockquote><p>
1436     "Accurate Garbage Collection in an Uncooperative Environment"
1437     Fergus Henderson, ISMM, 2002
1438   </p></blockquote>
1439 </div>
1440
1441 <!-------------------------------------------------------------------------- -->
1442 <div class="doc_subsection">
1443   <a name="lowerinvoke">Lower invoke and unwind, for unwindless code generators</a>
1444 </div>
1445 <div class="doc_text">
1446   <p>
1447   This transformation is designed for use by code generators which do not yet
1448   support stack unwinding.  This pass supports two models of exception handling
1449   lowering, the 'cheap' support and the 'expensive' support.
1450   </p>
1451
1452   <p>
1453   'Cheap' exception handling support gives the program the ability to execute
1454   any program which does not "throw an exception", by turning 'invoke'
1455   instructions into calls and by turning 'unwind' instructions into calls to
1456   abort().  If the program does dynamically use the unwind instruction, the
1457   program will print a message then abort.
1458   </p>
1459
1460   <p>
1461   'Expensive' exception handling support gives the full exception handling
1462   support to the program at the cost of making the 'invoke' instruction
1463   really expensive.  It basically inserts setjmp/longjmp calls to emulate the
1464   exception handling as necessary.
1465   </p>
1466
1467   <p>
1468   Because the 'expensive' support slows down programs a lot, and EH is only
1469   used for a subset of the programs, it must be specifically enabled by the
1470   <tt>-enable-correct-eh-support</tt> option.
1471   </p>
1472
1473   <p>
1474   Note that after this pass runs the CFG is not entirely accurate (exceptional
1475   control flow edges are not correct anymore) so only very simple things should
1476   be done after the lowerinvoke pass has run (like generation of native code).
1477   This should not be used as a general purpose "my LLVM-to-LLVM pass doesn't
1478   support the invoke instruction yet" lowering pass.
1479   </p>
1480 </div>
1481
1482 <!-------------------------------------------------------------------------- -->
1483 <div class="doc_subsection">
1484   <a name="lowerselect">Lower select instructions to branches</a>
1485 </div>
1486 <div class="doc_text">
1487   <p>
1488   Lowers select instructions into conditional branches for targets that do not
1489   have conditional moves or that have not implemented the select instruction
1490   yet.
1491   </p>
1492
1493   <p>
1494   Note that this pass could be improved.  In particular it turns every select
1495   instruction into a new conditional branch, even though some common cases have
1496   select instructions on the same predicate next to each other.  It would be
1497   better to use the same branch for the whole group of selects.
1498   </p>
1499 </div>
1500
1501 <!-------------------------------------------------------------------------- -->
1502 <div class="doc_subsection">
1503   <a name="lowersetjmp">Lower Set Jump</a>
1504 </div>
1505 <div class="doc_text">
1506   <p>
1507    Lowers <tt>setjmp</tt> and <tt>longjmp</tt> to use the LLVM invoke and unwind
1508    instructions as necessary.
1509   </p>
1510
1511   <p>
1512    Lowering of <tt>longjmp</tt> is fairly trivial. We replace the call with a
1513    call to the LLVM library function <tt>__llvm_sjljeh_throw_longjmp()</tt>.
1514    This unwinds the stack for us calling all of the destructors for
1515    objects allocated on the stack.
1516   </p>
1517
1518   <p>
1519    At a <tt>setjmp</tt> call, the basic block is split and the <tt>setjmp</tt>
1520    removed. The calls in a function that have a <tt>setjmp</tt> are converted to
1521    invoke where the except part checks to see if it's a <tt>longjmp</tt>
1522    exception and, if so, if it's handled in the function. If it is, then it gets
1523    the value returned by the <tt>longjmp</tt> and goes to where the basic block
1524    was split. <tt>invoke</tt> instructions are handled in a similar fashion with
1525    the original except block being executed if it isn't a <tt>longjmp</tt>
1526    except that is handled by that function.
1527   </p>
1528 </div>
1529
1530 <!-------------------------------------------------------------------------- -->
1531 <div class="doc_subsection">
1532   <a name="lowerswitch">Lower SwitchInst's to branches</a>
1533 </div>
1534 <div class="doc_text">
1535   <p>
1536   Rewrites <tt>switch</tt> instructions with a sequence of branches, which
1537   allows targets to get away with not implementing the switch instruction until
1538   it is convenient.
1539   </p>
1540 </div>
1541
1542 <!-------------------------------------------------------------------------- -->
1543 <div class="doc_subsection">
1544   <a name="mem2reg">Promote Memory to Register</a>
1545 </div>
1546 <div class="doc_text">
1547   <p>
1548   This file promotes memory references to be register references.  It promotes
1549   <tt>alloca</tt> instructions which only have <tt>load</tt>s and
1550   <tt>store</tt>s as uses.  An <tt>alloca</tt> is transformed by using dominator
1551   frontiers to place <tt>phi</tt> nodes, then traversing the function in
1552   depth-first order to rewrite <tt>load</tt>s and <tt>store</tt>s as
1553   appropriate. This is just the standard SSA construction algorithm to construct
1554   "pruned" SSA form.
1555   </p>
1556 </div>
1557
1558 <!-------------------------------------------------------------------------- -->
1559 <div class="doc_subsection">
1560   <a name="mergereturn">Unify function exit nodes</a>
1561 </div>
1562 <div class="doc_text">
1563   <p>
1564   Ensure that functions have at most one <tt>ret</tt> instruction in them.
1565   Additionally, it keeps track of which node is the new exit node of the CFG.
1566   </p>
1567 </div>
1568
1569 <!-------------------------------------------------------------------------- -->
1570 <div class="doc_subsection">
1571   <a name="predsimplify">Predicate Simplifier</a>
1572 </div>
1573 <div class="doc_text">
1574   <p>
1575   Path-sensitive optimizer. In a branch where <tt>x == y</tt>, replace uses of
1576   <tt>x</tt> with <tt>y</tt>. Permits further optimization, such as the
1577   elimination of the unreachable call:
1578   </p>
1579
1580 <blockquote><pre
1581 >void test(int *p, int *q)
1582 {
1583   if (p != q)
1584     return;
1585
1586   if (*p != *q)
1587     foo(); // unreachable
1588 }</pre></blockquote>
1589 </div>
1590
1591 <!-------------------------------------------------------------------------- -->
1592 <div class="doc_subsection">
1593   <a name="prune-eh">Remove unused exception handling info</a>
1594 </div>
1595 <div class="doc_text">
1596   <p>
1597   This file implements a simple interprocedural pass which walks the call-graph,
1598   turning <tt>invoke</tt> instructions into <tt>call</tt> instructions if and
1599   only if the callee cannot throw an exception. It implements this as a
1600   bottom-up traversal of the call-graph.
1601   </p>
1602 </div>
1603
1604 <!-------------------------------------------------------------------------- -->
1605 <div class="doc_subsection">
1606   <a name="raiseallocs">Raise allocations from calls to instructions</a>
1607 </div>
1608 <div class="doc_text">
1609   <p>
1610   Converts <tt>@malloc</tt> and <tt>@free</tt> calls to <tt>malloc</tt> and
1611   <tt>free</tt> instructions.
1612   </p>
1613 </div>
1614
1615 <!-------------------------------------------------------------------------- -->
1616 <div class="doc_subsection">
1617   <a name="reassociate">Reassociate expressions</a>
1618 </div>
1619 <div class="doc_text">
1620   <p>
1621   This pass reassociates commutative expressions in an order that is designed
1622   to promote better constant propagation, GCSE, LICM, PRE, etc.
1623   </p>
1624
1625   <p>
1626   For example: 4 + (<var>x</var> + 5) ⇒ <var>x</var> + (4 + 5)
1627   </p>
1628
1629   <p>
1630   In the implementation of this algorithm, constants are assigned rank = 0,
1631   function arguments are rank = 1, and other values are assigned ranks
1632   corresponding to the reverse post order traversal of current function
1633   (starting at 2), which effectively gives values in deep loops higher rank
1634   than values not in loops.
1635   </p>
1636 </div>
1637
1638 <!-------------------------------------------------------------------------- -->
1639 <div class="doc_subsection">
1640   <a name="reg2mem">Demote all values to stack slots</a>
1641 </div>
1642 <div class="doc_text">
1643   <p>
1644   This file demotes all registers to memory references.  It is intented to be
1645   the inverse of <a href="#mem2reg"><tt>-mem2reg</tt></a>.  By converting to
1646   <tt>load</tt> instructions, the only values live accross basic blocks are
1647   <tt>alloca</tt> instructions and <tt>load</tt> instructions before
1648   <tt>phi</tt> nodes. It is intended that this should make CFG hacking much
1649   easier. To make later hacking easier, the entry block is split into two, such
1650   that all introduced <tt>alloca</tt> instructions (and nothing else) are in the
1651   entry block.
1652   </p>
1653 </div>
1654
1655 <!-------------------------------------------------------------------------- -->
1656 <div class="doc_subsection">
1657   <a name="scalarrepl">Scalar Replacement of Aggregates</a>
1658 </div>
1659 <div class="doc_text">
1660   <p>
1661   The well-known scalar replacement of aggregates transformation.  This
1662   transform breaks up <tt>alloca</tt> instructions of aggregate type (structure
1663   or array) into individual <tt>alloca</tt> instructions for each member if
1664   possible.  Then, if possible, it transforms the individual <tt>alloca</tt>
1665   instructions into nice clean scalar SSA form.
1666   </p>
1667
1668   <p>
1669   This combines a simple scalar replacement of aggregates algorithm with the <a
1670   href="#mem2reg"><tt>mem2reg</tt></a> algorithm because often interact,
1671   especially for C++ programs.  As such, iterating between <tt>scalarrepl</tt>,
1672   then <a href="#mem2reg"><tt>mem2reg</tt></a> until we run out of things to
1673   promote works well.
1674   </p>
1675 </div>
1676
1677 <!-------------------------------------------------------------------------- -->
1678 <div class="doc_subsection">
1679   <a name="sccp">Sparse Conditional Constant Propagation</a>
1680 </div>
1681 <div class="doc_text">
1682   <p>
1683   Sparse conditional constant propagation and merging, which can be summarized
1684   as:
1685   </p>
1686
1687   <ol>
1688     <li>Assumes values are constant unless proven otherwise</li>
1689     <li>Assumes BasicBlocks are dead unless proven otherwise</li>
1690     <li>Proves values to be constant, and replaces them with constants</li>
1691     <li>Proves conditional branches to be unconditional</li>
1692   </ol>
1693
1694   <p>
1695   Note that this pass has a habit of making definitions be dead.  It is a good
1696   idea to to run a DCE pass sometime after running this pass.
1697   </p>
1698 </div>
1699
1700 <!-------------------------------------------------------------------------- -->
1701 <div class="doc_subsection">
1702   <a name="simplify-libcalls">Simplify well-known library calls</a>
1703 </div>
1704 <div class="doc_text">
1705   <p>
1706   Applies a variety of small optimizations for calls to specific well-known
1707   function calls (e.g. runtime library functions). For example, a call
1708    <tt>exit(3)</tt> that occurs within the <tt>main()</tt> function can be
1709    transformed into simply <tt>return 3</tt>.
1710   </p>
1711 </div>
1712
1713 <!-------------------------------------------------------------------------- -->
1714 <div class="doc_subsection">
1715   <a name="simplifycfg">Simplify the CFG</a>
1716 </div>
1717 <div class="doc_text">
1718   <p>
1719   Performs dead code elimination and basic block merging. Specifically:
1720   </p>
1721
1722   <ol>
1723     <li>Removes basic blocks with no predecessors.</li>
1724     <li>Merges a basic block into its predecessor if there is only one and the
1725         predecessor only has one successor.</li>
1726     <li>Eliminates PHI nodes for basic blocks with a single predecessor.</li>
1727     <li>Eliminates a basic block that only contains an unconditional
1728         branch.</li>
1729   </ol>
1730 </div>
1731
1732 <!-------------------------------------------------------------------------- -->
1733 <div class="doc_subsection">
1734   <a name="strip">Strip all symbols from a module</a>
1735 </div>
1736 <div class="doc_text">
1737   <p>
1738   Performs code stripping. This transformation can delete:
1739   </p>
1740
1741   <ol>
1742     <li>names for virtual registers</li>
1743     <li>symbols for internal globals and functions</li>
1744     <li>debug information</li>
1745   </ol>
1746
1747   <p>
1748   Note that this transformation makes code much less readable, so it should
1749   only be used in situations where the <tt>strip</tt> utility would be used,
1750   such as reducing code size or making it harder to reverse engineer code.
1751   </p>
1752 </div>
1753
1754 <!-------------------------------------------------------------------------- -->
1755 <div class="doc_subsection">
1756   <a name="tailcallelim">Tail Call Elimination</a>
1757 </div>
1758 <div class="doc_text">
1759   <p>
1760   This file transforms calls of the current function (self recursion) followed
1761   by a return instruction with a branch to the entry of the function, creating
1762   a loop.  This pass also implements the following extensions to the basic
1763   algorithm:
1764   </p>
1765
1766   <ul>
1767   <li>Trivial instructions between the call and return do not prevent the
1768       transformation from taking place, though currently the analysis cannot
1769       support moving any really useful instructions (only dead ones).
1770   <li>This pass transforms functions that are prevented from being tail
1771       recursive by an associative expression to use an accumulator variable,
1772       thus compiling the typical naive factorial or <tt>fib</tt> implementation
1773       into efficient code.
1774   <li>TRE is performed if the function returns void, if the return
1775       returns the result returned by the call, or if the function returns a
1776       run-time constant on all exits from the function.  It is possible, though
1777       unlikely, that the return returns something else (like constant 0), and
1778       can still be TRE'd.  It can be TRE'd if <em>all other</em> return
1779       instructions in the function return the exact same value.
1780   <li>If it can prove that callees do not access theier caller stack frame,
1781       they are marked as eligible for tail call elimination (by the code
1782       generator).
1783   </ul>
1784 </div>
1785
1786 <!-------------------------------------------------------------------------- -->
1787 <div class="doc_subsection">
1788   <a name="tailduplicate">Tail Duplication</a>
1789 </div>
1790 <div class="doc_text">
1791   <p>
1792   This pass performs a limited form of tail duplication, intended to simplify
1793   CFGs by removing some unconditional branches.  This pass is necessary to
1794   straighten out loops created by the C front-end, but also is capable of
1795   making other code nicer.  After this pass is run, the CFG simplify pass
1796   should be run to clean up the mess.
1797   </p>
1798 </div>
1799
1800 <!-- ======================================================================= -->
1801 <div class="doc_section"> <a name="transform">Utility Passes</a></div>
1802 <div class="doc_text">
1803   <p>This section describes the LLVM Utility Passes.</p>
1804 </div>
1805
1806 <!-------------------------------------------------------------------------- -->
1807 <div class="doc_subsection">
1808   <a name="deadarghaX0r">Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)</a>
1809 </div>
1810 <div class="doc_text">
1811   <p>
1812   Same as dead argument elimination, but deletes arguments to functions which
1813   are external.  This is only for use by <a
1814   href="Bugpoint.html">bugpoint</a>.</p>
1815 </div>
1816
1817 <!-------------------------------------------------------------------------- -->
1818 <div class="doc_subsection">
1819   <a name="extract-blocks">Extract Basic Blocks From Module (for bugpoint use)</a>
1820 </div>
1821 <div class="doc_text">
1822   <p>
1823   This pass is used by bugpoint to extract all blocks from the module into their
1824   own functions.</p>
1825 </div>
1826
1827 <!-------------------------------------------------------------------------- -->
1828 <div class="doc_subsection">
1829   <a name="preverify">Preliminary module verification</a>
1830 </div>
1831 <div class="doc_text">
1832   <p>
1833   Ensures that the module is in the form required by the <a
1834   href="#verifier">Module Verifier</a> pass.
1835   </p>
1836
1837   <p>
1838   Running the verifier runs this pass automatically, so there should be no need
1839   to use it directly.
1840   </p>
1841 </div>
1842
1843 <!-------------------------------------------------------------------------- -->
1844 <div class="doc_subsection">
1845   <a name="verify">Module Verifier</a>
1846 </div>
1847 <div class="doc_text">
1848   <p>
1849   Verifies an LLVM IR code. This is useful to run after an optimization which is
1850   undergoing testing. Note that <tt>llvm-as</tt> verifies its input before
1851   emitting bitcode, and also that malformed bitcode is likely to make LLVM
1852   crash. All language front-ends are therefore encouraged to verify their output
1853   before performing optimizing transformations.
1854   </p>
1855
1856   <ul>
1857     <li>Both of a binary operator's parameters are of the same type.</li>
1858     <li>Verify that the indices of mem access instructions match other
1859         operands.</li>
1860     <li>Verify that arithmetic and other things are only performed on
1861         first-class types.  Verify that shifts and logicals only happen on
1862         integrals f.e.</li>
1863     <li>All of the constants in a switch statement are of the correct type.</li>
1864     <li>The code is in valid SSA form.</li>
1865     <li>It should be illegal to put a label into any other type (like a
1866         structure) or to return one. [except constant arrays!]</li>
1867     <li>Only phi nodes can be self referential: <tt>%x = add int %x, %x</tt> is
1868         invalid.</li>
1869     <li>PHI nodes must have an entry for each predecessor, with no extras.</li>
1870     <li>PHI nodes must be the first thing in a basic block, all grouped
1871         together.</li>
1872     <li>PHI nodes must have at least one entry.</li>
1873     <li>All basic blocks should only end with terminator insts, not contain
1874         them.</li>
1875     <li>The entry node to a function must not have predecessors.</li>
1876     <li>All Instructions must be embedded into a basic block.</li>
1877     <li>Functions cannot take a void-typed parameter.</li>
1878     <li>Verify that a function's argument list agrees with its declared
1879         type.</li>
1880     <li>It is illegal to specify a name for a void value.</li>
1881     <li>It is illegal to have a internal global value with no initializer.</li>
1882     <li>It is illegal to have a ret instruction that returns a value that does
1883         not agree with the function return value type.</li>
1884     <li>Function call argument types match the function prototype.</li>
1885     <li>All other things that are tested by asserts spread about the code.</li>
1886   </ul>
1887
1888   <p>
1889   Note that this does not provide full security verification (like Java), but
1890   instead just tries to ensure that code is well-formed.
1891   </p>
1892 </div>
1893
1894 <!-------------------------------------------------------------------------- -->
1895 <div class="doc_subsection">
1896   <a name="view-cfg">View CFG of function</a>
1897 </div>
1898 <div class="doc_text">
1899   <p>
1900   Displays the control flow graph using the GraphViz tool.
1901   </p>
1902 </div>
1903
1904 <!-------------------------------------------------------------------------- -->
1905 <div class="doc_subsection">
1906   <a name="view-cfg-only">View CFG of function (with no function bodies)</a>
1907 </div>
1908 <div class="doc_text">
1909   <p>
1910   Displays the control flow graph using the GraphViz tool, but omitting function
1911   bodies.
1912   </p>
1913 </div>
1914
1915 <!-- *********************************************************************** -->
1916
1917 <hr>
1918 <address>
1919   <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1920   src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1921   <a href="http://validator.w3.org/check/referer"><img
1922   src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
1923
1924   <a href="mailto:rspencer@x10sys.com">Reid Spencer</a><br>
1925   <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
1926   Last modified: $Date$
1927 </address>
1928
1929 </body>
1930 </html>