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.. _passes-da: 87 88``-da``: Dependence Analysis 89---------------------------- 90 91Dependence analysis framework, which is used to detect dependences in memory 92accesses. 93 94``-debug-aa``: AA use debugger 95------------------------------ 96 97This simple pass checks alias analysis users to ensure that if they create a 98new value, they do not query AA without informing it of the value. It acts as 99a shim over any other AA pass you want. 100 101Yes keeping track of every value in the program is expensive, but this is a 102debugging pass. 103 104``-domfrontier``: Dominance Frontier Construction 105------------------------------------------------- 106 107This pass is a simple dominator construction algorithm for finding forward 108dominator frontiers. 109 110``-domtree``: Dominator Tree Construction 111----------------------------------------- 112 113This pass is a simple dominator construction algorithm for finding forward 114dominators. 115 116 117``-dot-callgraph``: Print Call Graph to "dot" file 118-------------------------------------------------- 119 120This pass, only available in ``opt``, prints the call graph into a ``.dot`` 121graph. This graph can then be processed with the "dot" tool to convert it to 122postscript or some other suitable format. 123 124``-dot-cfg``: Print CFG of function to "dot" file 125------------------------------------------------- 126 127This pass, only available in ``opt``, prints the control flow graph into a 128``.dot`` graph. This graph can then be processed with the :program:`dot` tool 129to convert it to postscript or some other suitable format. 130 131``-dot-cfg-only``: Print CFG of function to "dot" file (with no function bodies) 132-------------------------------------------------------------------------------- 133 134This pass, only available in ``opt``, prints the control flow graph into a 135``.dot`` graph, omitting the function bodies. This graph can then be processed 136with the :program:`dot` tool to convert it to postscript or some other suitable 137format. 138 139``-dot-dom``: Print dominance tree of function to "dot" file 140------------------------------------------------------------ 141 142This pass, only available in ``opt``, prints the dominator tree into a ``.dot`` 143graph. This graph can then be processed with the :program:`dot` tool to 144convert it to postscript or some other suitable format. 145 146``-dot-dom-only``: Print dominance tree of function to "dot" file (with no function bodies) 147------------------------------------------------------------------------------------------- 148 149This pass, only available in ``opt``, prints the dominator tree into a ``.dot`` 150graph, omitting the function bodies. This graph can then be processed with the 151:program:`dot` tool to convert it to postscript or some other suitable format. 152 153``-dot-postdom``: Print postdominance tree of function to "dot" file 154-------------------------------------------------------------------- 155 156This pass, only available in ``opt``, prints the post dominator tree into a 157``.dot`` graph. This graph can then be processed with the :program:`dot` tool 158to convert it to postscript or some other suitable format. 159 160``-dot-postdom-only``: Print postdominance tree of function to "dot" file (with no function bodies) 161--------------------------------------------------------------------------------------------------- 162 163This pass, only available in ``opt``, prints the post dominator tree into a 164``.dot`` graph, omitting the function bodies. This graph can then be processed 165with the :program:`dot` tool to convert it to postscript or some other suitable 166format. 167 168``-globalsmodref-aa``: Simple mod/ref analysis for globals 169---------------------------------------------------------- 170 171This simple pass provides alias and mod/ref information for global values that 172do not have their address taken, and keeps track of whether functions read or 173write memory (are "pure"). For this simple (but very common) case, we can 174provide pretty accurate and useful information. 175 176``-instcount``: Counts the various types of ``Instruction``\ s 177-------------------------------------------------------------- 178 179This pass collects the count of all instructions and reports them. 180 181``-intervals``: Interval Partition Construction 182----------------------------------------------- 183 184This analysis calculates and represents the interval partition of a function, 185or a preexisting interval partition. 186 187In this way, the interval partition may be used to reduce a flow graph down to 188its degenerate single node interval partition (unless it is irreducible). 189 190``-iv-users``: Induction Variable Users 191--------------------------------------- 192 193Bookkeeping for "interesting" users of expressions computed from induction 194variables. 195 196``-lazy-value-info``: Lazy Value Information Analysis 197----------------------------------------------------- 198 199Interface for lazy computation of value constraint information. 200 201``-libcall-aa``: LibCall Alias Analysis 202--------------------------------------- 203 204LibCall Alias Analysis. 205 206``-lint``: Statically lint-checks LLVM IR 207----------------------------------------- 208 209This pass statically checks for common and easily-identified constructs which 210produce undefined or likely unintended behavior in LLVM IR. 211 212It is not a guarantee of correctness, in two ways. First, it isn't 213comprehensive. There are checks which could be done statically which are not 214yet implemented. Some of these are indicated by TODO comments, but those 215aren't comprehensive either. Second, many conditions cannot be checked 216statically. This pass does no dynamic instrumentation, so it can't check for 217all possible problems. 218 219Another limitation is that it assumes all code will be executed. A store 220through a null pointer in a basic block which is never reached is harmless, but 221this pass will warn about it anyway. 222 223Optimization passes may make conditions that this pass checks for more or less 224obvious. If an optimization pass appears to be introducing a warning, it may 225be that the optimization pass is merely exposing an existing condition in the 226code. 227 228This code may be run before :ref:`instcombine <passes-instcombine>`. In many 229cases, instcombine checks for the same kinds of things and turns instructions 230with undefined behavior into unreachable (or equivalent). Because of this, 231this pass makes some effort to look through bitcasts and so on. 232 233``-loops``: Natural Loop Information 234------------------------------------ 235 236This analysis is used to identify natural loops and determine the loop depth of 237various nodes of the CFG. Note that the loops identified may actually be 238several natural loops that share the same header node... not just a single 239natural loop. 240 241``-memdep``: Memory Dependence Analysis 242--------------------------------------- 243 244An analysis that determines, for a given memory operation, what preceding 245memory operations it depends on. It builds on alias analysis information, and 246tries to provide a lazy, caching interface to a common kind of alias 247information query. 248 249``-module-debuginfo``: Decodes module-level debug info 250------------------------------------------------------ 251 252This pass decodes the debug info metadata in a module and prints in a 253(sufficiently-prepared-) human-readable form. 254 255For example, run this pass from ``opt`` along with the ``-analyze`` option, and 256it'll print to standard output. 257 258``-postdomfrontier``: Post-Dominance Frontier Construction 259---------------------------------------------------------- 260 261This pass is a simple post-dominator construction algorithm for finding 262post-dominator frontiers. 263 264``-postdomtree``: Post-Dominator Tree Construction 265-------------------------------------------------- 266 267This pass is a simple post-dominator construction algorithm for finding 268post-dominators. 269 270``-print-alias-sets``: Alias Set Printer 271---------------------------------------- 272 273Yet to be written. 274 275``-print-callgraph``: Print a call graph 276---------------------------------------- 277 278This pass, only available in ``opt``, prints the call graph to standard error 279in a human-readable form. 280 281``-print-callgraph-sccs``: Print SCCs of the Call Graph 282------------------------------------------------------- 283 284This pass, only available in ``opt``, prints the SCCs of the call graph to 285standard error in a human-readable form. 286 287``-print-cfg-sccs``: Print SCCs of each function CFG 288---------------------------------------------------- 289 290This pass, only available in ``opt``, printsthe SCCs of each function CFG to 291standard error in a human-readable fom. 292 293``-print-dom-info``: Dominator Info Printer 294------------------------------------------- 295 296Dominator Info Printer. 297 298``-print-externalfnconstants``: Print external fn callsites passed constants 299---------------------------------------------------------------------------- 300 301This pass, only available in ``opt``, prints out call sites to external 302functions that are called with constant arguments. This can be useful when 303looking for standard library functions we should constant fold or handle in 304alias analyses. 305 306``-print-function``: Print function to stderr 307--------------------------------------------- 308 309The ``PrintFunctionPass`` class is designed to be pipelined with other 310``FunctionPasses``, and prints out the functions of the module as they are 311processed. 312 313``-print-module``: Print module to stderr 314----------------------------------------- 315 316This pass simply prints out the entire module when it is executed. 317 318.. _passes-print-used-types: 319 320``-print-used-types``: Find Used Types 321-------------------------------------- 322 323This pass is used to seek out all of the types in use by the program. Note 324that this analysis explicitly does not include types only used by the symbol 325table. 326 327``-regions``: Detect single entry single exit regions 328----------------------------------------------------- 329 330The ``RegionInfo`` pass detects single entry single exit regions in a function, 331where a region is defined as any subgraph that is connected to the remaining 332graph at only two spots. Furthermore, an hierarchical region tree is built. 333 334``-scalar-evolution``: Scalar Evolution Analysis 335------------------------------------------------ 336 337The ``ScalarEvolution`` analysis can be used to analyze and catagorize scalar 338expressions in loops. It specializes in recognizing general induction 339variables, representing them with the abstract and opaque ``SCEV`` class. 340Given this analysis, trip counts of loops and other important properties can be 341obtained. 342 343This analysis is primarily useful for induction variable substitution and 344strength reduction. 345 346``-scev-aa``: ScalarEvolution-based Alias Analysis 347-------------------------------------------------- 348 349Simple alias analysis implemented in terms of ``ScalarEvolution`` queries. 350 351This differs from traditional loop dependence analysis in that it tests for 352dependencies within a single iteration of a loop, rather than dependencies 353between different iterations. 354 355``ScalarEvolution`` has a more complete understanding of pointer arithmetic 356than ``BasicAliasAnalysis``' collection of ad-hoc analyses. 357 358``-targetdata``: Target Data Layout 359----------------------------------- 360 361Provides other passes access to information on how the size and alignment 362required by the target ABI for various data types. 363 364Transform Passes 365================ 366 367This section describes the LLVM Transform Passes. 368 369``-adce``: Aggressive Dead Code Elimination 370------------------------------------------- 371 372ADCE aggressively tries to eliminate code. This pass is similar to :ref:`DCE 373<passes-dce>` but it assumes that values are dead until proven otherwise. This 374is similar to :ref:`SCCP <passes-sccp>`, except applied to the liveness of 375values. 376 377``-always-inline``: Inliner for ``always_inline`` functions 378----------------------------------------------------------- 379 380A custom inliner that handles only functions that are marked as "always 381inline". 382 383``-argpromotion``: Promote 'by reference' arguments to scalars 384-------------------------------------------------------------- 385 386This pass promotes "by reference" arguments to be "by value" arguments. In 387practice, this means looking for internal functions that have pointer 388arguments. If it can prove, through the use of alias analysis, that an 389argument is *only* loaded, then it can pass the value into the function instead 390of the address of the value. This can cause recursive simplification of code 391and lead to the elimination of allocas (especially in C++ template code like 392the STL). 393 394This pass also handles aggregate arguments that are passed into a function, 395scalarizing them if the elements of the aggregate are only loaded. Note that 396it refuses to scalarize aggregates which would require passing in more than 397three operands to the function, because passing thousands of operands for a 398large array or structure is unprofitable! 399 400Note that this transformation could also be done for arguments that are only 401stored to (returning the value instead), but does not currently. This case 402would be best handled when and if LLVM starts supporting multiple return values 403from functions. 404 405``-bb-vectorize``: Basic-Block Vectorization 406-------------------------------------------- 407 408This pass combines instructions inside basic blocks to form vector 409instructions. It iterates over each basic block, attempting to pair compatible 410instructions, repeating this process until no additional pairs are selected for 411vectorization. When the outputs of some pair of compatible instructions are 412used as inputs by some other pair of compatible instructions, those pairs are 413part of a potential vectorization chain. Instruction pairs are only fused into 414vector instructions when they are part of a chain longer than some threshold 415length. Moreover, the pass attempts to find the best possible chain for each 416pair of compatible instructions. These heuristics are intended to prevent 417vectorization in cases where it would not yield a performance increase of the 418resulting code. 419 420``-block-placement``: Profile Guided Basic Block Placement 421---------------------------------------------------------- 422 423This pass is a very simple profile guided basic block placement algorithm. The 424idea is to put frequently executed blocks together at the start of the function 425and hopefully increase the number of fall-through conditional branches. If 426there is no profile information for a particular function, this pass basically 427orders blocks in depth-first order. 428 429``-break-crit-edges``: Break critical edges in CFG 430-------------------------------------------------- 431 432Break all of the critical edges in the CFG by inserting a dummy basic block. 433It may be "required" by passes that cannot deal with critical edges. This 434transformation obviously invalidates the CFG, but can update forward dominator 435(set, immediate dominators, tree, and frontier) information. 436 437``-codegenprepare``: Optimize for code generation 438------------------------------------------------- 439 440This pass munges the code in the input function to better prepare it for 441SelectionDAG-based code generation. This works around limitations in its 442basic-block-at-a-time approach. It should eventually be removed. 443 444``-constmerge``: Merge Duplicate Global Constants 445------------------------------------------------- 446 447Merges duplicate global constants together into a single constant that is 448shared. This is useful because some passes (i.e., TraceValues) insert a lot of 449string constants into the program, regardless of whether or not an existing 450string is available. 451 452``-constprop``: Simple constant propagation 453------------------------------------------- 454 455This pass implements constant propagation and merging. It looks for 456instructions involving only constant operands and replaces them with a constant 457value instead of an instruction. For example: 458 459.. code-block:: llvm 460 461 add i32 1, 2 462 463becomes 464 465.. code-block:: llvm 466 467 i32 3 468 469NOTE: this pass has a habit of making definitions be dead. It is a good idea 470to run a :ref:`Dead Instruction Elimination <passes-die>` pass sometime after 471running this pass. 472 473.. _passes-dce: 474 475``-dce``: Dead Code Elimination 476------------------------------- 477 478Dead code elimination is similar to :ref:`dead instruction elimination 479<passes-die>`, but it rechecks instructions that were used by removed 480instructions to see if they are newly dead. 481 482``-deadargelim``: Dead Argument Elimination 483------------------------------------------- 484 485This pass deletes dead arguments from internal functions. Dead argument 486elimination removes arguments which are directly dead, as well as arguments 487only passed into function calls as dead arguments of other functions. This 488pass also deletes dead arguments in a similar way. 489 490This pass is often useful as a cleanup pass to run after aggressive 491interprocedural passes, which add possibly-dead arguments. 492 493``-deadtypeelim``: Dead Type Elimination 494---------------------------------------- 495 496This pass is used to cleanup the output of GCC. It eliminate names for types 497that are unused in the entire translation unit, using the :ref:`find used types 498<passes-print-used-types>` pass. 499 500.. _passes-die: 501 502``-die``: Dead Instruction Elimination 503-------------------------------------- 504 505Dead instruction elimination performs a single pass over the function, removing 506instructions that are obviously dead. 507 508``-dse``: Dead Store Elimination 509-------------------------------- 510 511A trivial dead store elimination that only considers basic-block local 512redundant stores. 513 514.. _passes-functionattrs: 515 516``-functionattrs``: Deduce function attributes 517---------------------------------------------- 518 519A simple interprocedural pass which walks the call-graph, looking for functions 520which do not access or only read non-local memory, and marking them 521``readnone``/``readonly``. In addition, it marks function arguments (of 522pointer type) "``nocapture``" if a call to the function does not create any 523copies of the pointer value that outlive the call. This more or less means 524that the pointer is only dereferenced, and not returned from the function or 525stored in a global. This pass is implemented as a bottom-up traversal of the 526call-graph. 527 528``-globaldce``: Dead Global Elimination 529--------------------------------------- 530 531This transform is designed to eliminate unreachable internal globals from the 532program. It uses an aggressive algorithm, searching out globals that are known 533to be alive. After it finds all of the globals which are needed, it deletes 534whatever is left over. This allows it to delete recursive chunks of the 535program which are unreachable. 536 537``-globalopt``: Global Variable Optimizer 538----------------------------------------- 539 540This pass transforms simple global variables that never have their address 541taken. If obviously true, it marks read/write globals as constant, deletes 542variables only stored to, etc. 543 544``-gvn``: Global Value Numbering 545-------------------------------- 546 547This pass performs global value numbering to eliminate fully and partially 548redundant instructions. It also performs redundant load elimination. 549 550.. _passes-indvars: 551 552``-indvars``: Canonicalize Induction Variables 553---------------------------------------------- 554 555This transformation analyzes and transforms the induction variables (and 556computations derived from them) into simpler forms suitable for subsequent 557analysis and transformation. 558 559This transformation makes the following changes to each loop with an 560identifiable induction variable: 561 562* All loops are transformed to have a *single* canonical induction variable 563 which starts at zero and steps by one. 564* The canonical induction variable is guaranteed to be the first PHI node in 565 the loop header block. 566* Any pointer arithmetic recurrences are raised to use array subscripts. 567 568If the trip count of a loop is computable, this pass also makes the following 569changes: 570 571* The exit condition for the loop is canonicalized to compare the induction 572 value against the exit value. This turns loops like: 573 574 .. code-block:: c++ 575 576 for (i = 7; i*i < 1000; ++i) 577 578 into 579 580 .. code-block:: c++ 581 582 for (i = 0; i != 25; ++i) 583 584* Any use outside of the loop of an expression derived from the indvar is 585 changed to compute the derived value outside of the loop, eliminating the 586 dependence on the exit value of the induction variable. If the only purpose 587 of the loop is to compute the exit value of some derived expression, this 588 transformation will make the loop dead. 589 590This transformation should be followed by strength reduction after all of the 591desired loop transformations have been performed. Additionally, on targets 592where it is profitable, the loop could be transformed to count down to zero 593(the "do loop" optimization). 594 595``-inline``: Function Integration/Inlining 596------------------------------------------ 597 598Bottom-up inlining of functions into callees. 599 600.. _passes-instcombine: 601 602``-instcombine``: Combine redundant instructions 603------------------------------------------------ 604 605Combine instructions to form fewer, simple instructions. This pass does not 606modify the CFG. This pass is where algebraic simplification happens. 607 608This pass combines things like: 609 610.. code-block:: llvm 611 612 %Y = add i32 %X, 1 613 %Z = add i32 %Y, 1 614 615into: 616 617.. code-block:: llvm 618 619 %Z = add i32 %X, 2 620 621This is a simple worklist driven algorithm. 622 623This pass guarantees that the following canonicalizations are performed on the 624program: 625 626#. If a binary operator has a constant operand, it is moved to the right-hand 627 side. 628#. Bitwise operators with constant operands are always grouped so that shifts 629 are performed first, then ``or``\ s, then ``and``\ s, then ``xor``\ s. 630#. Compare instructions are converted from ``<``, ``>``, ``≤``, or ``≥`` to 631 ``=`` or ``≠`` if possible. 632#. All ``cmp`` instructions on boolean values are replaced with logical 633 operations. 634#. ``add X, X`` is represented as ``mul X, 2`` ⇒ ``shl X, 1`` 635#. Multiplies with a constant power-of-two argument are transformed into 636 shifts. 637#. … etc. 638 639This pass can also simplify calls to specific well-known function calls (e.g. 640runtime library functions). For example, a call ``exit(3)`` that occurs within 641the ``main()`` function can be transformed into simply ``return 3``. Whether or 642not library calls are simplified is controlled by the 643:ref:`-functionattrs <passes-functionattrs>` pass and LLVM's knowledge of 644library calls on different targets. 645 646.. _passes-aggressive-instcombine: 647 648``-aggressive-instcombine``: Combine expression patterns 649-------------------------------------------------------- 650 651Combine expression patterns to form expressions with fewer, simple instructions. 652This pass does not modify the CFG. 653 654For example, this pass reduce width of expressions post-dominated by TruncInst 655into smaller width when applicable. 656 657It differs from instcombine pass in that it contains pattern optimization that 658requires higher complexity than the O(1), thus, it should run fewer times than 659instcombine pass. 660 661``-internalize``: Internalize Global Symbols 662-------------------------------------------- 663 664This pass loops over all of the functions in the input module, looking for a 665main function. If a main function is found, all other functions and all global 666variables with initializers are marked as internal. 667 668``-ipconstprop``: Interprocedural constant propagation 669------------------------------------------------------ 670 671This pass implements an *extremely* simple interprocedural constant propagation 672pass. It could certainly be improved in many different ways, like using a 673worklist. This pass makes arguments dead, but does not remove them. The 674existing dead argument elimination pass should be run after this to clean up 675the mess. 676 677``-ipsccp``: Interprocedural Sparse Conditional Constant Propagation 678-------------------------------------------------------------------- 679 680An interprocedural variant of :ref:`Sparse Conditional Constant Propagation 681<passes-sccp>`. 682 683``-jump-threading``: Jump Threading 684----------------------------------- 685 686Jump threading tries to find distinct threads of control flow running through a 687basic block. This pass looks at blocks that have multiple predecessors and 688multiple successors. If one or more of the predecessors of the block can be 689proven to always cause a jump to one of the successors, we forward the edge 690from the predecessor to the successor by duplicating the contents of this 691block. 692 693An example of when this can occur is code like this: 694 695.. code-block:: c++ 696 697 if () { ... 698 X = 4; 699 } 700 if (X < 3) { 701 702In this case, the unconditional branch at the end of the first if can be 703revectored to the false side of the second if. 704 705``-lcssa``: Loop-Closed SSA Form Pass 706------------------------------------- 707 708This pass transforms loops by placing phi nodes at the end of the loops for all 709values that are live across the loop boundary. For example, it turns the left 710into the right code: 711 712.. code-block:: c++ 713 714 for (...) for (...) 715 if (c) if (c) 716 X1 = ... X1 = ... 717 else else 718 X2 = ... X2 = ... 719 X3 = phi(X1, X2) X3 = phi(X1, X2) 720 ... = X3 + 4 X4 = phi(X3) 721 ... = X4 + 4 722 723This is still valid LLVM; the extra phi nodes are purely redundant, and will be 724trivially eliminated by ``InstCombine``. The major benefit of this 725transformation is that it makes many other loop optimizations, such as 726``LoopUnswitch``\ ing, simpler. 727 728.. _passes-licm: 729 730``-licm``: Loop Invariant Code Motion 731------------------------------------- 732 733This pass performs loop invariant code motion, attempting to remove as much 734code from the body of a loop as possible. It does this by either hoisting code 735into the preheader block, or by sinking code to the exit blocks if it is safe. 736This pass also promotes must-aliased memory locations in the loop to live in 737registers, thus hoisting and sinking "invariant" loads and stores. 738 739This pass uses alias analysis for two purposes: 740 741#. Moving loop invariant loads and calls out of loops. If we can determine 742 that a load or call inside of a loop never aliases anything stored to, we 743 can hoist it or sink it like any other instruction. 744 745#. Scalar Promotion of Memory. If there is a store instruction inside of the 746 loop, we try to move the store to happen AFTER the loop instead of inside of 747 the loop. This can only happen if a few conditions are true: 748 749 #. The pointer stored through is loop invariant. 750 #. There are no stores or loads in the loop which *may* alias the pointer. 751 There are no calls in the loop which mod/ref the pointer. 752 753 If these conditions are true, we can promote the loads and stores in the 754 loop of the pointer to use a temporary alloca'd variable. We then use the 755 :ref:`mem2reg <passes-mem2reg>` functionality to construct the appropriate 756 SSA form for the variable. 757 758``-loop-deletion``: Delete dead loops 759------------------------------------- 760 761This file implements the Dead Loop Deletion Pass. This pass is responsible for 762eliminating loops with non-infinite computable trip counts that have no side 763effects or volatile instructions, and do not contribute to the computation of 764the function's return value. 765 766.. _passes-loop-extract: 767 768``-loop-extract``: Extract loops into new functions 769--------------------------------------------------- 770 771A pass wrapper around the ``ExtractLoop()`` scalar transformation to extract 772each top-level loop into its own new function. If the loop is the *only* loop 773in a given function, it is not touched. This is a pass most useful for 774debugging via bugpoint. 775 776``-loop-extract-single``: Extract at most one loop into a new function 777---------------------------------------------------------------------- 778 779Similar to :ref:`Extract loops into new functions <passes-loop-extract>`, this 780pass extracts one natural loop from the program into a function if it can. 781This is used by :program:`bugpoint`. 782 783``-loop-reduce``: Loop Strength Reduction 784----------------------------------------- 785 786This pass performs a strength reduction on array references inside loops that 787have as one or more of their components the loop induction variable. This is 788accomplished by creating a new value to hold the initial value of the array 789access for the first iteration, and then creating a new GEP instruction in the 790loop to increment the value by the appropriate amount. 791 792``-loop-rotate``: Rotate Loops 793------------------------------ 794 795A simple loop rotation transformation. 796 797``-loop-simplify``: Canonicalize natural loops 798---------------------------------------------- 799 800This pass performs several transformations to transform natural loops into a 801simpler form, which makes subsequent analyses and transformations simpler and 802more effective. 803 804Loop pre-header insertion guarantees that there is a single, non-critical entry 805edge from outside of the loop to the loop header. This simplifies a number of 806analyses and transformations, such as :ref:`LICM <passes-licm>`. 807 808Loop exit-block insertion guarantees that all exit blocks from the loop (blocks 809which are outside of the loop that have predecessors inside of the loop) only 810have predecessors from inside of the loop (and are thus dominated by the loop 811header). This simplifies transformations such as store-sinking that are built 812into LICM. 813 814This pass also guarantees that loops will have exactly one backedge. 815 816Note that the :ref:`simplifycfg <passes-simplifycfg>` pass will clean up blocks 817which are split out but end up being unnecessary, so usage of this pass should 818not pessimize generated code. 819 820This pass obviously modifies the CFG, but updates loop information and 821dominator information. 822 823``-loop-unroll``: Unroll loops 824------------------------------ 825 826This pass implements a simple loop unroller. It works best when loops have 827been canonicalized by the :ref:`indvars <passes-indvars>` pass, allowing it to 828determine the trip counts of loops easily. 829 830``-loop-unroll-and-jam``: Unroll and Jam loops 831---------------------------------------------- 832 833This pass implements a simple unroll and jam classical loop optimisation pass. 834It transforms loop from: 835 836.. code-block:: c++ 837 838 for i.. i+= 1 for i.. i+= 4 839 for j.. for j.. 840 code(i, j) code(i, j) 841 code(i+1, j) 842 code(i+2, j) 843 code(i+3, j) 844 remainder loop 845 846Which can be seen as unrolling the outer loop and "jamming" (fusing) the inner 847loops into one. When variables or loads can be shared in the new inner loop, this 848can lead to significant performance improvements. It uses 849:ref:`Dependence Analysis <passes-da>` for proving the transformations are safe. 850 851``-loop-unswitch``: Unswitch loops 852---------------------------------- 853 854This pass transforms loops that contain branches on loop-invariant conditions 855to have multiple loops. For example, it turns the left into the right code: 856 857.. code-block:: c++ 858 859 for (...) if (lic) 860 A for (...) 861 if (lic) A; B; C 862 B else 863 C for (...) 864 A; C 865 866This can increase the size of the code exponentially (doubling it every time a 867loop is unswitched) so we only unswitch if the resultant code will be smaller 868than a threshold. 869 870This pass expects :ref:`LICM <passes-licm>` to be run before it to hoist 871invariant conditions out of the loop, to make the unswitching opportunity 872obvious. 873 874``-loweratomic``: Lower atomic intrinsics to non-atomic form 875------------------------------------------------------------ 876 877This pass lowers atomic intrinsics to non-atomic form for use in a known 878non-preemptible environment. 879 880The pass does not verify that the environment is non-preemptible (in general 881this would require knowledge of the entire call graph of the program including 882any libraries which may not be available in bitcode form); it simply lowers 883every atomic intrinsic. 884 885``-lowerinvoke``: Lower invokes to calls, for unwindless code generators 886------------------------------------------------------------------------ 887 888This transformation is designed for use by code generators which do not yet 889support stack unwinding. This pass converts ``invoke`` instructions to 890``call`` instructions, so that any exception-handling ``landingpad`` blocks 891become dead code (which can be removed by running the ``-simplifycfg`` pass 892afterwards). 893 894``-lowerswitch``: Lower ``SwitchInst``\ s to branches 895----------------------------------------------------- 896 897Rewrites switch instructions with a sequence of branches, which allows targets 898to get away with not implementing the switch instruction until it is 899convenient. 900 901.. _passes-mem2reg: 902 903``-mem2reg``: Promote Memory to Register 904---------------------------------------- 905 906This file promotes memory references to be register references. It promotes 907alloca instructions which only have loads and stores as uses. An ``alloca`` is 908transformed by using dominator frontiers to place phi nodes, then traversing 909the function in depth-first order to rewrite loads and stores as appropriate. 910This is just the standard SSA construction algorithm to construct "pruned" SSA 911form. 912 913``-memcpyopt``: MemCpy Optimization 914----------------------------------- 915 916This pass performs various transformations related to eliminating ``memcpy`` 917calls, or transforming sets of stores into ``memset``\ s. 918 919``-mergefunc``: Merge Functions 920------------------------------- 921 922This pass looks for equivalent functions that are mergable and folds them. 923 924Total-ordering is introduced among the functions set: we define comparison 925that answers for every two functions which of them is greater. It allows to 926arrange functions into the binary tree. 927 928For every new function we check for equivalent in tree. 929 930If equivalent exists we fold such functions. If both functions are overridable, 931we move the functionality into a new internal function and leave two 932overridable thunks to it. 933 934If there is no equivalent, then we add this function to tree. 935 936Lookup routine has O(log(n)) complexity, while whole merging process has 937complexity of O(n*log(n)). 938 939Read 940:doc:`this <MergeFunctions>` 941article for more details. 942 943``-mergereturn``: Unify function exit nodes 944------------------------------------------- 945 946Ensure that functions have at most one ``ret`` instruction in them. 947Additionally, it keeps track of which node is the new exit node of the CFG. 948 949``-partial-inliner``: Partial Inliner 950------------------------------------- 951 952This pass performs partial inlining, typically by inlining an ``if`` statement 953that surrounds the body of the function. 954 955``-prune-eh``: Remove unused exception handling info 956---------------------------------------------------- 957 958This file implements a simple interprocedural pass which walks the call-graph, 959turning invoke instructions into call instructions if and only if the callee 960cannot throw an exception. It implements this as a bottom-up traversal of the 961call-graph. 962 963``-reassociate``: Reassociate expressions 964----------------------------------------- 965 966This pass reassociates commutative expressions in an order that is designed to 967promote better constant propagation, GCSE, :ref:`LICM <passes-licm>`, PRE, etc. 968 969For example: 4 + (x + 5) ⇒ x + (4 + 5) 970 971In the implementation of this algorithm, constants are assigned rank = 0, 972function arguments are rank = 1, and other values are assigned ranks 973corresponding to the reverse post order traversal of current function (starting 974at 2), which effectively gives values in deep loops higher rank than values not 975in loops. 976 977``-reg2mem``: Demote all values to stack slots 978---------------------------------------------- 979 980This file demotes all registers to memory references. It is intended to be the 981inverse of :ref:`mem2reg <passes-mem2reg>`. By converting to ``load`` 982instructions, the only values live across basic blocks are ``alloca`` 983instructions and ``load`` instructions before ``phi`` nodes. It is intended 984that this should make CFG hacking much easier. To make later hacking easier, 985the entry block is split into two, such that all introduced ``alloca`` 986instructions (and nothing else) are in the entry block. 987 988``-sroa``: Scalar Replacement of Aggregates 989------------------------------------------------------ 990 991The well-known scalar replacement of aggregates transformation. This transform 992breaks up ``alloca`` instructions of aggregate type (structure or array) into 993individual ``alloca`` instructions for each member if possible. Then, if 994possible, it transforms the individual ``alloca`` instructions into nice clean 995scalar SSA form. 996 997.. _passes-sccp: 998 999``-sccp``: Sparse Conditional Constant Propagation 1000-------------------------------------------------- 1001 1002Sparse conditional constant propagation and merging, which can be summarized 1003as: 1004 1005* Assumes values are constant unless proven otherwise 1006* Assumes BasicBlocks are dead unless proven otherwise 1007* Proves values to be constant, and replaces them with constants 1008* Proves conditional branches to be unconditional 1009 1010Note that this pass has a habit of making definitions be dead. It is a good 1011idea to run a :ref:`DCE <passes-dce>` pass sometime after running this pass. 1012 1013.. _passes-simplifycfg: 1014 1015``-simplifycfg``: Simplify the CFG 1016---------------------------------- 1017 1018Performs dead code elimination and basic block merging. Specifically: 1019 1020* Removes basic blocks with no predecessors. 1021* Merges a basic block into its predecessor if there is only one and the 1022 predecessor only has one successor. 1023* Eliminates PHI nodes for basic blocks with a single predecessor. 1024* Eliminates a basic block that only contains an unconditional branch. 1025 1026``-sink``: Code sinking 1027----------------------- 1028 1029This pass moves instructions into successor blocks, when possible, so that they 1030aren't executed on paths where their results aren't needed. 1031 1032``-strip``: Strip all symbols from a module 1033------------------------------------------- 1034 1035Performs code stripping. This transformation 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-dead-debug-info``: Strip debug info for unused symbols 1046--------------------------------------------------------------- 1047 1048.. FIXME: this description is the same as for -strip 1049 1050performs code stripping. this transformation 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``-strip-dead-prototypes``: Strip Unused Function Prototypes 1061------------------------------------------------------------ 1062 1063This pass loops over all of the functions in the input module, looking for dead 1064declarations and removes them. Dead declarations are declarations of functions 1065for which no implementation is available (i.e., declarations for unused library 1066functions). 1067 1068``-strip-debug-declare``: Strip all ``llvm.dbg.declare`` intrinsics 1069------------------------------------------------------------------- 1070 1071.. FIXME: this description is the same as for -strip 1072 1073This pass implements code stripping. Specifically, it can delete: 1074 1075#. names for virtual registers 1076#. symbols for internal globals and functions 1077#. debug information 1078 1079Note that this transformation makes code much less readable, so it should only 1080be used in situations where the 'strip' utility would be used, such as reducing 1081code size or making it harder to reverse engineer code. 1082 1083``-strip-nondebug``: Strip all symbols, except dbg symbols, from a module 1084------------------------------------------------------------------------- 1085 1086.. FIXME: this description is the same as for -strip 1087 1088This pass implements code stripping. Specifically, it can delete: 1089 1090#. names for virtual registers 1091#. symbols for internal globals and functions 1092#. debug information 1093 1094Note that this transformation makes code much less readable, so it should only 1095be used in situations where the 'strip' utility would be used, such as reducing 1096code size or making it harder to reverse engineer code. 1097 1098``-tailcallelim``: Tail Call Elimination 1099---------------------------------------- 1100 1101This file transforms calls of the current function (self recursion) followed by 1102a return instruction with a branch to the entry of the function, creating a 1103loop. This pass also implements the following extensions to the basic 1104algorithm: 1105 1106#. Trivial instructions between the call and return do not prevent the 1107 transformation from taking place, though currently the analysis cannot 1108 support moving any really useful instructions (only dead ones). 1109#. This pass transforms functions that are prevented from being tail recursive 1110 by an associative expression to use an accumulator variable, thus compiling 1111 the typical naive factorial or fib implementation into efficient code. 1112#. TRE is performed if the function returns void, if the return returns the 1113 result returned by the call, or if the function returns a run-time constant 1114 on all exits from the function. It is possible, though unlikely, that the 1115 return returns something else (like constant 0), and can still be TRE'd. It 1116 can be TRE'd if *all other* return instructions in the function return the 1117 exact same value. 1118#. If it can prove that callees do not access theier caller stack frame, they 1119 are marked as eligible for tail call elimination (by the code generator). 1120 1121Utility Passes 1122============== 1123 1124This section describes the LLVM Utility Passes. 1125 1126``-deadarghaX0r``: Dead Argument Hacking (BUGPOINT USE ONLY; DO NOT USE) 1127------------------------------------------------------------------------ 1128 1129Same as dead argument elimination, but deletes arguments to functions which are 1130external. This is only for use by :doc:`bugpoint <Bugpoint>`. 1131 1132``-extract-blocks``: Extract Basic Blocks From Module (for bugpoint use) 1133------------------------------------------------------------------------ 1134 1135This pass is used by bugpoint to extract all blocks from the module into their 1136own functions. 1137 1138``-instnamer``: Assign names to anonymous instructions 1139------------------------------------------------------ 1140 1141This is a little utility pass that gives instructions names, this is mostly 1142useful when diffing the effect of an optimization because deleting an unnamed 1143instruction can change all other instruction numbering, making the diff very 1144noisy. 1145 1146.. _passes-verify: 1147 1148``-verify``: Module Verifier 1149---------------------------- 1150 1151Verifies an LLVM IR code. This is useful to run after an optimization which is 1152undergoing testing. Note that llvm-as verifies its input before emitting 1153bitcode, and also that malformed bitcode is likely to make LLVM crash. All 1154language front-ends are therefore encouraged to verify their output before 1155performing optimizing transformations. 1156 1157#. Both of a binary operator's parameters are of the same type. 1158#. Verify that the indices of mem access instructions match other operands. 1159#. Verify that arithmetic and other things are only performed on first-class 1160 types. Verify that shifts and logicals only happen on integrals f.e. 1161#. All of the constants in a switch statement are of the correct type. 1162#. The code is in valid SSA form. 1163#. It is illegal to put a label into any other type (like a structure) or to 1164 return one. 1165#. Only phi nodes can be self referential: ``%x = add i32 %x``, ``%x`` is 1166 invalid. 1167#. PHI nodes must have an entry for each predecessor, with no extras. 1168#. PHI nodes must be the first thing in a basic block, all grouped together. 1169#. PHI nodes must have at least one entry. 1170#. All basic blocks should only end with terminator insts, not contain them. 1171#. The entry node to a function must not have predecessors. 1172#. All Instructions must be embedded into a basic block. 1173#. Functions cannot take a void-typed parameter. 1174#. Verify that a function's argument list agrees with its declared type. 1175#. It is illegal to specify a name for a void value. 1176#. It is illegal to have an internal global value with no initializer. 1177#. It is illegal to have a ``ret`` instruction that returns a value that does 1178 not agree with the function return value type. 1179#. Function call argument types match the function prototype. 1180#. All other things that are tested by asserts spread about the code. 1181 1182Note that this does not provide full security verification (like Java), but 1183instead just tries to ensure that code is well-formed. 1184 1185``-view-cfg``: View CFG of function 1186----------------------------------- 1187 1188Displays the control flow graph using the GraphViz tool. 1189 1190``-view-cfg-only``: View CFG of function (with no function bodies) 1191------------------------------------------------------------------ 1192 1193Displays the control flow graph using the GraphViz tool, but omitting function 1194bodies. 1195 1196``-view-dom``: View dominance tree of function 1197---------------------------------------------- 1198 1199Displays the dominator tree using the GraphViz tool. 1200 1201``-view-dom-only``: View dominance tree of function (with no function bodies) 1202----------------------------------------------------------------------------- 1203 1204Displays the dominator tree using the GraphViz tool, but omitting function 1205bodies. 1206 1207``-view-postdom``: View postdominance tree of function 1208------------------------------------------------------ 1209 1210Displays the post dominator tree using the GraphViz tool. 1211 1212``-view-postdom-only``: View postdominance tree of function (with no function bodies) 1213------------------------------------------------------------------------------------- 1214 1215Displays the post dominator tree using the GraphViz tool, but omitting function 1216bodies. 1217 1218