/usr/share/doc/llvm-3.6-doc/html/MergeFunctions.html is in llvm-3.6-doc 1:3.6-2ubuntu1~trusty2.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 | <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<title>MergeFunctions pass, how it works — LLVM 3.6 documentation</title>
<link rel="stylesheet" href="_static/llvm-theme.css" type="text/css" />
<link rel="stylesheet" href="_static/pygments.css" type="text/css" />
<script type="text/javascript">
var DOCUMENTATION_OPTIONS = {
URL_ROOT: './',
VERSION: '3.6',
COLLAPSE_INDEX: false,
FILE_SUFFIX: '.html',
HAS_SOURCE: true
};
</script>
<script type="text/javascript" src="_static/jquery.js"></script>
<script type="text/javascript" src="_static/underscore.js"></script>
<script type="text/javascript" src="_static/doctools.js"></script>
<link rel="top" title="LLVM 3.6 documentation" href="index.html" />
<link rel="next" title="LLVM Developer Policy" href="DeveloperPolicy.html" />
<link rel="prev" title="Garbage Collection Safepoints in LLVM" href="Statepoints.html" />
<style type="text/css">
table.right { float: right; margin-left: 20px; }
table.right td { border: 1px solid #ccc; }
</style>
</head>
<body>
<div class="logo">
<a href="index.html">
<img src="_static/logo.png"
alt="LLVM Logo" width="250" height="88"/></a>
</div>
<div class="related">
<h3>Navigation</h3>
<ul>
<li class="right" style="margin-right: 10px">
<a href="genindex.html" title="General Index"
accesskey="I">index</a></li>
<li class="right" >
<a href="DeveloperPolicy.html" title="LLVM Developer Policy"
accesskey="N">next</a> |</li>
<li class="right" >
<a href="Statepoints.html" title="Garbage Collection Safepoints in LLVM"
accesskey="P">previous</a> |</li>
<li><a href="http://llvm.org/">LLVM Home</a> | </li>
<li><a href="index.html">Documentation</a>»</li>
</ul>
</div>
<div class="document">
<div class="documentwrapper">
<div class="body">
<div class="section" id="mergefunctions-pass-how-it-works">
<h1>MergeFunctions pass, how it works<a class="headerlink" href="#mergefunctions-pass-how-it-works" title="Permalink to this headline">¶</a></h1>
<div class="contents local topic" id="contents">
<ul class="simple">
<li><a class="reference internal" href="#introduction" id="id1">Introduction</a><ul>
<li><a class="reference internal" href="#why-would-i-want-to-read-this-document" id="id2">Why would I want to read this document?</a></li>
<li><a class="reference internal" href="#what-should-i-know-to-be-able-to-follow-along-with-this-document" id="id3">What should I know to be able to follow along with this document?</a></li>
<li><a class="reference internal" href="#what-i-gain-by-reading-this-document" id="id4">What I gain by reading this document?</a></li>
<li><a class="reference internal" href="#narrative-structure" id="id5">Narrative structure</a></li>
</ul>
</li>
<li><a class="reference internal" href="#basics" id="id6">Basics</a><ul>
<li><a class="reference internal" href="#how-to-do-it" id="id7">How to do it?</a><ul>
<li><a class="reference internal" href="#possible-solutions" id="id8">Possible solutions</a><ul>
<li><a class="reference internal" href="#random-access" id="id9">Random-access</a></li>
<li><a class="reference internal" href="#logarithmical-search" id="id10">Logarithmical search</a></li>
<li><a class="reference internal" href="#present-state" id="id11">Present state</a></li>
</ul>
</li>
<li><a class="reference internal" href="#mergefunctions-main-fields-and-runonmodule" id="id12">MergeFunctions, main fields and runOnModule</a><ul>
<li><a class="reference internal" href="#runonmodule" id="id13">runOnModule</a></li>
<li><a class="reference internal" href="#comparison-and-logarithmical-search" id="id14">Comparison and logarithmical search</a></li>
</ul>
</li>
</ul>
</li>
</ul>
</li>
<li><a class="reference internal" href="#functions-comparison" id="id15">Functions comparison</a><ul>
<li><a class="reference internal" href="#functioncomparator-compare-void" id="id16">FunctionComparator::compare(void)</a></li>
<li><a class="reference internal" href="#functioncomparator-cmptype" id="id17">FunctionComparator::cmpType</a></li>
<li><a class="reference internal" href="#cmpvalues-const-value-const-value" id="id18">cmpValues(const Value*, const Value*)</a><ul>
<li><a class="reference internal" href="#what-we-assiciate-in-cmpvalues" id="id19">What we assiciate in cmpValues?</a></li>
<li><a class="reference internal" href="#how-to-implement-cmpvalues" id="id20">How to implement cmpValues?</a></li>
</ul>
</li>
<li><a class="reference internal" href="#cmpconstants" id="id21">cmpConstants</a></li>
<li><a class="reference internal" href="#compare-const-basicblock-const-basicblock" id="id22">compare(const BasicBlock*, const BasicBlock*)</a></li>
<li><a class="reference internal" href="#cmpgep" id="id23">cmpGEP</a></li>
<li><a class="reference internal" href="#cmpoperation" id="id24">cmpOperation</a></li>
<li><a class="reference internal" href="#o-log-n" id="id25">O(log(N))</a></li>
</ul>
</li>
<li><a class="reference internal" href="#merging-process-mergetwofunctions" id="id26">Merging process, mergeTwoFunctions</a><ul>
<li><a class="reference internal" href="#if-f-may-be-overridden" id="id27">If “F” may be overridden</a><ul>
<li><a class="reference internal" href="#hasglobalaliases-removeusers" id="id28">HasGlobalAliases, removeUsers</a></li>
<li><a class="reference internal" href="#no-global-aliases-replacedirectcallers" id="id29">No global aliases, replaceDirectCallers</a><ul>
<li><a class="reference internal" href="#if-f-could-not-be-overridden-fix-it" id="id30">If “F” could not be overridden, fix it!</a></li>
</ul>
</li>
</ul>
</li>
</ul>
</li>
<li><a class="reference internal" href="#that-s-it" id="id31">That’s it.</a></li>
</ul>
</div>
<div class="section" id="introduction">
<h2><a class="toc-backref" href="#id1">Introduction</a><a class="headerlink" href="#introduction" title="Permalink to this headline">¶</a></h2>
<p>Sometimes code contains equal functions, or functions that does exactly the same
thing even though they are non-equal on the IR level (e.g.: multiplication on 2
and ‘shl 1’). It could happen due to several reasons: mainly, the usage of
templates and automatic code generators. Though, sometimes user itself could
write the same thing twice :-)</p>
<p>The main purpose of this pass is to recognize such functions and merge them.</p>
<div class="section" id="why-would-i-want-to-read-this-document">
<h3><a class="toc-backref" href="#id2">Why would I want to read this document?</a><a class="headerlink" href="#why-would-i-want-to-read-this-document" title="Permalink to this headline">¶</a></h3>
<p>Document is the extension to pass comments and describes the pass logic. It
describes algorithm that is used in order to compare functions, it also
explains how we could combine equal functions correctly, keeping module valid.</p>
<p>Material is brought in top-down form, so reader could start learn pass from
ideas and end up with low-level algorithm details, thus preparing him for
reading the sources.</p>
<p>So main goal is do describe algorithm and logic here; the concept. This document
is good for you, if you <em>don’t want</em> to read the source code, but want to
understand pass algorithms. Author tried not to repeat the source-code and
cover only common cases, and thus avoid cases when after minor code changes we
need to update this document.</p>
</div>
<div class="section" id="what-should-i-know-to-be-able-to-follow-along-with-this-document">
<h3><a class="toc-backref" href="#id3">What should I know to be able to follow along with this document?</a><a class="headerlink" href="#what-should-i-know-to-be-able-to-follow-along-with-this-document" title="Permalink to this headline">¶</a></h3>
<p>Reader should be familiar with common compile-engineering principles and LLVM
code fundamentals. In this article we suppose reader is familiar with
<a class="reference external" href="http://en.wikipedia.org/wiki/Static_single_assignment_form">Single Static Assingment</a>
concepts. Understanding of
<a class="reference external" href="http://llvm.org/docs/LangRef.html#high-level-structure">IR structure</a> is
also important.</p>
<p>We will use such terms as
“<a class="reference external" href="http://llvm.org/docs/LangRef.html#high-level-structure">module</a>”,
“<a class="reference external" href="http://llvm.org/docs/ProgrammersManual.html#the-function-class">function</a>”,
“<a class="reference external" href="http://en.wikipedia.org/wiki/Basic_block">basic block</a>”,
“<a class="reference external" href="http://llvm.org/docs/ProgrammersManual.html#the-user-class">user</a>”,
“<a class="reference external" href="http://llvm.org/docs/ProgrammersManual.html#the-value-class">value</a>”,
“<a class="reference external" href="http://llvm.org/docs/ProgrammersManual.html#the-instruction-class">instruction</a>”.</p>
<p>As a good start point, Kaleidoscope tutorial could be used:</p>
<p><a class="reference internal" href="tutorial/index.html"><em>LLVM Tutorial: Table of Contents</em></a></p>
<p>Especially it’s important to understand chapter 3 of tutorial:</p>
<p><a class="reference internal" href="tutorial/LangImpl3.html"><em>Kaleidoscope: Code generation to LLVM IR</em></a></p>
<p>Reader also should know how passes work in LLVM, he could use next article as a
reference and start point here:</p>
<p><a class="reference internal" href="WritingAnLLVMPass.html"><em>Writing an LLVM Pass</em></a></p>
<p>What else? Well perhaps reader also should have some experience in LLVM pass
debugging and bug-fixing.</p>
</div>
<div class="section" id="what-i-gain-by-reading-this-document">
<h3><a class="toc-backref" href="#id4">What I gain by reading this document?</a><a class="headerlink" href="#what-i-gain-by-reading-this-document" title="Permalink to this headline">¶</a></h3>
<p>Main purpose is to provide reader with comfortable form of algorithms
description, namely the human reading text. Since it could be hard to
understand algorithm straight from the source code: pass uses some principles
that have to be explained first.</p>
<p>Author wishes to everybody to avoid case, when you read code from top to bottom
again and again, and yet you don’t understand why we implemented it that way.</p>
<p>We hope that after this article reader could easily debug and improve
MergeFunctions pass and thus help LLVM project.</p>
</div>
<div class="section" id="narrative-structure">
<h3><a class="toc-backref" href="#id5">Narrative structure</a><a class="headerlink" href="#narrative-structure" title="Permalink to this headline">¶</a></h3>
<p>Article consists of three parts. First part explains pass functionality on the
top-level. Second part describes the comparison procedure itself. The third
part describes the merging process.</p>
<p>In every part author also tried to put the contents into the top-down form.
First, the top-level methods will be described, while the terminal ones will be
at the end, in the tail of each part. If reader will see the reference to the
method that wasn’t described yet, he will find its description a bit below.</p>
</div>
</div>
<div class="section" id="basics">
<h2><a class="toc-backref" href="#id6">Basics</a><a class="headerlink" href="#basics" title="Permalink to this headline">¶</a></h2>
<div class="section" id="how-to-do-it">
<h3><a class="toc-backref" href="#id7">How to do it?</a><a class="headerlink" href="#how-to-do-it" title="Permalink to this headline">¶</a></h3>
<p>Do we need to merge functions? Obvious thing is: yes that’s a quite possible
case, since usually we <em>do</em> have duplicates. And it would be good to get rid of
them. But how to detect such a duplicates? The idea is next: we split functions
onto small bricks (parts), then we compare “bricks” amount, and if it equal,
compare “bricks” themselves, and then do our conclusions about functions
themselves.</p>
<p>What the difference it could be? For example, on machine with 64-bit pointers
(let’s assume we have only one address space), one function stores 64-bit
integer, while another one stores a pointer. So if the target is a machine
mentioned above, and if functions are identical, except the parameter type (we
could consider it as a part of function type), then we can treat <tt class="docutils literal"><span class="pre">uint64_t</span></tt>
and``void*`` as equal.</p>
<p>It was just an example; possible details are described a bit below.</p>
<p>As another example reader may imagine two more functions. First function
performs multiplication on 2, while the second one performs arithmetic right
shift on 1.</p>
<div class="section" id="possible-solutions">
<h4><a class="toc-backref" href="#id8">Possible solutions</a><a class="headerlink" href="#possible-solutions" title="Permalink to this headline">¶</a></h4>
<p>Let’s briefly consider possible options about how and what we have to implement
in order to create full-featured functions merging, and also what it would
meant for us.</p>
<p>Equal functions detection, obviously supposes “detector” method to be
implemented, latter should answer the question “whether functions are equal”.
This “detector” method consists of tiny “sub-detectors”, each of them answers
exactly the same question, but for function parts.</p>
<p>As the second step, we should merge equal functions. So it should be a “merger”
method. “Merger” accepts two functions <em>F1</em> and <em>F2</em>, and produces <em>F1F2</em>
function, the result of merging.</p>
<p>Having such a routines in our hands, we can process whole module, and merge all
equal functions.</p>
<p>In this case, we have to compare every function with every another function. As
reader could notice, this way seems to be quite expensive. Of course we could
introduce hashing and other helpers, but it is still just an optimization, and
thus the level of O(N*N) complexity.</p>
<p>Can we reach another level? Could we introduce logarithmical search, or random
access lookup? The answer is: “yes”.</p>
<div class="section" id="random-access">
<h5><a class="toc-backref" href="#id9">Random-access</a><a class="headerlink" href="#random-access" title="Permalink to this headline">¶</a></h5>
<p>How it could be done? Just convert each function to number, and gather all of
them in special hash-table. Functions with equal hash are equal. Good hashing
means, that every function part must be taken into account. That means we have
to convert every function part into some number, and then add it into hash.
Lookup-up time would be small, but such approach adds some delay due to hashing
routine.</p>
</div>
<div class="section" id="logarithmical-search">
<h5><a class="toc-backref" href="#id10">Logarithmical search</a><a class="headerlink" href="#logarithmical-search" title="Permalink to this headline">¶</a></h5>
<p>We could introduce total ordering among the functions set, once we had it we
could then implement a logarithmical search. Lookup time still depends on N,
but adds a little of delay (<em>log(N)</em>).</p>
</div>
<div class="section" id="present-state">
<h5><a class="toc-backref" href="#id11">Present state</a><a class="headerlink" href="#present-state" title="Permalink to this headline">¶</a></h5>
<p>Both of approaches (random-access and logarithmical) has been implemented and
tested. And both of them gave a very good improvement. And what was most
surprising, logarithmical search was faster; sometimes up to 15%. Hashing needs
some extra CPU time, and it is the main reason why it works slower; in most of
cases total “hashing” time was greater than total “logarithmical-search” time.</p>
<p>So, preference has been granted to the “logarithmical search”.</p>
<p>Though in the case of need, <em>logarithmical-search</em> (read “total-ordering”) could
be used as a milestone on our way to the <em>random-access</em> implementation.</p>
<p>Every comparison is based either on the numbers or on flags comparison. In
<em>random-access</em> approach we could use the same comparison algorithm. During
comparison we exit once we find the difference, but here we might have to scan
whole function body every time (note, it could be slower). Like in
“total-ordering”, we will track every numbers and flags, but instead of
comparison, we should get numbers sequence and then create the hash number. So,
once again, <em>total-ordering</em> could be considered as a milestone for even faster
(in theory) random-access approach.</p>
</div>
</div>
<div class="section" id="mergefunctions-main-fields-and-runonmodule">
<h4><a class="toc-backref" href="#id12">MergeFunctions, main fields and runOnModule</a><a class="headerlink" href="#mergefunctions-main-fields-and-runonmodule" title="Permalink to this headline">¶</a></h4>
<p>There are two most important fields in class:</p>
<p><tt class="docutils literal"><span class="pre">FnTree</span></tt> – the set of all unique functions. It keeps items that couldn’t be
merged with each other. It is defined as:</p>
<p><tt class="docutils literal"><span class="pre">std::set<FunctionNode></span> <span class="pre">FnTree;</span></tt></p>
<p>Here <tt class="docutils literal"><span class="pre">FunctionNode</span></tt> is a wrapper for <tt class="docutils literal"><span class="pre">llvm::Function</span></tt> class, with
implemented “<” operator among the functions set (below we explain how it works
exactly; this is a key point in fast functions comparison).</p>
<p><tt class="docutils literal"><span class="pre">Deferred</span></tt> – merging process can affect bodies of functions that are in
<tt class="docutils literal"><span class="pre">FnTree</span></tt> already. Obviously such functions should be rechecked again. In this
case we remove them from <tt class="docutils literal"><span class="pre">FnTree</span></tt>, and mark them as to be rescanned, namely
put them into <tt class="docutils literal"><span class="pre">Deferred</span></tt> list.</p>
<div class="section" id="runonmodule">
<h5><a class="toc-backref" href="#id13">runOnModule</a><a class="headerlink" href="#runonmodule" title="Permalink to this headline">¶</a></h5>
<p>The algorithm is pretty simple:</p>
<ol class="arabic simple">
<li>Put all module’s functions into the <em>worklist</em>.</li>
</ol>
<p>2. Scan <em>worklist</em>‘s functions twice: first enumerate only strong functions and
then only weak ones:</p>
<blockquote>
<div>2.1. Loop body: take function from <em>worklist</em> (call it <em>FCur</em>) and try to
insert it into <em>FnTree</em>: check whether <em>FCur</em> is equal to one of functions
in <em>FnTree</em>. If there <em>is</em> equal function in <em>FnTree</em> (call it <em>FExists</em>):
merge function <em>FCur</em> with <em>FExists</em>. Otherwise add function from <em>worklist</em>
to <em>FnTree</em>.</div></blockquote>
<p>3. Once <em>worklist</em> scanning and merging operations is complete, check <em>Deferred</em>
list. If it is not empty: refill <em>worklist</em> contents with <em>Deferred</em> list and
do step 2 again, if <em>Deferred</em> is empty, then exit from method.</p>
</div>
<div class="section" id="comparison-and-logarithmical-search">
<h5><a class="toc-backref" href="#id14">Comparison and logarithmical search</a><a class="headerlink" href="#comparison-and-logarithmical-search" title="Permalink to this headline">¶</a></h5>
<p>Let’s recall our task: for every function <em>F</em> from module <em>M</em>, we have to find
equal functions <em>F`</em> in shortest time, and merge them into the single function.</p>
<p>Defining total ordering among the functions set allows to organize functions
into the binary tree. The lookup procedure complexity would be estimated as
O(log(N)) in this case. But how to define <em>total-ordering</em>?</p>
<p>We have to introduce a single rule applicable to every pair of functions, and
following this rule then evaluate which of them is greater. What kind of rule
it could be? Let’s declare it as “compare” method, that returns one of 3
possible values:</p>
<p>-1, left is <em>less</em> than right,</p>
<p>0, left and right are <em>equal</em>,</p>
<p>1, left is <em>greater</em> than right.</p>
<p>Of course it means, that we have to maintain
<em>strict and non-strict order relation properties</em>:</p>
<ul class="simple">
<li>reflexivity (<tt class="docutils literal"><span class="pre">a</span> <span class="pre"><=</span> <span class="pre">a</span></tt>, <tt class="docutils literal"><span class="pre">a</span> <span class="pre">==</span> <span class="pre">a</span></tt>, <tt class="docutils literal"><span class="pre">a</span> <span class="pre">>=</span> <span class="pre">a</span></tt>),</li>
<li>antisymmetry (if <tt class="docutils literal"><span class="pre">a</span> <span class="pre"><=</span> <span class="pre">b</span></tt> and <tt class="docutils literal"><span class="pre">b</span> <span class="pre"><=</span> <span class="pre">a</span></tt> then <tt class="docutils literal"><span class="pre">a</span> <span class="pre">==</span> <span class="pre">b</span></tt>),</li>
<li>transitivity (<tt class="docutils literal"><span class="pre">a</span> <span class="pre"><=</span> <span class="pre">b</span></tt> and <tt class="docutils literal"><span class="pre">b</span> <span class="pre"><=</span> <span class="pre">c</span></tt>, then <tt class="docutils literal"><span class="pre">a</span> <span class="pre"><=</span> <span class="pre">c</span></tt>)</li>
<li>asymmetry (if <tt class="docutils literal"><span class="pre">a</span> <span class="pre"><</span> <span class="pre">b</span></tt>, then <tt class="docutils literal"><span class="pre">a</span> <span class="pre">></span> <span class="pre">b</span></tt> or <tt class="docutils literal"><span class="pre">a</span> <span class="pre">==</span> <span class="pre">b</span></tt>).</li>
</ul>
<p>As it was mentioned before, comparison routine consists of
“sub-comparison-routines”, each of them also consists
“sub-comparison-routines”, and so on, finally it ends up with a primitives
comparison.</p>
<p>Below, we will use the next operations:</p>
<ol class="arabic simple">
<li><tt class="docutils literal"><span class="pre">cmpNumbers(number1,</span> <span class="pre">number2)</span></tt> is method that returns -1 if left is less
than right; 0, if left and right are equal; and 1 otherwise.</li>
<li><tt class="docutils literal"><span class="pre">cmpFlags(flag1,</span> <span class="pre">flag2)</span></tt> is hypothetical method that compares two flags.
The logic is the same as in <tt class="docutils literal"><span class="pre">cmpNumbers</span></tt>, where <tt class="docutils literal"><span class="pre">true</span></tt> is 1, and
<tt class="docutils literal"><span class="pre">false</span></tt> is 0.</li>
</ol>
<p>The rest of article is based on <em>MergeFunctions.cpp</em> source code
(<em><llvm_dir>/lib/Transforms/IPO/MergeFunctions.cpp</em>). We would like to ask
reader to keep this file open nearby, so we could use it as a reference for
further explanations.</p>
<p>Now we’re ready to proceed to the next chapter and see how it works.</p>
</div>
</div>
</div>
</div>
<div class="section" id="functions-comparison">
<h2><a class="toc-backref" href="#id15">Functions comparison</a><a class="headerlink" href="#functions-comparison" title="Permalink to this headline">¶</a></h2>
<p>At first, let’s define how exactly we compare complex objects.</p>
<p>Complex objects comparison (function, basic-block, etc) is mostly based on its
sub-objects comparison results. So it is similar to the next “tree” objects
comparison:</p>
<ol class="arabic simple">
<li>For two trees <em>T1</em> and <em>T2</em> we perform <em>depth-first-traversal</em> and have
two sequences as a product: “<em>T1Items</em>” and “<em>T2Items</em>”.</li>
<li>Then compare chains “<em>T1Items</em>” and “<em>T2Items</em>” in
most-significant-item-first order. Result of items comparison would be the
result of <em>T1</em> and <em>T2</em> comparison itself.</li>
</ol>
<div class="section" id="functioncomparator-compare-void">
<h3><a class="toc-backref" href="#id16">FunctionComparator::compare(void)</a><a class="headerlink" href="#functioncomparator-compare-void" title="Permalink to this headline">¶</a></h3>
<p>Brief look at the source code tells us, that comparison starts in
“<tt class="docutils literal"><span class="pre">int</span> <span class="pre">FunctionComparator::compare(void)</span></tt>” method.</p>
<p>1. First parts to be compared are function’s attributes and some properties that
outsides “attributes” term, but still could make function different without
changing its body. This part of comparison is usually done within simple
<em>cmpNumbers</em> or <em>cmpFlags</em> operations (e.g.
<tt class="docutils literal"><span class="pre">cmpFlags(F1->hasGC(),</span> <span class="pre">F2->hasGC())</span></tt>). Below is full list of function’s
properties to be compared on this stage:</p>
<blockquote>
<div><ul class="simple">
<li><em>Attributes</em> (those are returned by <tt class="docutils literal"><span class="pre">Function::getAttributes()</span></tt>
method).</li>
<li><em>GC</em>, for equivalence, <em>RHS</em> and <em>LHS</em> should be both either without
<em>GC</em> or with the same one.</li>
<li><em>Section</em>, just like a <em>GC</em>: <em>RHS</em> and <em>LHS</em> should be defined in the
same section.</li>
<li><em>Variable arguments</em>. <em>LHS</em> and <em>RHS</em> should be both either with or
without <em>var-args</em>.</li>
<li><em>Calling convention</em> should be the same.</li>
</ul>
</div></blockquote>
<p>2. Function type. Checked by <tt class="docutils literal"><span class="pre">FunctionComparator::cmpType(Type*,</span> <span class="pre">Type*)</span></tt>
method. It checks return type and parameters type; the method itself will be
described later.</p>
<p>3. Associate function formal parameters with each other. Then comparing function
bodies, if we see the usage of <em>LHS</em>‘s <em>i</em>-th argument in <em>LHS</em>‘s body, then,
we want to see usage of <em>RHS</em>‘s <em>i</em>-th argument at the same place in <em>RHS</em>‘s
body, otherwise functions are different. On this stage we grant the preference
to those we met later in function body (value we met first would be <em>less</em>).
This is done by “<tt class="docutils literal"><span class="pre">FunctionComparator::cmpValues(const</span> <span class="pre">Value*,</span> <span class="pre">const</span> <span class="pre">Value*)</span></tt>”
method (will be described a bit later).</p>
<ol class="arabic simple" start="4">
<li>Function body comparison. As it written in method comments:</li>
</ol>
<p>“We do a CFG-ordered walk since the actual ordering of the blocks in the linked
list is immaterial. Our walk starts at the entry block for both functions, then
takes each block from each terminator in order. As an artifact, this also means
that unreachable blocks are ignored.”</p>
<p>So, using this walk we get BBs from <em>left</em> and <em>right</em> in the same order, and
compare them by “<tt class="docutils literal"><span class="pre">FunctionComparator::compare(const</span> <span class="pre">BasicBlock*,</span> <span class="pre">const</span>
<span class="pre">BasicBlock*)</span></tt>” method.</p>
<p>We also associate BBs with each other, like we did it with function formal
arguments (see <tt class="docutils literal"><span class="pre">cmpValues</span></tt> method below).</p>
</div>
<div class="section" id="functioncomparator-cmptype">
<h3><a class="toc-backref" href="#id17">FunctionComparator::cmpType</a><a class="headerlink" href="#functioncomparator-cmptype" title="Permalink to this headline">¶</a></h3>
<p>Consider how types comparison works.</p>
<p>1. Coerce pointer to integer. If left type is a pointer, try to coerce it to the
integer type. It could be done if its address space is 0, or if address spaces
are ignored at all. Do the same thing for the right type.</p>
<p>2. If left and right types are equal, return 0. Otherwise we need to give
preference to one of them. So proceed to the next step.</p>
<p>3. If types are of different kind (different type IDs). Return result of type
IDs comparison, treating them as a numbers (use <tt class="docutils literal"><span class="pre">cmpNumbers</span></tt> operation).</p>
<p>4. If types are vectors or integers, return result of their pointers comparison,
comparing them as numbers.</p>
<ol class="arabic" start="5">
<li><p class="first">Check whether type ID belongs to the next group (call it equivalent-group):</p>
<ul class="simple">
<li>Void</li>
<li>Float</li>
<li>Double</li>
<li>X86_FP80</li>
<li>FP128</li>
<li>PPC_FP128</li>
<li>Label</li>
<li>Metadata.</li>
</ul>
<p>If ID belongs to group above, return 0. Since it’s enough to see that
types has the same <tt class="docutils literal"><span class="pre">TypeID</span></tt>. No additional information is required.</p>
</li>
</ol>
<p>6. Left and right are pointers. Return result of address space comparison
(numbers comparison).</p>
<p>7. Complex types (structures, arrays, etc.). Follow complex objects comparison
technique (see the very first paragraph of this chapter). Both <em>left</em> and
<em>right</em> are to be expanded and their element types will be checked the same
way. If we get -1 or 1 on some stage, return it. Otherwise return 0.</p>
<p>8. Steps 1-6 describe all the possible cases, if we passed steps 1-6 and didn’t
get any conclusions, then invoke <tt class="docutils literal"><span class="pre">llvm_unreachable</span></tt>, since it’s quite
unexpectable case.</p>
</div>
<div class="section" id="cmpvalues-const-value-const-value">
<h3><a class="toc-backref" href="#id18">cmpValues(const Value*, const Value*)</a><a class="headerlink" href="#cmpvalues-const-value-const-value" title="Permalink to this headline">¶</a></h3>
<p>Method that compares local values.</p>
<p>This method gives us an answer on a very curious quesion: whether we could treat
local values as equal, and which value is greater otherwise. It’s better to
start from example:</p>
<p>Consider situation when we’re looking at the same place in left function “<em>FL</em>”
and in right function “<em>FR</em>”. And every part of <em>left</em> place is equal to the
corresponding part of <em>right</em> place, and (!) both parts use <em>Value</em> instances,
for example:</p>
<div class="highlight-llvm"><div class="highlight"><pre>instr0 i32 %LV ; left side, function FL
instr0 i32 %RV ; right side, function FR
</pre></div>
</div>
<p>So, now our conclusion depends on <em>Value</em> instances comparison.</p>
<p>Main purpose of this method is to determine relation between such values.</p>
<p>What we expect from equal functions? At the same place, in functions “<em>FL</em>” and
“<em>FR</em>” we expect to see <em>equal</em> values, or values <em>defined</em> at the same place
in “<em>FL</em>” and “<em>FR</em>”.</p>
<p>Consider small example here:</p>
<div class="highlight-llvm"><div class="highlight"><pre>define void %f(i32 %pf0, i32 %pf1) {
instr0 i32 %pf0 instr1 i32 %pf1 instr2 i32 123
}
</pre></div>
</div>
<div class="highlight-llvm"><div class="highlight"><pre>define void %g(i32 %pg0, i32 %pg1) {
instr0 i32 %pg0 instr1 i32 %pg0 instr2 i32 123
}
</pre></div>
</div>
<p>In this example, <em>pf0</em> is associated with <em>pg0</em>, <em>pf1</em> is associated with <em>pg1</em>,
and we also declare that <em>pf0</em> < <em>pf1</em>, and thus <em>pg0</em> < <em>pf1</em>.</p>
<p>Instructions with opcode “<em>instr0</em>” would be <em>equal</em>, since their types and
opcodes are equal, and values are <em>associated</em>.</p>
<p>Instruction with opcode “<em>instr1</em>” from <em>f</em> is <em>greater</em> than instruction with
opcode “<em>instr1</em>” from <em>g</em>; here we have equal types and opcodes, but “<em>pf1</em> is
greater than “<em>pg0</em>”.</p>
<p>And instructions with opcode “<em>instr2</em>” are equal, because their opcodes and
types are equal, and the same constant is used as a value.</p>
<div class="section" id="what-we-assiciate-in-cmpvalues">
<h4><a class="toc-backref" href="#id19">What we assiciate in cmpValues?</a><a class="headerlink" href="#what-we-assiciate-in-cmpvalues" title="Permalink to this headline">¶</a></h4>
<ul class="simple">
<li>Function arguments. <em>i</em>-th argument from left function associated with
<em>i</em>-th argument from right function.</li>
<li>BasicBlock instances. In basic-block enumeration loop we associate <em>i</em>-th
BasicBlock from the left function with <em>i</em>-th BasicBlock from the right
function.</li>
<li>Instructions.</li>
<li>Instruction operands. Note, we can meet <em>Value</em> here we have never seen
before. In this case it is not a function argument, nor <em>BasicBlock</em>, nor
<em>Instruction</em>. It is global value. It is constant, since its the only
supposed global here. Method also compares:</li>
<li>Constants that are of the same type.</li>
<li>If right constant could be losslessly bit-casted to the left one, then we
also compare them.</li>
</ul>
</div>
<div class="section" id="how-to-implement-cmpvalues">
<h4><a class="toc-backref" href="#id20">How to implement cmpValues?</a><a class="headerlink" href="#how-to-implement-cmpvalues" title="Permalink to this headline">¶</a></h4>
<p><em>Association</em> is a case of equality for us. We just treat such values as equal.
But, in general, we need to implement antisymmetric relation. As it was
mentioned above, to understand what is <em>less</em>, we can use order in which we
meet values. If both of values has the same order in function (met at the same
time), then treat values as <em>associated</em>. Otherwise – it depends on who was
first.</p>
<p>Every time we run top-level compare method, we initialize two identical maps
(one for the left side, another one for the right side):</p>
<p><tt class="docutils literal"><span class="pre">map<Value,</span> <span class="pre">int></span> <span class="pre">sn_mapL,</span> <span class="pre">sn_mapR;</span></tt></p>
<p>The key of the map is the <em>Value</em> itself, the <em>value</em> – is its order (call it
<em>serial number</em>).</p>
<p>To add value <em>V</em> we need to perform the next procedure:</p>
<p><tt class="docutils literal"><span class="pre">sn_map.insert(std::make_pair(V,</span> <span class="pre">sn_map.size()));</span></tt></p>
<p>For the first <em>Value</em>, map will return <em>0</em>, for second <em>Value</em> map will return
<em>1</em>, and so on.</p>
<p>Then we can check whether left and right values met at the same time with simple
comparison:</p>
<p><tt class="docutils literal"><span class="pre">cmpNumbers(sn_mapL[Left],</span> <span class="pre">sn_mapR[Right]);</span></tt></p>
<p>Of course, we can combine insertion and comparison:</p>
<div class="highlight-c++"><div class="highlight"><pre><span class="n">std</span><span class="o">::</span><span class="n">pair</span><span class="o"><</span><span class="n">iterator</span><span class="p">,</span> <span class="kt">bool</span><span class="o">></span>
<span class="n">LeftRes</span> <span class="o">=</span> <span class="n">sn_mapL</span><span class="p">.</span><span class="n">insert</span><span class="p">(</span><span class="n">std</span><span class="o">::</span><span class="n">make_pair</span><span class="p">(</span><span class="n">Left</span><span class="p">,</span> <span class="n">sn_mapL</span><span class="p">.</span><span class="n">size</span><span class="p">())),</span> <span class="n">RightRes</span>
<span class="o">=</span> <span class="n">sn_mapR</span><span class="p">.</span><span class="n">insert</span><span class="p">(</span><span class="n">std</span><span class="o">::</span><span class="n">make_pair</span><span class="p">(</span><span class="n">Right</span><span class="p">,</span> <span class="n">sn_mapR</span><span class="p">.</span><span class="n">size</span><span class="p">()));</span>
<span class="k">return</span> <span class="nf">cmpNumbers</span><span class="p">(</span><span class="n">LeftRes</span><span class="p">.</span><span class="n">first</span><span class="o">-></span><span class="n">second</span><span class="p">,</span> <span class="n">RightRes</span><span class="p">.</span><span class="n">first</span><span class="o">-></span><span class="n">second</span><span class="p">);</span>
</pre></div>
</div>
<p>Let’s look, how whole method could be implemented.</p>
<p>1. we have to start from the bad news. Consider function self and
cross-referencing cases:</p>
<div class="highlight-c++"><div class="highlight"><pre><span class="c1">// self-reference unsigned fact0(unsigned n) { return n > 1 ? n</span>
<span class="o">*</span> <span class="n">fact0</span><span class="p">(</span><span class="n">n</span><span class="o">-</span><span class="mi">1</span><span class="p">)</span> <span class="o">:</span> <span class="mi">1</span><span class="p">;</span> <span class="p">}</span> <span class="kt">unsigned</span> <span class="n">fact1</span><span class="p">(</span><span class="kt">unsigned</span> <span class="n">n</span><span class="p">)</span> <span class="p">{</span> <span class="k">return</span> <span class="n">n</span> <span class="o">></span> <span class="mi">1</span> <span class="o">?</span> <span class="n">n</span> <span class="o">*</span>
<span class="n">fact1</span><span class="p">(</span><span class="n">n</span><span class="o">-</span><span class="mi">1</span><span class="p">)</span> <span class="o">:</span> <span class="mi">1</span><span class="p">;</span> <span class="p">}</span>
<span class="c1">// cross-reference unsigned ping(unsigned n) { return n!= 0 ? pong(n-1) : 0;</span>
<span class="p">}</span> <span class="kt">unsigned</span> <span class="n">pong</span><span class="p">(</span><span class="kt">unsigned</span> <span class="n">n</span><span class="p">)</span> <span class="p">{</span> <span class="k">return</span> <span class="n">n</span><span class="o">!=</span> <span class="mi">0</span> <span class="o">?</span> <span class="n">ping</span><span class="p">(</span><span class="n">n</span><span class="o">-</span><span class="mi">1</span><span class="p">)</span> <span class="o">:</span> <span class="mi">0</span><span class="p">;</span> <span class="p">}</span>
</pre></div>
</div>
<blockquote>
<div>This comparison has been implemented in initial <em>MergeFunctions</em> pass
version. But, unfortunately, it is not transitive. And this is the only case
we can’t convert to less-equal-greater comparison. It is a seldom case, 4-5
functions of 10000 (checked on test-suite), and, we hope, reader would
forgive us for such a sacrifice in order to get the O(log(N)) pass time.</div></blockquote>
<p>2. If left/right <em>Value</em> is a constant, we have to compare them. Return 0 if it
is the same constant, or use <tt class="docutils literal"><span class="pre">cmpConstants</span></tt> method otherwise.</p>
<p>3. If left/right is <em>InlineAsm</em> instance. Return result of <em>Value</em> pointers
comparison.</p>
<p>4. Explicit association of <em>L</em> (left value) and <em>R</em> (right value). We need to
find out whether values met at the same time, and thus are <em>associated</em>. Or we
need to put the rule: when we treat <em>L</em> < <em>R</em>. Now it is easy: just return
result of numbers comparison:</p>
<div class="highlight-c++"><div class="highlight"><pre><span class="n">std</span><span class="o">::</span><span class="n">pair</span><span class="o"><</span><span class="n">iterator</span><span class="p">,</span> <span class="kt">bool</span><span class="o">></span>
<span class="n">LeftRes</span> <span class="o">=</span> <span class="n">sn_mapL</span><span class="p">.</span><span class="n">insert</span><span class="p">(</span><span class="n">std</span><span class="o">::</span><span class="n">make_pair</span><span class="p">(</span><span class="n">Left</span><span class="p">,</span> <span class="n">sn_mapL</span><span class="p">.</span><span class="n">size</span><span class="p">())),</span>
<span class="n">RightRes</span> <span class="o">=</span> <span class="n">sn_mapR</span><span class="p">.</span><span class="n">insert</span><span class="p">(</span><span class="n">std</span><span class="o">::</span><span class="n">make_pair</span><span class="p">(</span><span class="n">Right</span><span class="p">,</span> <span class="n">sn_mapR</span><span class="p">.</span><span class="n">size</span><span class="p">()));</span>
<span class="k">if</span> <span class="p">(</span><span class="n">LeftRes</span><span class="p">.</span><span class="n">first</span><span class="o">-></span><span class="n">second</span> <span class="o">==</span> <span class="n">RightRes</span><span class="p">.</span><span class="n">first</span><span class="o">-></span><span class="n">second</span><span class="p">)</span> <span class="k">return</span> <span class="mi">0</span><span class="p">;</span>
<span class="k">if</span> <span class="p">(</span><span class="n">LeftRes</span><span class="p">.</span><span class="n">first</span><span class="o">-></span><span class="n">second</span> <span class="o"><</span> <span class="n">RightRes</span><span class="p">.</span><span class="n">first</span><span class="o">-></span><span class="n">second</span><span class="p">)</span> <span class="k">return</span> <span class="o">-</span><span class="mi">1</span><span class="p">;</span>
<span class="k">return</span> <span class="mi">1</span><span class="p">;</span>
</pre></div>
</div>
<p>Now when <em>cmpValues</em> returns 0, we can proceed comparison procedure. Otherwise,
if we get (-1 or 1), we need to pass this result to the top level, and finish
comparison procedure.</p>
</div>
</div>
<div class="section" id="cmpconstants">
<h3><a class="toc-backref" href="#id21">cmpConstants</a><a class="headerlink" href="#cmpconstants" title="Permalink to this headline">¶</a></h3>
<p>Performs constants comparison as follows:</p>
<p>1. Compare constant types using <tt class="docutils literal"><span class="pre">cmpType</span></tt> method. If result is -1 or 1, goto
step 2, otherwise proceed to step 3.</p>
<p>2. If types are different, we still can check whether constants could be
losslessly bitcasted to each other. The further explanation is modification of
<tt class="docutils literal"><span class="pre">canLosslesslyBitCastTo</span></tt> method.</p>
<blockquote>
<div><p>2.1 Check whether constants are of the first class types
(<tt class="docutils literal"><span class="pre">isFirstClassType</span></tt> check):</p>
<p>2.1.1. If both constants are <em>not</em> of the first class type: return result
of <tt class="docutils literal"><span class="pre">cmpType</span></tt>.</p>
<p>2.1.2. Otherwise, if left type is not of the first class, return -1. If
right type is not of the first class, return 1.</p>
<p>2.1.3. If both types are of the first class type, proceed to the next step
(2.1.3.1).</p>
<p>2.1.3.1. If types are vectors, compare their bitwidth using the
<em>cmpNumbers</em>. If result is not 0, return it.</p>
<p>2.1.3.2. Different types, but not a vectors:</p>
<ul class="simple">
<li>if both of them are pointers, good for us, we can proceed to step 3.</li>
<li>if one of types is pointer, return result of <em>isPointer</em> flags
comparison (<em>cmpFlags</em> operation).</li>
<li>otherwise we have no methods to prove bitcastability, and thus return
result of types comparison (-1 or 1).</li>
</ul>
</div></blockquote>
<p>Steps below are for the case when types are equal, or case when constants are
bitcastable:</p>
<p>3. One of constants is a “<em>null</em>” value. Return the result of
<tt class="docutils literal"><span class="pre">cmpFlags(L->isNullValue,</span> <span class="pre">R->isNullValue)</span></tt> comparison.</p>
<ol class="arabic simple" start="4">
<li>Compare value IDs, and return result if it is not 0:</li>
</ol>
<div class="highlight-c++"><div class="highlight"><pre><span class="k">if</span> <span class="p">(</span><span class="kt">int</span> <span class="n">Res</span> <span class="o">=</span> <span class="n">cmpNumbers</span><span class="p">(</span><span class="n">L</span><span class="o">-></span><span class="n">getValueID</span><span class="p">(),</span> <span class="n">R</span><span class="o">-></span><span class="n">getValueID</span><span class="p">()))</span>
<span class="k">return</span> <span class="n">Res</span><span class="p">;</span>
</pre></div>
</div>
<p>5. Compare the contents of constants. The comparison depends on kind of
constants, but on this stage it is just a lexicographical comparison. Just see
how it was described in the beginning of “<em>Functions comparison</em>” paragraph.
Mathematically it is equal to the next case: we encode left constant and right
constant (with similar way <em>bitcode-writer</em> does). Then compare left code
sequence and right code sequence.</p>
</div>
<div class="section" id="compare-const-basicblock-const-basicblock">
<h3><a class="toc-backref" href="#id22">compare(const BasicBlock*, const BasicBlock*)</a><a class="headerlink" href="#compare-const-basicblock-const-basicblock" title="Permalink to this headline">¶</a></h3>
<p>Compares two <em>BasicBlock</em> instances.</p>
<p>It enumerates instructions from left <em>BB</em> and right <em>BB</em>.</p>
<p>1. It assigns serial numbers to the left and right instructions, using
<tt class="docutils literal"><span class="pre">cmpValues</span></tt> method.</p>
<p>2. If one of left or right is <em>GEP</em> (<tt class="docutils literal"><span class="pre">GetElementPtr</span></tt>), then treat <em>GEP</em> as
greater than other instructions, if both instructions are <em>GEPs</em> use <tt class="docutils literal"><span class="pre">cmpGEP</span></tt>
method for comparison. If result is -1 or 1, pass it to the top-level
comparison (return it).</p>
<blockquote>
<div><p>3.1. Compare operations. Call <tt class="docutils literal"><span class="pre">cmpOperation</span></tt> method. If result is -1 or
1, return it.</p>
<p>3.2. Compare number of operands, if result is -1 or 1, return it.</p>
<p>3.3. Compare operands themselves, use <tt class="docutils literal"><span class="pre">cmpValues</span></tt> method. Return result
if it is -1 or 1.</p>
<p>3.4. Compare type of operands, using <tt class="docutils literal"><span class="pre">cmpType</span></tt> method. Return result if
it is -1 or 1.</p>
<p>3.5. Proceed to the next instruction.</p>
</div></blockquote>
<ol class="arabic" start="4">
<li><p class="first">We can finish instruction enumeration in 3 cases:</p>
<p>4.1. We reached the end of both left and right basic-blocks. We didn’t
exit on steps 1-3, so contents is equal, return 0.</p>
<p>4.2. We have reached the end of the left basic-block. Return -1.</p>
<p>4.3. Return 1 (the end of the right basic block).</p>
</li>
</ol>
</div>
<div class="section" id="cmpgep">
<h3><a class="toc-backref" href="#id23">cmpGEP</a><a class="headerlink" href="#cmpgep" title="Permalink to this headline">¶</a></h3>
<p>Compares two GEPs (<tt class="docutils literal"><span class="pre">getelementptr</span></tt> instructions).</p>
<p>It differs from regular operations comparison with the only thing: possibility
to use <tt class="docutils literal"><span class="pre">accumulateConstantOffset</span></tt> method.</p>
<p>So, if we get constant offset for both left and right <em>GEPs</em>, then compare it as
numbers, and return comparison result.</p>
<p>Otherwise treat it like a regular operation (see previous paragraph).</p>
</div>
<div class="section" id="cmpoperation">
<h3><a class="toc-backref" href="#id24">cmpOperation</a><a class="headerlink" href="#cmpoperation" title="Permalink to this headline">¶</a></h3>
<p>Compares instruction opcodes and some important operation properties.</p>
<ol class="arabic simple">
<li>Compare opcodes, if it differs return the result.</li>
<li>Compare number of operands. If it differs – return the result.</li>
</ol>
<p>3. Compare operation types, use <em>cmpType</em>. All the same – if types are
different, return result.</p>
<p>4. Compare <em>subclassOptionalData</em>, get it with <tt class="docutils literal"><span class="pre">getRawSubclassOptionalData</span></tt>
method, and compare it like a numbers.</p>
<ol class="arabic simple" start="5">
<li>Compare operand types.</li>
</ol>
<p>6. For some particular instructions check equivalence (relation in our case) of
some significant attributes. For example we have to compare alignment for
<tt class="docutils literal"><span class="pre">load</span></tt> instructions.</p>
</div>
<div class="section" id="o-log-n">
<h3><a class="toc-backref" href="#id25">O(log(N))</a><a class="headerlink" href="#o-log-n" title="Permalink to this headline">¶</a></h3>
<p>Methods described above implement order relationship. And latter, could be used
for nodes comparison in a binary tree. So we can organize functions set into
the binary tree and reduce the cost of lookup procedure from
O(N*N) to O(log(N)).</p>
</div>
</div>
<div class="section" id="merging-process-mergetwofunctions">
<h2><a class="toc-backref" href="#id26">Merging process, mergeTwoFunctions</a><a class="headerlink" href="#merging-process-mergetwofunctions" title="Permalink to this headline">¶</a></h2>
<p>Once <em>MergeFunctions</em> detected that current function (<em>G</em>) is equal to one that
were analyzed before (function <em>F</em>) it calls <tt class="docutils literal"><span class="pre">mergeTwoFunctions(Function*,</span>
<span class="pre">Function*)</span></tt>.</p>
<p>Operation affects <tt class="docutils literal"><span class="pre">FnTree</span></tt> contents with next way: <em>F</em> will stay in
<tt class="docutils literal"><span class="pre">FnTree</span></tt>. <em>G</em> being equal to <em>F</em> will not be added to <tt class="docutils literal"><span class="pre">FnTree</span></tt>. Calls of
<em>G</em> would be replaced with something else. It changes bodies of callers. So,
functions that calls <em>G</em> would be put into <tt class="docutils literal"><span class="pre">Deferred</span></tt> set and removed from
<tt class="docutils literal"><span class="pre">FnTree</span></tt>, and analyzed again.</p>
<p>The approach is next:</p>
<p>1. Most wished case: when we can use alias and both of <em>F</em> and <em>G</em> are weak. We
make both of them with aliases to the third strong function <em>H</em>. Actually <em>H</em>
is <em>F</em>. See below how it’s made (but it’s better to look straight into the
source code). Well, this is a case when we can just replace <em>G</em> with <em>F</em>
everywhere, we use <tt class="docutils literal"><span class="pre">replaceAllUsesWith</span></tt> operation here (<em>RAUW</em>).</p>
<p>2. <em>F</em> could not be overridden, while <em>G</em> could. It would be good to do the
next: after merging the places where overridable function were used, still use
overridable stub. So try to make <em>G</em> alias to <em>F</em>, or create overridable tail
call wrapper around <em>F</em> and replace <em>G</em> with that call.</p>
<p>3. Neither <em>F</em> nor <em>G</em> could be overridden. We can’t use <em>RAUW</em>. We can just
change the callers: call <em>F</em> instead of <em>G</em>. That’s what
<tt class="docutils literal"><span class="pre">replaceDirectCallers</span></tt> does.</p>
<p>Below is detailed body description.</p>
<div class="section" id="if-f-may-be-overridden">
<h3><a class="toc-backref" href="#id27">If “F” may be overridden</a><a class="headerlink" href="#if-f-may-be-overridden" title="Permalink to this headline">¶</a></h3>
<p>As follows from <tt class="docutils literal"><span class="pre">mayBeOverridden</span></tt> comments: “whether the definition of this
global may be replaced by something non-equivalent at link time”. If so, thats
ok: we can use alias to <em>F</em> instead of <em>G</em> or change call instructions itself.</p>
<div class="section" id="hasglobalaliases-removeusers">
<h4><a class="toc-backref" href="#id28">HasGlobalAliases, removeUsers</a><a class="headerlink" href="#hasglobalaliases-removeusers" title="Permalink to this headline">¶</a></h4>
<p>First consider the case when we have global aliases of one function name to
another. Our purpose is make both of them with aliases to the third strong
function. Though if we keep <em>F</em> alive and without major changes we can leave it
in <tt class="docutils literal"><span class="pre">FnTree</span></tt>. Try to combine these two goals.</p>
<p>Do stub replacement of <em>F</em> itself with an alias to <em>F</em>.</p>
<p>1. Create stub function <em>H</em>, with the same name and attributes like function
<em>F</em>. It takes maximum alignment of <em>F</em> and <em>G</em>.</p>
<p>2. Replace all uses of function <em>F</em> with uses of function <em>H</em>. It is the two
steps procedure instead. First of all, we must take into account, all functions
from whom <em>F</em> is called would be changed: since we change the call argument
(from <em>F</em> to <em>H</em>). If so we must to review these caller functions again after
this procedure. We remove callers from <tt class="docutils literal"><span class="pre">FnTree</span></tt>, method with name
<tt class="docutils literal"><span class="pre">removeUsers(F)</span></tt> does that (don’t confuse with <tt class="docutils literal"><span class="pre">replaceAllUsesWith</span></tt>):</p>
<blockquote>
<div><p>2.1. <tt class="docutils literal"><span class="pre">Inside</span> <span class="pre">removeUsers(Value*</span>
<span class="pre">V)</span></tt> we go through the all values that use value <em>V</em> (or <em>F</em> in our context).
If value is instruction, we go to function that holds this instruction and
mark it as to-be-analyzed-again (put to <tt class="docutils literal"><span class="pre">Deferred</span></tt> set), we also remove
caller from <tt class="docutils literal"><span class="pre">FnTree</span></tt>.</p>
<p>2.2. Now we can do the replacement: call <tt class="docutils literal"><span class="pre">F->replaceAllUsesWith(H)</span></tt>.</p>
</div></blockquote>
<p>3. <em>H</em> (that now “officially” plays <em>F</em>‘s role) is replaced with alias to <em>F</em>.
Do the same with <em>G</em>: replace it with alias to <em>F</em>. So finally everywhere <em>F</em>
was used, we use <em>H</em> and it is alias to <em>F</em>, and everywhere <em>G</em> was used we
also have alias to <em>F</em>.</p>
<ol class="arabic simple" start="4">
<li>Set <em>F</em> linkage to private. Make it strong :-)</li>
</ol>
</div>
<div class="section" id="no-global-aliases-replacedirectcallers">
<h4><a class="toc-backref" href="#id29">No global aliases, replaceDirectCallers</a><a class="headerlink" href="#no-global-aliases-replacedirectcallers" title="Permalink to this headline">¶</a></h4>
<p>If global aliases are not supported. We call <tt class="docutils literal"><span class="pre">replaceDirectCallers</span></tt> then. Just
go through all calls of <em>G</em> and replace it with calls of <em>F</em>. If you look into
method you will see that it scans all uses of <em>G</em> too, and if use is callee (if
user is call instruction and <em>G</em> is used as what to be called), we replace it
with use of <em>F</em>.</p>
<div class="section" id="if-f-could-not-be-overridden-fix-it">
<h5><a class="toc-backref" href="#id30">If “F” could not be overridden, fix it!</a><a class="headerlink" href="#if-f-could-not-be-overridden-fix-it" title="Permalink to this headline">¶</a></h5>
<p>We call <tt class="docutils literal"><span class="pre">writeThunkOrAlias(Function</span> <span class="pre">*F,</span> <span class="pre">Function</span> <span class="pre">*G)</span></tt>. Here we try to replace
<em>G</em> with alias to <em>F</em> first. Next conditions are essential:</p>
<ul class="simple">
<li>target should support global aliases,</li>
<li>the address itself of <em>G</em> should be not significant, not named and not
referenced anywhere,</li>
<li>function should come with external, local or weak linkage.</li>
</ul>
<p>Otherwise we write thunk: some wrapper that has <em>G’s</em> interface and calls <em>F</em>,
so <em>G</em> could be replaced with this wrapper.</p>
<p><em>writeAlias</em></p>
<p>As follows from <em>llvm</em> reference:</p>
<p>“Aliases act as <em>second name</em> for the aliasee value”. So we just want to create
second name for <em>F</em> and use it instead of <em>G</em>:</p>
<ol class="arabic">
<li><p class="first">create global alias itself (<em>GA</em>),</p>
</li>
<li><p class="first">adjust alignment of <em>F</em> so it must be maximum of current and <em>G’s</em> alignment;</p>
</li>
<li><p class="first">replace uses of <em>G</em>:</p>
<p>3.1. first mark all callers of <em>G</em> as to-be-analyzed-again, using
<tt class="docutils literal"><span class="pre">removeUsers</span></tt> method (see chapter above),</p>
<p>3.2. call <tt class="docutils literal"><span class="pre">G->replaceAllUsesWith(GA)</span></tt>.</p>
</li>
<li><p class="first">Get rid of <em>G</em>.</p>
</li>
</ol>
<p><em>writeThunk</em></p>
<p>As it written in method comments:</p>
<p>“Replace G with a simple tail call to bitcast(F). Also replace direct uses of G
with bitcast(F). Deletes G.”</p>
<p>In general it does the same as usual when we want to replace callee, except the
first point:</p>
<p>1. We generate tail call wrapper around <em>F</em>, but with interface that allows use
it instead of <em>G</em>.</p>
<ol class="arabic simple" start="2">
<li>“As-usual”: <tt class="docutils literal"><span class="pre">removeUsers</span></tt> and <tt class="docutils literal"><span class="pre">replaceAllUsesWith</span></tt> then.</li>
<li>Get rid of <em>G</em>.</li>
</ol>
</div>
</div>
</div>
</div>
<div class="section" id="that-s-it">
<h2><a class="toc-backref" href="#id31">That’s it.</a><a class="headerlink" href="#that-s-it" title="Permalink to this headline">¶</a></h2>
<p>We have described how to detect equal functions, and how to merge them, and in
first chapter we have described how it works all-together. Author hopes, reader
have some picture from now, and it helps him improve and debug this pass.</p>
<p>Reader is welcomed to send us any questions and proposals ;-)</p>
</div>
</div>
</div>
</div>
<div class="clearer"></div>
</div>
<div class="related">
<h3>Navigation</h3>
<ul>
<li class="right" style="margin-right: 10px">
<a href="genindex.html" title="General Index"
>index</a></li>
<li class="right" >
<a href="DeveloperPolicy.html" title="LLVM Developer Policy"
>next</a> |</li>
<li class="right" >
<a href="Statepoints.html" title="Garbage Collection Safepoints in LLVM"
>previous</a> |</li>
<li><a href="http://llvm.org/">LLVM Home</a> | </li>
<li><a href="index.html">Documentation</a>»</li>
</ul>
</div>
<div class="footer">
© Copyright 2003-2014, LLVM Project.
Last updated on 2018-07-24.
Created using <a href="http://sphinx-doc.org/">Sphinx</a> 1.2.2.
</div>
</body>
</html>
|