• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1..
2    If Passes.html is up to date, the following "one-liner" should print
3    an empty diff.
4
5    egrep -e '^<tr><td><a href="#.*">-.*</a></td><td>.*</td></tr>$' \
6          -e '^  <a name=".*">.*</a>$' < Passes.html >html; \
7    perl >help <<'EOT' && diff -u help html; rm -f help html
8    open HTML, "<Passes.html" or die "open: Passes.html: $!\n";
9    while (<HTML>) {
10      m:^<tr><td><a href="#(.*)">-.*</a></td><td>.*</td></tr>$: or next;
11      $order{$1} = sprintf("%03d", 1 + int %order);
12    }
13    open HELP, "../Release/bin/opt -help|" or die "open: opt -help: $!\n";
14    while (<HELP>) {
15      m:^    -([^ ]+) +- (.*)$: or next;
16      my $o = $order{$1};
17      $o = "000" unless defined $o;
18      push @x, "$o<tr><td><a href=\"#$1\">-$1</a></td><td>$2</td></tr>\n";
19      push @y, "$o  <a name=\"$1\">-$1: $2</a>\n";
20    }
21    @x = map { s/^\d\d\d//; $_ } sort @x;
22    @y = map { s/^\d\d\d//; $_ } sort @y;
23    print @x, @y;
24    EOT
25
26    This (real) one-liner can also be helpful when converting comments to HTML:
27
28    perl -e '$/ = 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'
29
30====================================
31LLVM's Analysis and Transform Passes
32====================================
33
34.. contents::
35    :local:
36
37Introduction
38============
39
40This document serves as a high level summary of the optimization features that
41LLVM provides.  Optimizations are implemented as Passes that traverse some
42portion of a program to either collect information or transform the program.
43The table below divides the passes that LLVM provides into three categories.
44Analysis passes compute information that other passes can use or for debugging
45or program visualization purposes.  Transform passes can use (or invalidate)
46the analysis passes.  Transform passes all mutate the program in some way.
47Utility passes provides some utility but don't otherwise fit categorization.
48For example passes to extract functions to bitcode or write a module to bitcode
49are neither analysis nor transform passes.  The table of contents above
50provides a quick summary of each pass and links to the more complete pass
51description later in the document.
52
53Analysis Passes
54===============
55
56This section describes the LLVM Analysis Passes.
57
58``-aa-eval``: Exhaustive Alias Analysis Precision Evaluator
59-----------------------------------------------------------
60
61This is a simple N^2 alias analysis accuracy evaluator.  Basically, for each
62function in the program, it simply queries to see how the alias analysis
63implementation answers alias queries between each pair of pointers in the
64function.
65
66This is inspired and adapted from code by: Naveen Neelakantam, Francesco
67Spadini, and Wojciech Stryjewski.
68
69``-basicaa``: Basic Alias Analysis (stateless AA impl)
70------------------------------------------------------
71
72A basic alias analysis pass that implements identities (two different globals
73cannot alias, etc), but does no stateful analysis.
74
75``-basiccg``: Basic CallGraph Construction
76------------------------------------------
77
78Yet to be written.
79
80``-count-aa``: Count Alias Analysis Query Responses
81---------------------------------------------------
82
83A pass which can be used to count how many alias queries are being made and how
84the alias analysis implementation being used responds.
85
86``-da``: Dependence Analysis
87----------------------------
88
89Dependence analysis framework, which is used to detect dependences in memory
90accesses.
91
92``-debug-aa``: AA use debugger
93------------------------------
94
95This simple pass checks alias analysis users to ensure that if they create a
96new value, they do not query AA without informing it of the value.  It acts as
97a shim over any other AA pass you want.
98
99Yes keeping track of every value in the program is expensive, but this is a
100debugging pass.
101
102``-domfrontier``: Dominance Frontier Construction
103-------------------------------------------------
104
105This pass is a simple dominator construction algorithm for finding forward
106dominator frontiers.
107
108``-domtree``: Dominator Tree Construction
109-----------------------------------------
110
111This pass is a simple dominator construction algorithm for finding forward
112dominators.
113
114
115``-dot-callgraph``: Print Call Graph to "dot" file
116--------------------------------------------------
117
118This pass, only available in ``opt``, prints the call graph into a ``.dot``
119graph.  This graph can then be processed with the "dot" tool to convert it to
120postscript or some other suitable format.
121
122``-dot-cfg``: Print CFG of function to "dot" file
123-------------------------------------------------
124
125This pass, only available in ``opt``, prints the control flow graph into a
126``.dot`` graph.  This graph can then be processed with the :program:`dot` tool
127to convert it to postscript or some other suitable format.
128
129``-dot-cfg-only``: Print CFG of function to "dot" file (with no function bodies)
130--------------------------------------------------------------------------------
131
132This pass, only available in ``opt``, prints the control flow graph into a
133``.dot`` graph, omitting the function bodies.  This graph can then be processed
134with the :program:`dot` tool to convert it to postscript or some other suitable
135format.
136
137``-dot-dom``: Print dominance tree of function to "dot" file
138------------------------------------------------------------
139
140This pass, only available in ``opt``, prints the dominator tree into a ``.dot``
141graph.  This graph can then be processed with the :program:`dot` tool to
142convert it to postscript or some other suitable format.
143
144``-dot-dom-only``: Print dominance tree of function to "dot" file (with no function bodies)
145-------------------------------------------------------------------------------------------
146
147This pass, only available in ``opt``, prints the dominator tree into a ``.dot``
148graph, omitting the function bodies.  This graph can then be processed with the
149:program:`dot` tool to convert it to postscript or some other suitable format.
150
151``-dot-postdom``: Print postdominance tree of function to "dot" file
152--------------------------------------------------------------------
153
154This pass, only available in ``opt``, prints the post dominator tree into a
155``.dot`` graph.  This graph can then be processed with the :program:`dot` tool
156to convert it to postscript or some other suitable format.
157
158``-dot-postdom-only``: Print postdominance tree of function to "dot" file (with no function bodies)
159---------------------------------------------------------------------------------------------------
160
161This pass, only available in ``opt``, prints the post dominator tree into a
162``.dot`` graph, omitting the function bodies.  This graph can then be processed
163with the :program:`dot` tool to convert it to postscript or some other suitable
164format.
165
166``-globalsmodref-aa``: Simple mod/ref analysis for globals
167----------------------------------------------------------
168
169This simple pass provides alias and mod/ref information for global values that
170do not have their address taken, and keeps track of whether functions read or
171write memory (are "pure").  For this simple (but very common) case, we can
172provide pretty accurate and useful information.
173
174``-instcount``: Counts the various types of ``Instruction``\ s
175--------------------------------------------------------------
176
177This pass collects the count of all instructions and reports them.
178
179``-intervals``: Interval Partition Construction
180-----------------------------------------------
181
182This analysis calculates and represents the interval partition of a function,
183or a preexisting interval partition.
184
185In this way, the interval partition may be used to reduce a flow graph down to
186its degenerate single node interval partition (unless it is irreducible).
187
188``-iv-users``: Induction Variable Users
189---------------------------------------
190
191Bookkeeping for "interesting" users of expressions computed from induction
192variables.
193
194``-lazy-value-info``: Lazy Value Information Analysis
195-----------------------------------------------------
196
197Interface for lazy computation of value constraint information.
198
199``-libcall-aa``: LibCall Alias Analysis
200---------------------------------------
201
202LibCall Alias Analysis.
203
204``-lint``: Statically lint-checks LLVM IR
205-----------------------------------------
206
207This pass statically checks for common and easily-identified constructs which
208produce undefined or likely unintended behavior in LLVM IR.
209
210It is not a guarantee of correctness, in two ways.  First, it isn't
211comprehensive.  There are checks which could be done statically which are not
212yet implemented.  Some of these are indicated by TODO comments, but those
213aren't comprehensive either.  Second, many conditions cannot be checked
214statically.  This pass does no dynamic instrumentation, so it can't check for
215all possible problems.
216
217Another limitation is that it assumes all code will be executed.  A store
218through a null pointer in a basic block which is never reached is harmless, but
219this pass will warn about it anyway.
220
221Optimization passes may make conditions that this pass checks for more or less
222obvious.  If an optimization pass appears to be introducing a warning, it may
223be that the optimization pass is merely exposing an existing condition in the
224code.
225
226This code may be run before :ref:`instcombine <passes-instcombine>`.  In many
227cases, instcombine checks for the same kinds of things and turns instructions
228with undefined behavior into unreachable (or equivalent).  Because of this,
229this pass makes some effort to look through bitcasts and so on.
230
231``-loops``: Natural Loop Information
232------------------------------------
233
234This analysis is used to identify natural loops and determine the loop depth of
235various nodes of the CFG.  Note that the loops identified may actually be
236several natural loops that share the same header node... not just a single
237natural loop.
238
239``-memdep``: Memory Dependence Analysis
240---------------------------------------
241
242An analysis that determines, for a given memory operation, what preceding
243memory operations it depends on.  It builds on alias analysis information, and
244tries to provide a lazy, caching interface to a common kind of alias
245information query.
246
247``-module-debuginfo``: Decodes module-level debug info
248------------------------------------------------------
249
250This pass decodes the debug info metadata in a module and prints in a
251(sufficiently-prepared-) human-readable form.
252
253For example, run this pass from ``opt`` along with the ``-analyze`` option, and
254it'll print to standard output.
255
256``-postdomfrontier``: Post-Dominance Frontier Construction
257----------------------------------------------------------
258
259This pass is a simple post-dominator construction algorithm for finding
260post-dominator frontiers.
261
262``-postdomtree``: Post-Dominator Tree Construction
263--------------------------------------------------
264
265This pass is a simple post-dominator construction algorithm for finding
266post-dominators.
267
268``-print-alias-sets``: Alias Set Printer
269----------------------------------------
270
271Yet to be written.
272
273``-print-callgraph``: Print a call graph
274----------------------------------------
275
276This pass, only available in ``opt``, prints the call graph to standard error
277in a human-readable form.
278
279``-print-callgraph-sccs``: Print SCCs of the Call Graph
280-------------------------------------------------------
281
282This pass, only available in ``opt``, prints the SCCs of the call graph to
283standard error in a human-readable form.
284
285``-print-cfg-sccs``: Print SCCs of each function CFG
286----------------------------------------------------
287
288This pass, only available in ``opt``, printsthe SCCs of each function CFG to
289standard error in a human-readable fom.
290
291``-print-dom-info``: Dominator Info Printer
292-------------------------------------------
293
294Dominator Info Printer.
295
296``-print-externalfnconstants``: Print external fn callsites passed constants
297----------------------------------------------------------------------------
298
299This pass, only available in ``opt``, prints out call sites to external
300functions that are called with constant arguments.  This can be useful when
301looking for standard library functions we should constant fold or handle in
302alias analyses.
303
304``-print-function``: Print function to stderr
305---------------------------------------------
306
307The ``PrintFunctionPass`` class is designed to be pipelined with other
308``FunctionPasses``, and prints out the functions of the module as they are
309processed.
310
311``-print-module``: Print module to stderr
312-----------------------------------------
313
314This pass simply prints out the entire module when it is executed.
315
316.. _passes-print-used-types:
317
318``-print-used-types``: Find Used Types
319--------------------------------------
320
321This pass is used to seek out all of the types in use by the program.  Note
322that this analysis explicitly does not include types only used by the symbol
323table.
324
325``-regions``: Detect single entry single exit regions
326-----------------------------------------------------
327
328The ``RegionInfo`` pass detects single entry single exit regions in a function,
329where a region is defined as any subgraph that is connected to the remaining
330graph at only two spots.  Furthermore, an hierarchical region tree is built.
331
332``-scalar-evolution``: Scalar Evolution Analysis
333------------------------------------------------
334
335The ``ScalarEvolution`` analysis can be used to analyze and catagorize scalar
336expressions in loops.  It specializes in recognizing general induction
337variables, representing them with the abstract and opaque ``SCEV`` class.
338Given this analysis, trip counts of loops and other important properties can be
339obtained.
340
341This analysis is primarily useful for induction variable substitution and
342strength reduction.
343
344``-scev-aa``: ScalarEvolution-based Alias Analysis
345--------------------------------------------------
346
347Simple alias analysis implemented in terms of ``ScalarEvolution`` queries.
348
349This differs from traditional loop dependence analysis in that it tests for
350dependencies within a single iteration of a loop, rather than dependencies
351between different iterations.
352
353``ScalarEvolution`` has a more complete understanding of pointer arithmetic
354than ``BasicAliasAnalysis``' collection of ad-hoc analyses.
355
356``-targetdata``: Target Data Layout
357-----------------------------------
358
359Provides other passes access to information on how the size and alignment
360required by the target ABI for various data types.
361
362Transform Passes
363================
364
365This section describes the LLVM Transform Passes.
366
367``-adce``: Aggressive Dead Code Elimination
368-------------------------------------------
369
370ADCE aggressively tries to eliminate code.  This pass is similar to :ref:`DCE
371<passes-dce>` but it assumes that values are dead until proven otherwise.  This
372is similar to :ref:`SCCP <passes-sccp>`, except applied to the liveness of
373values.
374
375``-always-inline``: Inliner for ``always_inline`` functions
376-----------------------------------------------------------
377
378A custom inliner that handles only functions that are marked as "always
379inline".
380
381``-argpromotion``: Promote 'by reference' arguments to scalars
382--------------------------------------------------------------
383
384This pass promotes "by reference" arguments to be "by value" arguments.  In
385practice, this means looking for internal functions that have pointer
386arguments.  If it can prove, through the use of alias analysis, that an
387argument is *only* loaded, then it can pass the value into the function instead
388of the address of the value.  This can cause recursive simplification of code
389and lead to the elimination of allocas (especially in C++ template code like
390the STL).
391
392This pass also handles aggregate arguments that are passed into a function,
393scalarizing them if the elements of the aggregate are only loaded.  Note that
394it refuses to scalarize aggregates which would require passing in more than
395three operands to the function, because passing thousands of operands for a
396large array or structure is unprofitable!
397
398Note that this transformation could also be done for arguments that are only
399stored to (returning the value instead), but does not currently.  This case
400would be best handled when and if LLVM starts supporting multiple return values
401from functions.
402
403``-bb-vectorize``: Basic-Block Vectorization
404--------------------------------------------
405
406This pass combines instructions inside basic blocks to form vector
407instructions.  It iterates over each basic block, attempting to pair compatible
408instructions, repeating this process until no additional pairs are selected for
409vectorization.  When the outputs of some pair of compatible instructions are
410used as inputs by some other pair of compatible instructions, those pairs are
411part of a potential vectorization chain.  Instruction pairs are only fused into
412vector instructions when they are part of a chain longer than some threshold
413length.  Moreover, the pass attempts to find the best possible chain for each
414pair of compatible instructions.  These heuristics are intended to prevent
415vectorization in cases where it would not yield a performance increase of the
416resulting code.
417
418``-block-placement``: Profile Guided Basic Block Placement
419----------------------------------------------------------
420
421This pass is a very simple profile guided basic block placement algorithm.  The
422idea is to put frequently executed blocks together at the start of the function
423and hopefully increase the number of fall-through conditional branches.  If
424there is no profile information for a particular function, this pass basically
425orders blocks in depth-first order.
426
427``-break-crit-edges``: Break critical edges in CFG
428--------------------------------------------------
429
430Break all of the critical edges in the CFG by inserting a dummy basic block.
431It may be "required" by passes that cannot deal with critical edges.  This
432transformation obviously invalidates the CFG, but can update forward dominator
433(set, immediate dominators, tree, and frontier) information.
434
435``-codegenprepare``: Optimize for code generation
436-------------------------------------------------
437
438This pass munges the code in the input function to better prepare it for
439SelectionDAG-based code generation.  This works around limitations in its
440basic-block-at-a-time approach.  It should eventually be removed.
441
442``-constmerge``: Merge Duplicate Global Constants
443-------------------------------------------------
444
445Merges duplicate global constants together into a single constant that is
446shared.  This is useful because some passes (i.e., TraceValues) insert a lot of
447string constants into the program, regardless of whether or not an existing
448string is available.
449
450``-constprop``: Simple constant propagation
451-------------------------------------------
452
453This pass implements constant propagation and merging.  It looks for
454instructions involving only constant operands and replaces them with a constant
455value instead of an instruction.  For example:
456
457.. code-block:: llvm
458
459  add i32 1, 2
460
461becomes
462
463.. code-block:: llvm
464
465  i32 3
466
467NOTE: this pass has a habit of making definitions be dead.  It is a good idea
468to run a :ref:`Dead Instruction Elimination <passes-die>` pass sometime after
469running this pass.
470
471.. _passes-dce:
472
473``-dce``: Dead Code Elimination
474-------------------------------
475
476Dead code elimination is similar to :ref:`dead instruction elimination
477<passes-die>`, but it rechecks instructions that were used by removed
478instructions to see if they are newly dead.
479
480``-deadargelim``: Dead Argument Elimination
481-------------------------------------------
482
483This pass deletes dead arguments from internal functions.  Dead argument
484elimination removes arguments which are directly dead, as well as arguments
485only passed into function calls as dead arguments of other functions.  This
486pass also deletes dead arguments in a similar way.
487
488This pass is often useful as a cleanup pass to run after aggressive
489interprocedural passes, which add possibly-dead arguments.
490
491``-deadtypeelim``: Dead Type Elimination
492----------------------------------------
493
494This pass is used to cleanup the output of GCC.  It eliminate names for types
495that are unused in the entire translation unit, using the :ref:`find used types
496<passes-print-used-types>` pass.
497
498.. _passes-die:
499
500``-die``: Dead Instruction Elimination
501--------------------------------------
502
503Dead instruction elimination performs a single pass over the function, removing
504instructions that are obviously dead.
505
506``-dse``: Dead Store Elimination
507--------------------------------
508
509A trivial dead store elimination that only considers basic-block local
510redundant stores.
511
512.. _passes-functionattrs:
513
514``-functionattrs``: Deduce function attributes
515----------------------------------------------
516
517A simple interprocedural pass which walks the call-graph, looking for functions
518which do not access or only read non-local memory, and marking them
519``readnone``/``readonly``.  In addition, it marks function arguments (of
520pointer type) "``nocapture``" if a call to the function does not create any
521copies of the pointer value that outlive the call.  This more or less means
522that the pointer is only dereferenced, and not returned from the function or
523stored in a global.  This pass is implemented as a bottom-up traversal of the
524call-graph.
525
526``-globaldce``: Dead Global Elimination
527---------------------------------------
528
529This transform is designed to eliminate unreachable internal globals from the
530program.  It uses an aggressive algorithm, searching out globals that are known
531to be alive.  After it finds all of the globals which are needed, it deletes
532whatever is left over.  This allows it to delete recursive chunks of the
533program which are unreachable.
534
535``-globalopt``: Global Variable Optimizer
536-----------------------------------------
537
538This pass transforms simple global variables that never have their address
539taken.  If obviously true, it marks read/write globals as constant, deletes
540variables only stored to, etc.
541
542``-gvn``: Global Value Numbering
543--------------------------------
544
545This pass performs global value numbering to eliminate fully and partially
546redundant instructions.  It also performs redundant load elimination.
547
548.. _passes-indvars:
549
550``-indvars``: Canonicalize Induction Variables
551----------------------------------------------
552
553This transformation analyzes and transforms the induction variables (and
554computations derived from them) into simpler forms suitable for subsequent
555analysis and transformation.
556
557This transformation makes the following changes to each loop with an
558identifiable induction variable:
559
560* All loops are transformed to have a *single* canonical induction variable
561  which starts at zero and steps by one.
562* The canonical induction variable is guaranteed to be the first PHI node in
563  the loop header block.
564* Any pointer arithmetic recurrences are raised to use array subscripts.
565
566If the trip count of a loop is computable, this pass also makes the following
567changes:
568
569* The exit condition for the loop is canonicalized to compare the induction
570  value against the exit value.  This turns loops like:
571
572  .. code-block:: c++
573
574    for (i = 7; i*i < 1000; ++i)
575
576    into
577
578  .. code-block:: c++
579
580    for (i = 0; i != 25; ++i)
581
582* Any use outside of the loop of an expression derived from the indvar is
583  changed to compute the derived value outside of the loop, eliminating the
584  dependence on the exit value of the induction variable.  If the only purpose
585  of the loop is to compute the exit value of some derived expression, this
586  transformation will make the loop dead.
587
588This transformation should be followed by strength reduction after all of the
589desired loop transformations have been performed.  Additionally, on targets
590where it is profitable, the loop could be transformed to count down to zero
591(the "do loop" optimization).
592
593``-inline``: Function Integration/Inlining
594------------------------------------------
595
596Bottom-up inlining of functions into callees.
597
598.. _passes-instcombine:
599
600``-instcombine``: Combine redundant instructions
601------------------------------------------------
602
603Combine instructions to form fewer, simple instructions.  This pass does not
604modify the CFG. This pass is where algebraic simplification happens.
605
606This pass combines things like:
607
608.. code-block:: llvm
609
610  %Y = add i32 %X, 1
611  %Z = add i32 %Y, 1
612
613into:
614
615.. code-block:: llvm
616
617  %Z = add i32 %X, 2
618
619This is a simple worklist driven algorithm.
620
621This pass guarantees that the following canonicalizations are performed on the
622program:
623
624#. If a binary operator has a constant operand, it is moved to the right-hand
625   side.
626#. Bitwise operators with constant operands are always grouped so that shifts
627   are performed first, then ``or``\ s, then ``and``\ s, then ``xor``\ s.
628#. Compare instructions are converted from ``<``, ``>``, ``≤``, or ``≥`` to
629   ``=`` or ``≠`` if possible.
630#. All ``cmp`` instructions on boolean values are replaced with logical
631   operations.
632#. ``add X, X`` is represented as ``mul X, 2`` ⇒ ``shl X, 1``
633#. Multiplies with a constant power-of-two argument are transformed into
634   shifts.
635#. … etc.
636
637This pass can also simplify calls to specific well-known function calls (e.g.
638runtime library functions).  For example, a call ``exit(3)`` that occurs within
639the ``main()`` function can be transformed into simply ``return 3``. Whether or
640not library calls are simplified is controlled by the
641:ref:`-functionattrs <passes-functionattrs>` pass and LLVM's knowledge of
642library calls on different targets.
643
644``-internalize``: Internalize Global Symbols
645--------------------------------------------
646
647This pass loops over all of the functions in the input module, looking for a
648main function.  If a main function is found, all other functions and all global
649variables with initializers are marked as internal.
650
651``-ipconstprop``: Interprocedural constant propagation
652------------------------------------------------------
653
654This pass implements an *extremely* simple interprocedural constant propagation
655pass.  It could certainly be improved in many different ways, like using a
656worklist.  This pass makes arguments dead, but does not remove them.  The
657existing dead argument elimination pass should be run after this to clean up
658the mess.
659
660``-ipsccp``: Interprocedural Sparse Conditional Constant Propagation
661--------------------------------------------------------------------
662
663An interprocedural variant of :ref:`Sparse Conditional Constant Propagation
664<passes-sccp>`.
665
666``-jump-threading``: Jump Threading
667-----------------------------------
668
669Jump threading tries to find distinct threads of control flow running through a
670basic block.  This pass looks at blocks that have multiple predecessors and
671multiple successors.  If one or more of the predecessors of the block can be
672proven to always cause a jump to one of the successors, we forward the edge
673from the predecessor to the successor by duplicating the contents of this
674block.
675
676An example of when this can occur is code like this:
677
678.. code-block:: c++
679
680  if () { ...
681    X = 4;
682  }
683  if (X < 3) {
684
685In this case, the unconditional branch at the end of the first if can be
686revectored to the false side of the second if.
687
688``-lcssa``: Loop-Closed SSA Form Pass
689-------------------------------------
690
691This pass transforms loops by placing phi nodes at the end of the loops for all
692values that are live across the loop boundary.  For example, it turns the left
693into the right code:
694
695.. code-block:: c++
696
697  for (...)                for (...)
698      if (c)                   if (c)
699          X1 = ...                 X1 = ...
700      else                     else
701          X2 = ...                 X2 = ...
702      X3 = phi(X1, X2)         X3 = phi(X1, X2)
703  ... = X3 + 4              X4 = phi(X3)
704                              ... = X4 + 4
705
706This is still valid LLVM; the extra phi nodes are purely redundant, and will be
707trivially eliminated by ``InstCombine``.  The major benefit of this
708transformation is that it makes many other loop optimizations, such as
709``LoopUnswitch``\ ing, simpler.
710
711.. _passes-licm:
712
713``-licm``: Loop Invariant Code Motion
714-------------------------------------
715
716This pass performs loop invariant code motion, attempting to remove as much
717code from the body of a loop as possible.  It does this by either hoisting code
718into the preheader block, or by sinking code to the exit blocks if it is safe.
719This pass also promotes must-aliased memory locations in the loop to live in
720registers, thus hoisting and sinking "invariant" loads and stores.
721
722This pass uses alias analysis for two purposes:
723
724#. Moving loop invariant loads and calls out of loops.  If we can determine
725   that a load or call inside of a loop never aliases anything stored to, we
726   can hoist it or sink it like any other instruction.
727
728#. Scalar Promotion of Memory.  If there is a store instruction inside of the
729   loop, we try to move the store to happen AFTER the loop instead of inside of
730   the loop.  This can only happen if a few conditions are true:
731
732   #. The pointer stored through is loop invariant.
733   #. There are no stores or loads in the loop which *may* alias the pointer.
734      There are no calls in the loop which mod/ref the pointer.
735
736   If these conditions are true, we can promote the loads and stores in the
737   loop of the pointer to use a temporary alloca'd variable.  We then use the
738   :ref:`mem2reg <passes-mem2reg>` functionality to construct the appropriate
739   SSA form for the variable.
740
741``-loop-deletion``: Delete dead loops
742-------------------------------------
743
744This file implements the Dead Loop Deletion Pass.  This pass is responsible for
745eliminating loops with non-infinite computable trip counts that have no side
746effects or volatile instructions, and do not contribute to the computation of
747the function's return value.
748
749.. _passes-loop-extract:
750
751``-loop-extract``: Extract loops into new functions
752---------------------------------------------------
753
754A pass wrapper around the ``ExtractLoop()`` scalar transformation to extract
755each top-level loop into its own new function.  If the loop is the *only* loop
756in a given function, it is not touched.  This is a pass most useful for
757debugging via bugpoint.
758
759``-loop-extract-single``: Extract at most one loop into a new function
760----------------------------------------------------------------------
761
762Similar to :ref:`Extract loops into new functions <passes-loop-extract>`, this
763pass extracts one natural loop from the program into a function if it can.
764This is used by :program:`bugpoint`.
765
766``-loop-reduce``: Loop Strength Reduction
767-----------------------------------------
768
769This pass performs a strength reduction on array references inside loops that
770have as one or more of their components the loop induction variable.  This is
771accomplished by creating a new value to hold the initial value of the array
772access for the first iteration, and then creating a new GEP instruction in the
773loop to increment the value by the appropriate amount.
774
775``-loop-rotate``: Rotate Loops
776------------------------------
777
778A simple loop rotation transformation.
779
780``-loop-simplify``: Canonicalize natural loops
781----------------------------------------------
782
783This pass performs several transformations to transform natural loops into a
784simpler form, which makes subsequent analyses and transformations simpler and
785more effective.
786
787Loop pre-header insertion guarantees that there is a single, non-critical entry
788edge from outside of the loop to the loop header.  This simplifies a number of
789analyses and transformations, such as :ref:`LICM <passes-licm>`.
790
791Loop exit-block insertion guarantees that all exit blocks from the loop (blocks
792which are outside of the loop that have predecessors inside of the loop) only
793have predecessors from inside of the loop (and are thus dominated by the loop
794header).  This simplifies transformations such as store-sinking that are built
795into LICM.
796
797This pass also guarantees that loops will have exactly one backedge.
798
799Note that the :ref:`simplifycfg <passes-simplifycfg>` pass will clean up blocks
800which are split out but end up being unnecessary, so usage of this pass should
801not pessimize generated code.
802
803This pass obviously modifies the CFG, but updates loop information and
804dominator information.
805
806``-loop-unroll``: Unroll loops
807------------------------------
808
809This pass implements a simple loop unroller.  It works best when loops have
810been canonicalized by the :ref:`indvars <passes-indvars>` pass, allowing it to
811determine the trip counts of loops easily.
812
813``-loop-unswitch``: Unswitch loops
814----------------------------------
815
816This pass transforms loops that contain branches on loop-invariant conditions
817to have multiple loops.  For example, it turns the left into the right code:
818
819.. code-block:: c++
820
821  for (...)                  if (lic)
822      A                          for (...)
823      if (lic)                       A; B; C
824          B                  else
825      C                          for (...)
826                                     A; C
827
828This can increase the size of the code exponentially (doubling it every time a
829loop is unswitched) so we only unswitch if the resultant code will be smaller
830than a threshold.
831
832This pass expects :ref:`LICM <passes-licm>` to be run before it to hoist
833invariant conditions out of the loop, to make the unswitching opportunity
834obvious.
835
836``-loweratomic``: Lower atomic intrinsics to non-atomic form
837------------------------------------------------------------
838
839This pass lowers atomic intrinsics to non-atomic form for use in a known
840non-preemptible environment.
841
842The pass does not verify that the environment is non-preemptible (in general
843this would require knowledge of the entire call graph of the program including
844any libraries which may not be available in bitcode form); it simply lowers
845every atomic intrinsic.
846
847``-lowerinvoke``: Lower invokes to calls, for unwindless code generators
848------------------------------------------------------------------------
849
850This transformation is designed for use by code generators which do not yet
851support stack unwinding.  This pass converts ``invoke`` instructions to
852``call`` instructions, so that any exception-handling ``landingpad`` blocks
853become dead code (which can be removed by running the ``-simplifycfg`` pass
854afterwards).
855
856``-lowerswitch``: Lower ``SwitchInst``\ s to branches
857-----------------------------------------------------
858
859Rewrites switch instructions with a sequence of branches, which allows targets
860to get away with not implementing the switch instruction until it is
861convenient.
862
863.. _passes-mem2reg:
864
865``-mem2reg``: Promote Memory to Register
866----------------------------------------
867
868This file promotes memory references to be register references.  It promotes
869alloca instructions which only have loads and stores as uses.  An ``alloca`` is
870transformed by using dominator frontiers to place phi nodes, then traversing
871the function in depth-first order to rewrite loads and stores as appropriate.
872This is just the standard SSA construction algorithm to construct "pruned" SSA
873form.
874
875``-memcpyopt``: MemCpy Optimization
876-----------------------------------
877
878This pass performs various transformations related to eliminating ``memcpy``
879calls, or transforming sets of stores into ``memset``\ s.
880
881``-mergefunc``: Merge Functions
882-------------------------------
883
884This pass looks for equivalent functions that are mergable and folds them.
885
886Total-ordering is introduced among the functions set: we define comparison
887that answers for every two functions which of them is greater. It allows to
888arrange functions into the binary tree.
889
890For every new function we check for equivalent in tree.
891
892If equivalent exists we fold such functions. If both functions are overridable,
893we move the functionality into a new internal function and leave two
894overridable thunks to it.
895
896If there is no equivalent, then we add this function to tree.
897
898Lookup routine has O(log(n)) complexity, while whole merging process has
899complexity of O(n*log(n)).
900
901Read
902:doc:`this <MergeFunctions>`
903article for more details.
904
905``-mergereturn``: Unify function exit nodes
906-------------------------------------------
907
908Ensure that functions have at most one ``ret`` instruction in them.
909Additionally, it keeps track of which node is the new exit node of the CFG.
910
911``-partial-inliner``: Partial Inliner
912-------------------------------------
913
914This pass performs partial inlining, typically by inlining an ``if`` statement
915that surrounds the body of the function.
916
917``-prune-eh``: Remove unused exception handling info
918----------------------------------------------------
919
920This file implements a simple interprocedural pass which walks the call-graph,
921turning invoke instructions into call instructions if and only if the callee
922cannot throw an exception.  It implements this as a bottom-up traversal of the
923call-graph.
924
925``-reassociate``: Reassociate expressions
926-----------------------------------------
927
928This pass reassociates commutative expressions in an order that is designed to
929promote better constant propagation, GCSE, :ref:`LICM <passes-licm>`, PRE, etc.
930
931For example: 4 + (x + 5) ⇒ x + (4 + 5)
932
933In the implementation of this algorithm, constants are assigned rank = 0,
934function arguments are rank = 1, and other values are assigned ranks
935corresponding to the reverse post order traversal of current function (starting
936at 2), which effectively gives values in deep loops higher rank than values not
937in loops.
938
939``-reg2mem``: Demote all values to stack slots
940----------------------------------------------
941
942This file demotes all registers to memory references.  It is intended to be the
943inverse of :ref:`mem2reg <passes-mem2reg>`.  By converting to ``load``
944instructions, the only values live across basic blocks are ``alloca``
945instructions and ``load`` instructions before ``phi`` nodes.  It is intended
946that this should make CFG hacking much easier.  To make later hacking easier,
947the entry block is split into two, such that all introduced ``alloca``
948instructions (and nothing else) are in the entry block.
949
950``-sroa``: Scalar Replacement of Aggregates
951------------------------------------------------------
952
953The well-known scalar replacement of aggregates transformation.  This transform
954breaks up ``alloca`` instructions of aggregate type (structure or array) into
955individual ``alloca`` instructions for each member if possible.  Then, if
956possible, it transforms the individual ``alloca`` instructions into nice clean
957scalar SSA form.
958
959.. _passes-sccp:
960
961``-sccp``: Sparse Conditional Constant Propagation
962--------------------------------------------------
963
964Sparse conditional constant propagation and merging, which can be summarized
965as:
966
967* Assumes values are constant unless proven otherwise
968* Assumes BasicBlocks are dead unless proven otherwise
969* Proves values to be constant, and replaces them with constants
970* Proves conditional branches to be unconditional
971
972Note that this pass has a habit of making definitions be dead.  It is a good
973idea to run a :ref:`DCE <passes-dce>` pass sometime after running this pass.
974
975.. _passes-simplifycfg:
976
977``-simplifycfg``: Simplify the CFG
978----------------------------------
979
980Performs dead code elimination and basic block merging.  Specifically:
981
982* Removes basic blocks with no predecessors.
983* Merges a basic block into its predecessor if there is only one and the
984  predecessor only has one successor.
985* Eliminates PHI nodes for basic blocks with a single predecessor.
986* Eliminates a basic block that only contains an unconditional branch.
987
988``-sink``: Code sinking
989-----------------------
990
991This pass moves instructions into successor blocks, when possible, so that they
992aren't executed on paths where their results aren't needed.
993
994``-strip``: Strip all symbols from a module
995-------------------------------------------
996
997Performs code stripping.  This transformation can delete:
998
999* names for virtual registers
1000* symbols for internal globals and functions
1001* debug information
1002
1003Note that this transformation makes code much less readable, so it should only
1004be used in situations where the strip utility would be used, such as reducing
1005code size or making it harder to reverse engineer code.
1006
1007``-strip-dead-debug-info``: Strip debug info for unused symbols
1008---------------------------------------------------------------
1009
1010.. FIXME: this description is the same as for -strip
1011
1012performs code stripping. this transformation can delete:
1013
1014* names for virtual registers
1015* symbols for internal globals and functions
1016* debug information
1017
1018note that this transformation makes code much less readable, so it should only
1019be used in situations where the strip utility would be used, such as reducing
1020code size or making it harder to reverse engineer code.
1021
1022``-strip-dead-prototypes``: Strip Unused Function Prototypes
1023------------------------------------------------------------
1024
1025This pass loops over all of the functions in the input module, looking for dead
1026declarations and removes them.  Dead declarations are declarations of functions
1027for which no implementation is available (i.e., declarations for unused library
1028functions).
1029
1030``-strip-debug-declare``: Strip all ``llvm.dbg.declare`` intrinsics
1031-------------------------------------------------------------------
1032
1033.. FIXME: this description is the same as for -strip
1034
1035This pass implements code stripping.  Specifically, it can delete:
1036
1037#. names for virtual registers
1038#. symbols for internal globals and functions
1039#. debug information
1040
1041Note that this transformation makes code much less readable, so it should only
1042be used in situations where the 'strip' utility would be used, such as reducing
1043code size or making it harder to reverse engineer code.
1044
1045``-strip-nondebug``: Strip all symbols, except dbg symbols, from a module
1046-------------------------------------------------------------------------
1047
1048.. FIXME: this description is the same as for -strip
1049
1050This pass implements code stripping.  Specifically, it can delete:
1051
1052#. names for virtual registers
1053#. symbols for internal globals and functions
1054#. debug information
1055
1056Note that this transformation makes code much less readable, so it should only
1057be used in situations where the 'strip' utility would be used, such as reducing
1058code size or making it harder to reverse engineer code.
1059
1060``-tailcallelim``: Tail Call Elimination
1061----------------------------------------
1062
1063This file transforms calls of the current function (self recursion) followed by
1064a return instruction with a branch to the entry of the function, creating a
1065loop.  This pass also implements the following extensions to the basic
1066algorithm:
1067
1068#. Trivial instructions between the call and return do not prevent the
1069   transformation from taking place, though currently the analysis cannot
1070   support moving any really useful instructions (only dead ones).
1071#. This pass transforms functions that are prevented from being tail recursive
1072   by an associative expression to use an accumulator variable, thus compiling
1073   the typical naive factorial or fib implementation into efficient code.
1074#. TRE is performed if the function returns void, if the return returns the
1075   result returned by the call, or if the function returns a run-time constant
1076   on all exits from the function.  It is possible, though unlikely, that the
1077   return returns something else (like constant 0), and can still be TRE'd.  It
1078   can be TRE'd if *all other* return instructions in the function return the
1079   exact same value.
1080#. If it can prove that callees do not access theier caller stack frame, they
1081   are marked as eligible for tail call elimination (by the code generator).
1082
1083Utility Passes
1084==============
1085
1086This section describes the LLVM Utility Passes.
1087
1088``-deadarghaX0r``: Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE)
1089------------------------------------------------------------------------
1090
1091Same as dead argument elimination, but deletes arguments to functions which are
1092external.  This is only for use by :doc:`bugpoint <Bugpoint>`.
1093
1094``-extract-blocks``: Extract Basic Blocks From Module (for bugpoint use)
1095------------------------------------------------------------------------
1096
1097This pass is used by bugpoint to extract all blocks from the module into their
1098own functions.
1099
1100``-instnamer``: Assign names to anonymous instructions
1101------------------------------------------------------
1102
1103This is a little utility pass that gives instructions names, this is mostly
1104useful when diffing the effect of an optimization because deleting an unnamed
1105instruction can change all other instruction numbering, making the diff very
1106noisy.
1107
1108.. _passes-verify:
1109
1110``-verify``: Module Verifier
1111----------------------------
1112
1113Verifies an LLVM IR code.  This is useful to run after an optimization which is
1114undergoing testing.  Note that llvm-as verifies its input before emitting
1115bitcode, and also that malformed bitcode is likely to make LLVM crash.  All
1116language front-ends are therefore encouraged to verify their output before
1117performing optimizing transformations.
1118
1119#. Both of a binary operator's parameters are of the same type.
1120#. Verify that the indices of mem access instructions match other operands.
1121#. Verify that arithmetic and other things are only performed on first-class
1122   types.  Verify that shifts and logicals only happen on integrals f.e.
1123#. All of the constants in a switch statement are of the correct type.
1124#. The code is in valid SSA form.
1125#. It is illegal to put a label into any other type (like a structure) or to
1126   return one.
1127#. Only phi nodes can be self referential: ``%x = add i32 %x``, ``%x`` is
1128   invalid.
1129#. PHI nodes must have an entry for each predecessor, with no extras.
1130#. PHI nodes must be the first thing in a basic block, all grouped together.
1131#. PHI nodes must have at least one entry.
1132#. All basic blocks should only end with terminator insts, not contain them.
1133#. The entry node to a function must not have predecessors.
1134#. All Instructions must be embedded into a basic block.
1135#. Functions cannot take a void-typed parameter.
1136#. Verify that a function's argument list agrees with its declared type.
1137#. It is illegal to specify a name for a void value.
1138#. It is illegal to have an internal global value with no initializer.
1139#. It is illegal to have a ``ret`` instruction that returns a value that does
1140   not agree with the function return value type.
1141#. Function call argument types match the function prototype.
1142#. All other things that are tested by asserts spread about the code.
1143
1144Note that this does not provide full security verification (like Java), but
1145instead just tries to ensure that code is well-formed.
1146
1147``-view-cfg``: View CFG of function
1148-----------------------------------
1149
1150Displays the control flow graph using the GraphViz tool.
1151
1152``-view-cfg-only``: View CFG of function (with no function bodies)
1153------------------------------------------------------------------
1154
1155Displays the control flow graph using the GraphViz tool, but omitting function
1156bodies.
1157
1158``-view-dom``: View dominance tree of function
1159----------------------------------------------
1160
1161Displays the dominator tree using the GraphViz tool.
1162
1163``-view-dom-only``: View dominance tree of function (with no function bodies)
1164-----------------------------------------------------------------------------
1165
1166Displays the dominator tree using the GraphViz tool, but omitting function
1167bodies.
1168
1169``-view-postdom``: View postdominance tree of function
1170------------------------------------------------------
1171
1172Displays the post dominator tree using the GraphViz tool.
1173
1174``-view-postdom-only``: View postdominance tree of function (with no function bodies)
1175-------------------------------------------------------------------------------------
1176
1177Displays the post dominator tree using the GraphViz tool, but omitting function
1178bodies.
1179
1180