1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" 2 "http://www.w3.org/TR/html4/strict.dtd"> 3<html> 4<head> 5 <meta http-equiv="Content-Type" Content="text/html; charset=UTF-8" > 6 <title>Accurate Garbage Collection with LLVM</title> 7 <link rel="stylesheet" href="llvm.css" type="text/css"> 8 <style type="text/css"> 9 .rowhead { text-align: left; background: inherit; } 10 .indent { padding-left: 1em; } 11 .optl { color: #BFBFBF; } 12 </style> 13</head> 14<body> 15 16<h1> 17 Accurate Garbage Collection with LLVM 18</h1> 19 20<ol> 21 <li><a href="#introduction">Introduction</a> 22 <ul> 23 <li><a href="#feature">Goals and non-goals</a></li> 24 </ul> 25 </li> 26 27 <li><a href="#quickstart">Getting started</a> 28 <ul> 29 <li><a href="#quickstart-compiler">In your compiler</a></li> 30 <li><a href="#quickstart-runtime">In your runtime library</a></li> 31 <li><a href="#shadow-stack">About the shadow stack</a></li> 32 </ul> 33 </li> 34 35 <li><a href="#core">Core support</a> 36 <ul> 37 <li><a href="#gcattr">Specifying GC code generation: 38 <tt>gc "..."</tt></a></li> 39 <li><a href="#gcroot">Identifying GC roots on the stack: 40 <tt>llvm.gcroot</tt></a></li> 41 <li><a href="#barriers">Reading and writing references in the heap</a> 42 <ul> 43 <li><a href="#gcwrite">Write barrier: <tt>llvm.gcwrite</tt></a></li> 44 <li><a href="#gcread">Read barrier: <tt>llvm.gcread</tt></a></li> 45 </ul> 46 </li> 47 </ul> 48 </li> 49 50 <li><a href="#plugin">Compiler plugin interface</a> 51 <ul> 52 <li><a href="#collector-algos">Overview of available features</a></li> 53 <li><a href="#stack-map">Computing stack maps</a></li> 54 <li><a href="#init-roots">Initializing roots to null: 55 <tt>InitRoots</tt></a></li> 56 <li><a href="#custom">Custom lowering of intrinsics: <tt>CustomRoots</tt>, 57 <tt>CustomReadBarriers</tt>, and <tt>CustomWriteBarriers</tt></a></li> 58 <li><a href="#safe-points">Generating safe points: 59 <tt>NeededSafePoints</tt></a></li> 60 <li><a href="#assembly">Emitting assembly code: 61 <tt>GCMetadataPrinter</tt></a></li> 62 </ul> 63 </li> 64 65 <li><a href="#runtime-impl">Implementing a collector runtime</a> 66 <ul> 67 <li><a href="#gcdescriptors">Tracing GC pointers from heap 68 objects</a></li> 69 </ul> 70 </li> 71 72 <li><a href="#references">References</a></li> 73 74</ol> 75 76<div class="doc_author"> 77 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a> and 78 Gordon Henriksen</p> 79</div> 80 81<!-- *********************************************************************** --> 82<h2> 83 <a name="introduction">Introduction</a> 84</h2> 85<!-- *********************************************************************** --> 86 87<div> 88 89<p>Garbage collection is a widely used technique that frees the programmer from 90having to know the lifetimes of heap objects, making software easier to produce 91and maintain. Many programming languages rely on garbage collection for 92automatic memory management. There are two primary forms of garbage collection: 93conservative and accurate.</p> 94 95<p>Conservative garbage collection often does not require any special support 96from either the language or the compiler: it can handle non-type-safe 97programming languages (such as C/C++) and does not require any special 98information from the compiler. The 99<a href="http://www.hpl.hp.com/personal/Hans_Boehm/gc/">Boehm collector</a> is 100an example of a state-of-the-art conservative collector.</p> 101 102<p>Accurate garbage collection requires the ability to identify all pointers in 103the program at run-time (which requires that the source-language be type-safe in 104most cases). Identifying pointers at run-time requires compiler support to 105locate all places that hold live pointer variables at run-time, including the 106<a href="#gcroot">processor stack and registers</a>.</p> 107 108<p>Conservative garbage collection is attractive because it does not require any 109special compiler support, but it does have problems. In particular, because the 110conservative garbage collector cannot <i>know</i> that a particular word in the 111machine is a pointer, it cannot move live objects in the heap (preventing the 112use of compacting and generational GC algorithms) and it can occasionally suffer 113from memory leaks due to integer values that happen to point to objects in the 114program. In addition, some aggressive compiler transformations can break 115conservative garbage collectors (though these seem rare in practice).</p> 116 117<p>Accurate garbage collectors do not suffer from any of these problems, but 118they can suffer from degraded scalar optimization of the program. In particular, 119because the runtime must be able to identify and update all pointers active in 120the program, some optimizations are less effective. In practice, however, the 121locality and performance benefits of using aggressive garbage collection 122techniques dominates any low-level losses.</p> 123 124<p>This document describes the mechanisms and interfaces provided by LLVM to 125support accurate garbage collection.</p> 126 127<!-- ======================================================================= --> 128<h3> 129 <a name="feature">Goals and non-goals</a> 130</h3> 131 132<div> 133 134<p>LLVM's intermediate representation provides <a href="#intrinsics">garbage 135collection intrinsics</a> that offer support for a broad class of 136collector models. For instance, the intrinsics permit:</p> 137 138<ul> 139 <li>semi-space collectors</li> 140 <li>mark-sweep collectors</li> 141 <li>generational collectors</li> 142 <li>reference counting</li> 143 <li>incremental collectors</li> 144 <li>concurrent collectors</li> 145 <li>cooperative collectors</li> 146</ul> 147 148<p>We hope that the primitive support built into the LLVM IR is sufficient to 149support a broad class of garbage collected languages including Scheme, ML, Java, 150C#, Perl, Python, Lua, Ruby, other scripting languages, and more.</p> 151 152<p>However, LLVM does not itself provide a garbage collector—this should 153be part of your language's runtime library. LLVM provides a framework for 154compile time <a href="#plugin">code generation plugins</a>. The role of these 155plugins is to generate code and data structures which conforms to the <em>binary 156interface</em> specified by the <em>runtime library</em>. This is similar to the 157relationship between LLVM and DWARF debugging info, for example. The 158difference primarily lies in the lack of an established standard in the domain 159of garbage collection—thus the plugins.</p> 160 161<p>The aspects of the binary interface with which LLVM's GC support is 162concerned are:</p> 163 164<ul> 165 <li>Creation of GC-safe points within code where collection is allowed to 166 execute safely.</li> 167 <li>Computation of the stack map. For each safe point in the code, object 168 references within the stack frame must be identified so that the 169 collector may traverse and perhaps update them.</li> 170 <li>Write barriers when storing object references to the heap. These are 171 commonly used to optimize incremental scans in generational 172 collectors.</li> 173 <li>Emission of read barriers when loading object references. These are 174 useful for interoperating with concurrent collectors.</li> 175</ul> 176 177<p>There are additional areas that LLVM does not directly address:</p> 178 179<ul> 180 <li>Registration of global roots with the runtime.</li> 181 <li>Registration of stack map entries with the runtime.</li> 182 <li>The functions used by the program to allocate memory, trigger a 183 collection, etc.</li> 184 <li>Computation or compilation of type maps, or registration of them with 185 the runtime. These are used to crawl the heap for object 186 references.</li> 187</ul> 188 189<p>In general, LLVM's support for GC does not include features which can be 190adequately addressed with other features of the IR and does not specify a 191particular binary interface. On the plus side, this means that you should be 192able to integrate LLVM with an existing runtime. On the other hand, it leaves 193a lot of work for the developer of a novel language. However, it's easy to get 194started quickly and scale up to a more sophisticated implementation as your 195compiler matures.</p> 196 197</div> 198 199</div> 200 201<!-- *********************************************************************** --> 202<h2> 203 <a name="quickstart">Getting started</a> 204</h2> 205<!-- *********************************************************************** --> 206 207<div> 208 209<p>Using a GC with LLVM implies many things, for example:</p> 210 211<ul> 212 <li>Write a runtime library or find an existing one which implements a GC 213 heap.<ol> 214 <li>Implement a memory allocator.</li> 215 <li>Design a binary interface for the stack map, used to identify 216 references within a stack frame on the machine stack.*</li> 217 <li>Implement a stack crawler to discover functions on the call stack.*</li> 218 <li>Implement a registry for global roots.</li> 219 <li>Design a binary interface for type maps, used to identify references 220 within heap objects.</li> 221 <li>Implement a collection routine bringing together all of the above.</li> 222 </ol></li> 223 <li>Emit compatible code from your compiler.<ul> 224 <li>Initialization in the main function.</li> 225 <li>Use the <tt>gc "..."</tt> attribute to enable GC code generation 226 (or <tt>F.setGC("...")</tt>).</li> 227 <li>Use <tt>@llvm.gcroot</tt> to mark stack roots.</li> 228 <li>Use <tt>@llvm.gcread</tt> and/or <tt>@llvm.gcwrite</tt> to 229 manipulate GC references, if necessary.</li> 230 <li>Allocate memory using the GC allocation routine provided by the 231 runtime library.</li> 232 <li>Generate type maps according to your runtime's binary interface.</li> 233 </ul></li> 234 <li>Write a compiler plugin to interface LLVM with the runtime library.*<ul> 235 <li>Lower <tt>@llvm.gcread</tt> and <tt>@llvm.gcwrite</tt> to appropriate 236 code sequences.*</li> 237 <li>Compile LLVM's stack map to the binary form expected by the 238 runtime.</li> 239 </ul></li> 240 <li>Load the plugin into the compiler. Use <tt>llc -load</tt> or link the 241 plugin statically with your language's compiler.*</li> 242 <li>Link program executables with the runtime.</li> 243</ul> 244 245<p>To help with several of these tasks (those indicated with a *), LLVM 246includes a highly portable, built-in ShadowStack code generator. It is compiled 247into <tt>llc</tt> and works even with the interpreter and C backends.</p> 248 249<!-- ======================================================================= --> 250<h3> 251 <a name="quickstart-compiler">In your compiler</a> 252</h3> 253 254<div> 255 256<p>To turn the shadow stack on for your functions, first call:</p> 257 258<div class="doc_code"><pre 259>F.setGC("shadow-stack");</pre></div> 260 261<p>for each function your compiler emits. Since the shadow stack is built into 262LLVM, you do not need to load a plugin.</p> 263 264<p>Your compiler must also use <tt>@llvm.gcroot</tt> as documented. 265Don't forget to create a root for each intermediate value that is generated 266when evaluating an expression. In <tt>h(f(), g())</tt>, the result of 267<tt>f()</tt> could easily be collected if evaluating <tt>g()</tt> triggers a 268collection.</p> 269 270<p>There's no need to use <tt>@llvm.gcread</tt> and <tt>@llvm.gcwrite</tt> over 271plain <tt>load</tt> and <tt>store</tt> for now. You will need them when 272switching to a more advanced GC.</p> 273 274</div> 275 276<!-- ======================================================================= --> 277<h3> 278 <a name="quickstart-runtime">In your runtime</a> 279</h3> 280 281<div> 282 283<p>The shadow stack doesn't imply a memory allocation algorithm. A semispace 284collector or building atop <tt>malloc</tt> are great places to start, and can 285be implemented with very little code.</p> 286 287<p>When it comes time to collect, however, your runtime needs to traverse the 288stack roots, and for this it needs to integrate with the shadow stack. Luckily, 289doing so is very simple. (This code is heavily commented to help you 290understand the data structure, but there are only 20 lines of meaningful 291code.)</p> 292 293<pre class="doc_code"> 294/// @brief The map for a single function's stack frame. One of these is 295/// compiled as constant data into the executable for each function. 296/// 297/// Storage of metadata values is elided if the %metadata parameter to 298/// @llvm.gcroot is null. 299struct FrameMap { 300 int32_t NumRoots; //< Number of roots in stack frame. 301 int32_t NumMeta; //< Number of metadata entries. May be < NumRoots. 302 const void *Meta[0]; //< Metadata for each root. 303}; 304 305/// @brief A link in the dynamic shadow stack. One of these is embedded in the 306/// stack frame of each function on the call stack. 307struct StackEntry { 308 StackEntry *Next; //< Link to next stack entry (the caller's). 309 const FrameMap *Map; //< Pointer to constant FrameMap. 310 void *Roots[0]; //< Stack roots (in-place array). 311}; 312 313/// @brief The head of the singly-linked list of StackEntries. Functions push 314/// and pop onto this in their prologue and epilogue. 315/// 316/// Since there is only a global list, this technique is not threadsafe. 317StackEntry *llvm_gc_root_chain; 318 319/// @brief Calls Visitor(root, meta) for each GC root on the stack. 320/// root and meta are exactly the values passed to 321/// <tt>@llvm.gcroot</tt>. 322/// 323/// Visitor could be a function to recursively mark live objects. Or it 324/// might copy them to another heap or generation. 325/// 326/// @param Visitor A function to invoke for every GC root on the stack. 327void visitGCRoots(void (*Visitor)(void **Root, const void *Meta)) { 328 for (StackEntry *R = llvm_gc_root_chain; R; R = R->Next) { 329 unsigned i = 0; 330 331 // For roots [0, NumMeta), the metadata pointer is in the FrameMap. 332 for (unsigned e = R->Map->NumMeta; i != e; ++i) 333 Visitor(&R->Roots[i], R->Map->Meta[i]); 334 335 // For roots [NumMeta, NumRoots), the metadata pointer is null. 336 for (unsigned e = R->Map->NumRoots; i != e; ++i) 337 Visitor(&R->Roots[i], NULL); 338 } 339}</pre> 340 341</div> 342 343<!-- ======================================================================= --> 344<h3> 345 <a name="shadow-stack">About the shadow stack</a> 346</h3> 347 348<div> 349 350<p>Unlike many GC algorithms which rely on a cooperative code generator to 351compile stack maps, this algorithm carefully maintains a linked list of stack 352roots [<a href="#henderson02">Henderson2002</a>]. This so-called "shadow stack" 353mirrors the machine stack. Maintaining this data structure is slower than using 354a stack map compiled into the executable as constant data, but has a significant 355portability advantage because it requires no special support from the target 356code generator, and does not require tricky platform-specific code to crawl 357the machine stack.</p> 358 359<p>The tradeoff for this simplicity and portability is:</p> 360 361<ul> 362 <li>High overhead per function call.</li> 363 <li>Not thread-safe.</li> 364</ul> 365 366<p>Still, it's an easy way to get started. After your compiler and runtime are 367up and running, writing a <a href="#plugin">plugin</a> will allow you to take 368advantage of <a href="#collector-algos">more advanced GC features</a> of LLVM 369in order to improve performance.</p> 370 371</div> 372 373</div> 374 375<!-- *********************************************************************** --> 376<h2> 377 <a name="core">IR features</a><a name="intrinsics"></a> 378</h2> 379<!-- *********************************************************************** --> 380 381<div> 382 383<p>This section describes the garbage collection facilities provided by the 384<a href="LangRef.html">LLVM intermediate representation</a>. The exact behavior 385of these IR features is specified by the binary interface implemented by a 386<a href="#plugin">code generation plugin</a>, not by this document.</p> 387 388<p>These facilities are limited to those strictly necessary; they are not 389intended to be a complete interface to any garbage collector. A program will 390need to interface with the GC library using the facilities provided by that 391program.</p> 392 393<!-- ======================================================================= --> 394<h3> 395 <a name="gcattr">Specifying GC code generation: <tt>gc "..."</tt></a> 396</h3> 397 398<div> 399 400<div class="doc_code"><tt> 401 define <i>ty</i> @<i>name</i>(...) <span style="text-decoration: underline">gc "<i>name</i>"</span> { ... 402</tt></div> 403 404<p>The <tt>gc</tt> function attribute is used to specify the desired GC style 405to the compiler. Its programmatic equivalent is the <tt>setGC</tt> method of 406<tt>Function</tt>.</p> 407 408<p>Setting <tt>gc "<i>name</i>"</tt> on a function triggers a search for a 409matching code generation plugin "<i>name</i>"; it is that plugin which defines 410the exact nature of the code generated to support GC. If none is found, the 411compiler will raise an error.</p> 412 413<p>Specifying the GC style on a per-function basis allows LLVM to link together 414programs that use different garbage collection algorithms (or none at all).</p> 415 416</div> 417 418<!-- ======================================================================= --> 419<h3> 420 <a name="gcroot">Identifying GC roots on the stack: <tt>llvm.gcroot</tt></a> 421</h3> 422 423<div> 424 425<div class="doc_code"><tt> 426 void @llvm.gcroot(i8** %ptrloc, i8* %metadata) 427</tt></div> 428 429<p>The <tt>llvm.gcroot</tt> intrinsic is used to inform LLVM that a stack 430variable references an object on the heap and is to be tracked for garbage 431collection. The exact impact on generated code is specified by a <a 432href="#plugin">compiler plugin</a>. All calls to <tt>llvm.gcroot</tt> <b>must</b> reside 433 inside the first basic block.</p> 434 435<p>A compiler which uses mem2reg to raise imperative code using <tt>alloca</tt> 436into SSA form need only add a call to <tt>@llvm.gcroot</tt> for those variables 437which a pointers into the GC heap.</p> 438 439<p>It is also important to mark intermediate values with <tt>llvm.gcroot</tt>. 440For example, consider <tt>h(f(), g())</tt>. Beware leaking the result of 441<tt>f()</tt> in the case that <tt>g()</tt> triggers a collection. Note, that 442stack variables must be initialized and marked with <tt>llvm.gcroot</tt> in 443function's prologue.</p> 444 445<p>The first argument <b>must</b> be a value referring to an alloca instruction 446or a bitcast of an alloca. The second contains a pointer to metadata that 447should be associated with the pointer, and <b>must</b> be a constant or global 448value address. If your target collector uses tags, use a null pointer for 449metadata.</p> 450 451<p>The <tt>%metadata</tt> argument can be used to avoid requiring heap objects 452to have 'isa' pointers or tag bits. [<a href="#appel89">Appel89</a>, <a 453href="#goldberg91">Goldberg91</a>, <a href="#tolmach94">Tolmach94</a>] If 454specified, its value will be tracked along with the location of the pointer in 455the stack frame.</p> 456 457<p>Consider the following fragment of Java code:</p> 458 459<pre class="doc_code"> 460 { 461 Object X; // A null-initialized reference to an object 462 ... 463 } 464</pre> 465 466<p>This block (which may be located in the middle of a function or in a loop 467nest), could be compiled to this LLVM code:</p> 468 469<pre class="doc_code"> 470Entry: 471 ;; In the entry block for the function, allocate the 472 ;; stack space for X, which is an LLVM pointer. 473 %X = alloca %Object* 474 475 ;; Tell LLVM that the stack space is a stack root. 476 ;; Java has type-tags on objects, so we pass null as metadata. 477 %tmp = bitcast %Object** %X to i8** 478 call void @llvm.gcroot(i8** %X, i8* null) 479 ... 480 481 ;; "CodeBlock" is the block corresponding to the start 482 ;; of the scope above. 483CodeBlock: 484 ;; Java null-initializes pointers. 485 store %Object* null, %Object** %X 486 487 ... 488 489 ;; As the pointer goes out of scope, store a null value into 490 ;; it, to indicate that the value is no longer live. 491 store %Object* null, %Object** %X 492 ... 493</pre> 494 495</div> 496 497<!-- ======================================================================= --> 498<h3> 499 <a name="barriers">Reading and writing references in the heap</a> 500</h3> 501 502<div> 503 504<p>Some collectors need to be informed when the mutator (the program that needs 505garbage collection) either reads a pointer from or writes a pointer to a field 506of a heap object. The code fragments inserted at these points are called 507<em>read barriers</em> and <em>write barriers</em>, respectively. The amount of 508code that needs to be executed is usually quite small and not on the critical 509path of any computation, so the overall performance impact of the barrier is 510tolerable.</p> 511 512<p>Barriers often require access to the <em>object pointer</em> rather than the 513<em>derived pointer</em> (which is a pointer to the field within the 514object). Accordingly, these intrinsics take both pointers as separate arguments 515for completeness. In this snippet, <tt>%object</tt> is the object pointer, and 516<tt>%derived</tt> is the derived pointer:</p> 517 518<blockquote><pre> 519 ;; An array type. 520 %class.Array = type { %class.Object, i32, [0 x %class.Object*] } 521 ... 522 523 ;; Load the object pointer from a gcroot. 524 %object = load %class.Array** %object_addr 525 526 ;; Compute the derived pointer. 527 %derived = getelementptr %object, i32 0, i32 2, i32 %n</pre></blockquote> 528 529<p>LLVM does not enforce this relationship between the object and derived 530pointer (although a <a href="#plugin">plugin</a> might). However, it would be 531an unusual collector that violated it.</p> 532 533<p>The use of these intrinsics is naturally optional if the target GC does 534require the corresponding barrier. Such a GC plugin will replace the intrinsic 535calls with the corresponding <tt>load</tt> or <tt>store</tt> instruction if they 536are used.</p> 537 538<!-- ======================================================================= --> 539<h4> 540 <a name="gcwrite">Write barrier: <tt>llvm.gcwrite</tt></a> 541</h4> 542 543<div> 544 545<div class="doc_code"><tt> 546void @llvm.gcwrite(i8* %value, i8* %object, i8** %derived) 547</tt></div> 548 549<p>For write barriers, LLVM provides the <tt>llvm.gcwrite</tt> intrinsic 550function. It has exactly the same semantics as a non-volatile <tt>store</tt> to 551the derived pointer (the third argument). The exact code generated is specified 552by a <a href="#plugin">compiler plugin</a>.</p> 553 554<p>Many important algorithms require write barriers, including generational 555and concurrent collectors. Additionally, write barriers could be used to 556implement reference counting.</p> 557 558</div> 559 560<!-- ======================================================================= --> 561<h4> 562 <a name="gcread">Read barrier: <tt>llvm.gcread</tt></a> 563</h4> 564 565<div> 566 567<div class="doc_code"><tt> 568i8* @llvm.gcread(i8* %object, i8** %derived)<br> 569</tt></div> 570 571<p>For read barriers, LLVM provides the <tt>llvm.gcread</tt> intrinsic function. 572It has exactly the same semantics as a non-volatile <tt>load</tt> from the 573derived pointer (the second argument). The exact code generated is specified by 574a <a href="#plugin">compiler plugin</a>.</p> 575 576<p>Read barriers are needed by fewer algorithms than write barriers, and may 577have a greater performance impact since pointer reads are more frequent than 578writes.</p> 579 580</div> 581 582</div> 583 584</div> 585 586<!-- *********************************************************************** --> 587<h2> 588 <a name="plugin">Implementing a collector plugin</a> 589</h2> 590<!-- *********************************************************************** --> 591 592<div> 593 594<p>User code specifies which GC code generation to use with the <tt>gc</tt> 595function attribute or, equivalently, with the <tt>setGC</tt> method of 596<tt>Function</tt>.</p> 597 598<p>To implement a GC plugin, it is necessary to subclass 599<tt>llvm::GCStrategy</tt>, which can be accomplished in a few lines of 600boilerplate code. LLVM's infrastructure provides access to several important 601algorithms. For an uncontroversial collector, all that remains may be to 602compile LLVM's computed stack map to assembly code (using the binary 603representation expected by the runtime library). This can be accomplished in 604about 100 lines of code.</p> 605 606<p>This is not the appropriate place to implement a garbage collected heap or a 607garbage collector itself. That code should exist in the language's runtime 608library. The compiler plugin is responsible for generating code which 609conforms to the binary interface defined by library, most essentially the 610<a href="#stack-map">stack map</a>.</p> 611 612<p>To subclass <tt>llvm::GCStrategy</tt> and register it with the compiler:</p> 613 614<blockquote><pre>// lib/MyGC/MyGC.cpp - Example LLVM GC plugin 615 616#include "llvm/CodeGen/GCStrategy.h" 617#include "llvm/CodeGen/GCMetadata.h" 618#include "llvm/Support/Compiler.h" 619 620using namespace llvm; 621 622namespace { 623 class LLVM_LIBRARY_VISIBILITY MyGC : public GCStrategy { 624 public: 625 MyGC() {} 626 }; 627 628 GCRegistry::Add<MyGC> 629 X("mygc", "My bespoke garbage collector."); 630}</pre></blockquote> 631 632<p>This boilerplate collector does nothing. More specifically:</p> 633 634<ul> 635 <li><tt>llvm.gcread</tt> calls are replaced with the corresponding 636 <tt>load</tt> instruction.</li> 637 <li><tt>llvm.gcwrite</tt> calls are replaced with the corresponding 638 <tt>store</tt> instruction.</li> 639 <li>No safe points are added to the code.</li> 640 <li>The stack map is not compiled into the executable.</li> 641</ul> 642 643<p>Using the LLVM makefiles (like the <a 644href="http://llvm.org/viewvc/llvm-project/llvm/trunk/projects/sample/">sample 645project</a>), this code can be compiled as a plugin using a simple 646makefile:</p> 647 648<blockquote><pre 649># lib/MyGC/Makefile 650 651LEVEL := ../.. 652LIBRARYNAME = <var>MyGC</var> 653LOADABLE_MODULE = 1 654 655include $(LEVEL)/Makefile.common</pre></blockquote> 656 657<p>Once the plugin is compiled, code using it may be compiled using <tt>llc 658-load=<var>MyGC.so</var></tt> (though <var>MyGC.so</var> may have some other 659platform-specific extension):</p> 660 661<blockquote><pre 662>$ cat sample.ll 663define void @f() gc "mygc" { 664entry: 665 ret void 666} 667$ llvm-as < sample.ll | llc -load=MyGC.so</pre></blockquote> 668 669<p>It is also possible to statically link the collector plugin into tools, such 670as a language-specific compiler front-end.</p> 671 672<!-- ======================================================================= --> 673<h3> 674 <a name="collector-algos">Overview of available features</a> 675</h3> 676 677<div> 678 679<p><tt>GCStrategy</tt> provides a range of features through which a plugin 680may do useful work. Some of these are callbacks, some are algorithms that can 681be enabled, disabled, or customized. This matrix summarizes the supported (and 682planned) features and correlates them with the collection techniques which 683typically require them.</p> 684 685<table> 686 <tr> 687 <th>Algorithm</th> 688 <th>Done</th> 689 <th>shadow stack</th> 690 <th>refcount</th> 691 <th>mark-sweep</th> 692 <th>copying</th> 693 <th>incremental</th> 694 <th>threaded</th> 695 <th>concurrent</th> 696 </tr> 697 <tr> 698 <th class="rowhead"><a href="#stack-map">stack map</a></th> 699 <td>✔</td> 700 <td></td> 701 <td></td> 702 <td>✘</td> 703 <td>✘</td> 704 <td>✘</td> 705 <td>✘</td> 706 <td>✘</td> 707 </tr> 708 <tr> 709 <th class="rowhead"><a href="#init-roots">initialize roots</a></th> 710 <td>✔</td> 711 <td>✘</td> 712 <td>✘</td> 713 <td>✘</td> 714 <td>✘</td> 715 <td>✘</td> 716 <td>✘</td> 717 <td>✘</td> 718 </tr> 719 <tr class="doc_warning"> 720 <th class="rowhead">derived pointers</th> 721 <td>NO</td> 722 <td></td> 723 <td></td> 724 <td></td> 725 <td></td> 726 <td></td> 727 <td>✘*</td> 728 <td>✘*</td> 729 </tr> 730 <tr> 731 <th class="rowhead"><em><a href="#custom">custom lowering</a></em></th> 732 <td>✔</td> 733 <th></th> 734 <th></th> 735 <th></th> 736 <th></th> 737 <th></th> 738 <th></th> 739 <th></th> 740 </tr> 741 <tr> 742 <th class="rowhead indent">gcroot</th> 743 <td>✔</td> 744 <td>✘</td> 745 <td>✘</td> 746 <td></td> 747 <td></td> 748 <td></td> 749 <td></td> 750 <td></td> 751 </tr> 752 <tr> 753 <th class="rowhead indent">gcwrite</th> 754 <td>✔</td> 755 <td></td> 756 <td>✘</td> 757 <td></td> 758 <td></td> 759 <td>✘</td> 760 <td></td> 761 <td>✘</td> 762 </tr> 763 <tr> 764 <th class="rowhead indent">gcread</th> 765 <td>✔</td> 766 <td></td> 767 <td></td> 768 <td></td> 769 <td></td> 770 <td></td> 771 <td></td> 772 <td>✘</td> 773 </tr> 774 <tr> 775 <th class="rowhead"><em><a href="#safe-points">safe points</a></em></th> 776 <td></td> 777 <th></th> 778 <th></th> 779 <th></th> 780 <th></th> 781 <th></th> 782 <th></th> 783 <th></th> 784 </tr> 785 <tr> 786 <th class="rowhead indent">in calls</th> 787 <td>✔</td> 788 <td></td> 789 <td></td> 790 <td>✘</td> 791 <td>✘</td> 792 <td>✘</td> 793 <td>✘</td> 794 <td>✘</td> 795 </tr> 796 <tr> 797 <th class="rowhead indent">before calls</th> 798 <td>✔</td> 799 <td></td> 800 <td></td> 801 <td></td> 802 <td></td> 803 <td></td> 804 <td>✘</td> 805 <td>✘</td> 806 </tr> 807 <tr class="doc_warning"> 808 <th class="rowhead indent">for loops</th> 809 <td>NO</td> 810 <td></td> 811 <td></td> 812 <td></td> 813 <td></td> 814 <td></td> 815 <td>✘</td> 816 <td>✘</td> 817 </tr> 818 <tr> 819 <th class="rowhead indent">before escape</th> 820 <td>✔</td> 821 <td></td> 822 <td></td> 823 <td></td> 824 <td></td> 825 <td></td> 826 <td>✘</td> 827 <td>✘</td> 828 </tr> 829 <tr class="doc_warning"> 830 <th class="rowhead">emit code at safe points</th> 831 <td>NO</td> 832 <td></td> 833 <td></td> 834 <td></td> 835 <td></td> 836 <td></td> 837 <td>✘</td> 838 <td>✘</td> 839 </tr> 840 <tr> 841 <th class="rowhead"><em>output</em></th> 842 <td></td> 843 <th></th> 844 <th></th> 845 <th></th> 846 <th></th> 847 <th></th> 848 <th></th> 849 <th></th> 850 </tr> 851 <tr> 852 <th class="rowhead indent"><a href="#assembly">assembly</a></th> 853 <td>✔</td> 854 <td></td> 855 <td></td> 856 <td>✘</td> 857 <td>✘</td> 858 <td>✘</td> 859 <td>✘</td> 860 <td>✘</td> 861 </tr> 862 <tr class="doc_warning"> 863 <th class="rowhead indent">JIT</th> 864 <td>NO</td> 865 <td></td> 866 <td></td> 867 <td class="optl">✘</td> 868 <td class="optl">✘</td> 869 <td class="optl">✘</td> 870 <td class="optl">✘</td> 871 <td class="optl">✘</td> 872 </tr> 873 <tr class="doc_warning"> 874 <th class="rowhead indent">obj</th> 875 <td>NO</td> 876 <td></td> 877 <td></td> 878 <td class="optl">✘</td> 879 <td class="optl">✘</td> 880 <td class="optl">✘</td> 881 <td class="optl">✘</td> 882 <td class="optl">✘</td> 883 </tr> 884 <tr class="doc_warning"> 885 <th class="rowhead">live analysis</th> 886 <td>NO</td> 887 <td></td> 888 <td></td> 889 <td class="optl">✘</td> 890 <td class="optl">✘</td> 891 <td class="optl">✘</td> 892 <td class="optl">✘</td> 893 <td class="optl">✘</td> 894 </tr> 895 <tr class="doc_warning"> 896 <th class="rowhead">register map</th> 897 <td>NO</td> 898 <td></td> 899 <td></td> 900 <td class="optl">✘</td> 901 <td class="optl">✘</td> 902 <td class="optl">✘</td> 903 <td class="optl">✘</td> 904 <td class="optl">✘</td> 905 </tr> 906 <tr> 907 <td colspan="10"> 908 <div><span class="doc_warning">*</span> Derived pointers only pose a 909 hazard to copying collectors.</div> 910 <div><span class="optl">✘</span> in gray denotes a feature which 911 could be utilized if available.</div> 912 </td> 913 </tr> 914</table> 915 916<p>To be clear, the collection techniques above are defined as:</p> 917 918<dl> 919 <dt>Shadow Stack</dt> 920 <dd>The mutator carefully maintains a linked list of stack roots.</dd> 921 <dt>Reference Counting</dt> 922 <dd>The mutator maintains a reference count for each object and frees an 923 object when its count falls to zero.</dd> 924 <dt>Mark-Sweep</dt> 925 <dd>When the heap is exhausted, the collector marks reachable objects starting 926 from the roots, then deallocates unreachable objects in a sweep 927 phase.</dd> 928 <dt>Copying</dt> 929 <dd>As reachability analysis proceeds, the collector copies objects from one 930 heap area to another, compacting them in the process. Copying collectors 931 enable highly efficient "bump pointer" allocation and can improve locality 932 of reference.</dd> 933 <dt>Incremental</dt> 934 <dd>(Including generational collectors.) Incremental collectors generally have 935 all the properties of a copying collector (regardless of whether the 936 mature heap is compacting), but bring the added complexity of requiring 937 write barriers.</dd> 938 <dt>Threaded</dt> 939 <dd>Denotes a multithreaded mutator; the collector must still stop the mutator 940 ("stop the world") before beginning reachability analysis. Stopping a 941 multithreaded mutator is a complicated problem. It generally requires 942 highly platform specific code in the runtime, and the production of 943 carefully designed machine code at safe points.</dd> 944 <dt>Concurrent</dt> 945 <dd>In this technique, the mutator and the collector run concurrently, with 946 the goal of eliminating pause times. In a <em>cooperative</em> collector, 947 the mutator further aids with collection should a pause occur, allowing 948 collection to take advantage of multiprocessor hosts. The "stop the world" 949 problem of threaded collectors is generally still present to a limited 950 extent. Sophisticated marking algorithms are necessary. Read barriers may 951 be necessary.</dd> 952</dl> 953 954<p>As the matrix indicates, LLVM's garbage collection infrastructure is already 955suitable for a wide variety of collectors, but does not currently extend to 956multithreaded programs. This will be added in the future as there is 957interest.</p> 958 959</div> 960 961<!-- ======================================================================= --> 962<h3> 963 <a name="stack-map">Computing stack maps</a> 964</h3> 965 966<div> 967 968<p>LLVM automatically computes a stack map. One of the most important features 969of a <tt>GCStrategy</tt> is to compile this information into the executable in 970the binary representation expected by the runtime library.</p> 971 972<p>The stack map consists of the location and identity of each GC root in the 973each function in the module. For each root:</p> 974 975<ul> 976 <li><tt>RootNum</tt>: The index of the root.</li> 977 <li><tt>StackOffset</tt>: The offset of the object relative to the frame 978 pointer.</li> 979 <li><tt>RootMetadata</tt>: The value passed as the <tt>%metadata</tt> 980 parameter to the <a href="#gcroot"><tt>@llvm.gcroot</tt></a> intrinsic.</li> 981</ul> 982 983<p>Also, for the function as a whole:</p> 984 985<ul> 986 <li><tt>getFrameSize()</tt>: The overall size of the function's initial 987 stack frame, not accounting for any dynamic allocation.</li> 988 <li><tt>roots_size()</tt>: The count of roots in the function.</li> 989</ul> 990 991<p>To access the stack map, use <tt>GCFunctionMetadata::roots_begin()</tt> and 992-<tt>end()</tt> from the <tt><a 993href="#assembly">GCMetadataPrinter</a></tt>:</p> 994 995<blockquote><pre 996>for (iterator I = begin(), E = end(); I != E; ++I) { 997 GCFunctionInfo *FI = *I; 998 unsigned FrameSize = FI->getFrameSize(); 999 size_t RootCount = FI->roots_size(); 1000 1001 for (GCFunctionInfo::roots_iterator RI = FI->roots_begin(), 1002 RE = FI->roots_end(); 1003 RI != RE; ++RI) { 1004 int RootNum = RI->Num; 1005 int RootStackOffset = RI->StackOffset; 1006 Constant *RootMetadata = RI->Metadata; 1007 } 1008}</pre></blockquote> 1009 1010<p>If the <tt>llvm.gcroot</tt> intrinsic is eliminated before code generation by 1011a custom lowering pass, LLVM will compute an empty stack map. This may be useful 1012for collector plugins which implement reference counting or a shadow stack.</p> 1013 1014</div> 1015 1016 1017<!-- ======================================================================= --> 1018<h3> 1019 <a name="init-roots">Initializing roots to null: <tt>InitRoots</tt></a> 1020</h3> 1021 1022<div> 1023 1024<blockquote><pre 1025>MyGC::MyGC() { 1026 InitRoots = true; 1027}</pre></blockquote> 1028 1029<p>When set, LLVM will automatically initialize each root to <tt>null</tt> upon 1030entry to the function. This prevents the GC's sweep phase from visiting 1031uninitialized pointers, which will almost certainly cause it to crash. This 1032initialization occurs before custom lowering, so the two may be used 1033together.</p> 1034 1035<p>Since LLVM does not yet compute liveness information, there is no means of 1036distinguishing an uninitialized stack root from an initialized one. Therefore, 1037this feature should be used by all GC plugins. It is enabled by default.</p> 1038 1039</div> 1040 1041 1042<!-- ======================================================================= --> 1043<h3> 1044 <a name="custom">Custom lowering of intrinsics: <tt>CustomRoots</tt>, 1045 <tt>CustomReadBarriers</tt>, and <tt>CustomWriteBarriers</tt></a> 1046</h3> 1047 1048<div> 1049 1050<p>For GCs which use barriers or unusual treatment of stack roots, these 1051flags allow the collector to perform arbitrary transformations of the LLVM 1052IR:</p> 1053 1054<blockquote><pre 1055>class MyGC : public GCStrategy { 1056public: 1057 MyGC() { 1058 CustomRoots = true; 1059 CustomReadBarriers = true; 1060 CustomWriteBarriers = true; 1061 } 1062 1063 virtual bool initializeCustomLowering(Module &M); 1064 virtual bool performCustomLowering(Function &F); 1065};</pre></blockquote> 1066 1067<p>If any of these flags are set, then LLVM suppresses its default lowering for 1068the corresponding intrinsics and instead calls 1069<tt>performCustomLowering</tt>.</p> 1070 1071<p>LLVM's default action for each intrinsic is as follows:</p> 1072 1073<ul> 1074 <li><tt>llvm.gcroot</tt>: Leave it alone. The code generator must see it 1075 or the stack map will not be computed.</li> 1076 <li><tt>llvm.gcread</tt>: Substitute a <tt>load</tt> instruction.</li> 1077 <li><tt>llvm.gcwrite</tt>: Substitute a <tt>store</tt> instruction.</li> 1078</ul> 1079 1080<p>If <tt>CustomReadBarriers</tt> or <tt>CustomWriteBarriers</tt> are specified, 1081then <tt>performCustomLowering</tt> <strong>must</strong> eliminate the 1082corresponding barriers.</p> 1083 1084<p><tt>performCustomLowering</tt> must comply with the same restrictions as <a 1085href="WritingAnLLVMPass.html#runOnFunction"><tt 1086>FunctionPass::runOnFunction</tt></a>. 1087Likewise, <tt>initializeCustomLowering</tt> has the same semantics as <a 1088href="WritingAnLLVMPass.html#doInitialization_mod"><tt 1089>Pass::doInitialization(Module&)</tt></a>.</p> 1090 1091<p>The following can be used as a template:</p> 1092 1093<blockquote><pre 1094>#include "llvm/Module.h" 1095#include "llvm/IntrinsicInst.h" 1096 1097bool MyGC::initializeCustomLowering(Module &M) { 1098 return false; 1099} 1100 1101bool MyGC::performCustomLowering(Function &F) { 1102 bool MadeChange = false; 1103 1104 for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) 1105 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ) 1106 if (IntrinsicInst *CI = dyn_cast<IntrinsicInst>(II++)) 1107 if (Function *F = CI->getCalledFunction()) 1108 switch (F->getIntrinsicID()) { 1109 case Intrinsic::gcwrite: 1110 // Handle llvm.gcwrite. 1111 CI->eraseFromParent(); 1112 MadeChange = true; 1113 break; 1114 case Intrinsic::gcread: 1115 // Handle llvm.gcread. 1116 CI->eraseFromParent(); 1117 MadeChange = true; 1118 break; 1119 case Intrinsic::gcroot: 1120 // Handle llvm.gcroot. 1121 CI->eraseFromParent(); 1122 MadeChange = true; 1123 break; 1124 } 1125 1126 return MadeChange; 1127}</pre></blockquote> 1128 1129</div> 1130 1131 1132<!-- ======================================================================= --> 1133<h3> 1134 <a name="safe-points">Generating safe points: <tt>NeededSafePoints</tt></a> 1135</h3> 1136 1137<div> 1138 1139<p>LLVM can compute four kinds of safe points:</p> 1140 1141<blockquote><pre 1142>namespace GC { 1143 /// PointKind - The type of a collector-safe point. 1144 /// 1145 enum PointKind { 1146 Loop, //< Instr is a loop (backwards branch). 1147 Return, //< Instr is a return instruction. 1148 PreCall, //< Instr is a call instruction. 1149 PostCall //< Instr is the return address of a call. 1150 }; 1151}</pre></blockquote> 1152 1153<p>A collector can request any combination of the four by setting the 1154<tt>NeededSafePoints</tt> mask:</p> 1155 1156<blockquote><pre 1157>MyGC::MyGC() { 1158 NeededSafePoints = 1 << GC::Loop 1159 | 1 << GC::Return 1160 | 1 << GC::PreCall 1161 | 1 << GC::PostCall; 1162}</pre></blockquote> 1163 1164<p>It can then use the following routines to access safe points.</p> 1165 1166<blockquote><pre 1167>for (iterator I = begin(), E = end(); I != E; ++I) { 1168 GCFunctionInfo *MD = *I; 1169 size_t PointCount = MD->size(); 1170 1171 for (GCFunctionInfo::iterator PI = MD->begin(), 1172 PE = MD->end(); PI != PE; ++PI) { 1173 GC::PointKind PointKind = PI->Kind; 1174 unsigned PointNum = PI->Num; 1175 } 1176} 1177</pre></blockquote> 1178 1179<p>Almost every collector requires <tt>PostCall</tt> safe points, since these 1180correspond to the moments when the function is suspended during a call to a 1181subroutine.</p> 1182 1183<p>Threaded programs generally require <tt>Loop</tt> safe points to guarantee 1184that the application will reach a safe point within a bounded amount of time, 1185even if it is executing a long-running loop which contains no function 1186calls.</p> 1187 1188<p>Threaded collectors may also require <tt>Return</tt> and <tt>PreCall</tt> 1189safe points to implement "stop the world" techniques using self-modifying code, 1190where it is important that the program not exit the function without reaching a 1191safe point (because only the topmost function has been patched).</p> 1192 1193</div> 1194 1195 1196<!-- ======================================================================= --> 1197<h3> 1198 <a name="assembly">Emitting assembly code: <tt>GCMetadataPrinter</tt></a> 1199</h3> 1200 1201<div> 1202 1203<p>LLVM allows a plugin to print arbitrary assembly code before and after the 1204rest of a module's assembly code. At the end of the module, the GC can compile 1205the LLVM stack map into assembly code. (At the beginning, this information is not 1206yet computed.)</p> 1207 1208<p>Since AsmWriter and CodeGen are separate components of LLVM, a separate 1209abstract base class and registry is provided for printing assembly code, the 1210<tt>GCMetadaPrinter</tt> and <tt>GCMetadataPrinterRegistry</tt>. The AsmWriter 1211will look for such a subclass if the <tt>GCStrategy</tt> sets 1212<tt>UsesMetadata</tt>:</p> 1213 1214<blockquote><pre 1215>MyGC::MyGC() { 1216 UsesMetadata = true; 1217}</pre></blockquote> 1218 1219<p>This separation allows JIT-only clients to be smaller.</p> 1220 1221<p>Note that LLVM does not currently have analogous APIs to support code 1222generation in the JIT, nor using the object writers.</p> 1223 1224<blockquote><pre 1225>// lib/MyGC/MyGCPrinter.cpp - Example LLVM GC printer 1226 1227#include "llvm/CodeGen/GCMetadataPrinter.h" 1228#include "llvm/Support/Compiler.h" 1229 1230using namespace llvm; 1231 1232namespace { 1233 class LLVM_LIBRARY_VISIBILITY MyGCPrinter : public GCMetadataPrinter { 1234 public: 1235 virtual void beginAssembly(std::ostream &OS, AsmPrinter &AP, 1236 const TargetAsmInfo &TAI); 1237 1238 virtual void finishAssembly(std::ostream &OS, AsmPrinter &AP, 1239 const TargetAsmInfo &TAI); 1240 }; 1241 1242 GCMetadataPrinterRegistry::Add<MyGCPrinter> 1243 X("mygc", "My bespoke garbage collector."); 1244}</pre></blockquote> 1245 1246<p>The collector should use <tt>AsmPrinter</tt> and <tt>TargetAsmInfo</tt> to 1247print portable assembly code to the <tt>std::ostream</tt>. The collector itself 1248contains the stack map for the entire module, and may access the 1249<tt>GCFunctionInfo</tt> using its own <tt>begin()</tt> and <tt>end()</tt> 1250methods. Here's a realistic example:</p> 1251 1252<blockquote><pre 1253>#include "llvm/CodeGen/AsmPrinter.h" 1254#include "llvm/Function.h" 1255#include "llvm/Target/TargetMachine.h" 1256#include "llvm/Target/TargetData.h" 1257#include "llvm/Target/TargetAsmInfo.h" 1258 1259void MyGCPrinter::beginAssembly(std::ostream &OS, AsmPrinter &AP, 1260 const TargetAsmInfo &TAI) { 1261 // Nothing to do. 1262} 1263 1264void MyGCPrinter::finishAssembly(std::ostream &OS, AsmPrinter &AP, 1265 const TargetAsmInfo &TAI) { 1266 // Set up for emitting addresses. 1267 const char *AddressDirective; 1268 int AddressAlignLog; 1269 if (AP.TM.getTargetData()->getPointerSize() == sizeof(int32_t)) { 1270 AddressDirective = TAI.getData32bitsDirective(); 1271 AddressAlignLog = 2; 1272 } else { 1273 AddressDirective = TAI.getData64bitsDirective(); 1274 AddressAlignLog = 3; 1275 } 1276 1277 // Put this in the data section. 1278 AP.SwitchToDataSection(TAI.getDataSection()); 1279 1280 // For each function... 1281 for (iterator FI = begin(), FE = end(); FI != FE; ++FI) { 1282 GCFunctionInfo &MD = **FI; 1283 1284 // Emit this data structure: 1285 // 1286 // struct { 1287 // int32_t PointCount; 1288 // struct { 1289 // void *SafePointAddress; 1290 // int32_t LiveCount; 1291 // int32_t LiveOffsets[LiveCount]; 1292 // } Points[PointCount]; 1293 // } __gcmap_<FUNCTIONNAME>; 1294 1295 // Align to address width. 1296 AP.EmitAlignment(AddressAlignLog); 1297 1298 // Emit the symbol by which the stack map entry can be found. 1299 std::string Symbol; 1300 Symbol += TAI.getGlobalPrefix(); 1301 Symbol += "__gcmap_"; 1302 Symbol += MD.getFunction().getName(); 1303 if (const char *GlobalDirective = TAI.getGlobalDirective()) 1304 OS << GlobalDirective << Symbol << "\n"; 1305 OS << TAI.getGlobalPrefix() << Symbol << ":\n"; 1306 1307 // Emit PointCount. 1308 AP.EmitInt32(MD.size()); 1309 AP.EOL("safe point count"); 1310 1311 // And each safe point... 1312 for (GCFunctionInfo::iterator PI = MD.begin(), 1313 PE = MD.end(); PI != PE; ++PI) { 1314 // Align to address width. 1315 AP.EmitAlignment(AddressAlignLog); 1316 1317 // Emit the address of the safe point. 1318 OS << AddressDirective 1319 << TAI.getPrivateGlobalPrefix() << "label" << PI->Num; 1320 AP.EOL("safe point address"); 1321 1322 // Emit the stack frame size. 1323 AP.EmitInt32(MD.getFrameSize()); 1324 AP.EOL("stack frame size"); 1325 1326 // Emit the number of live roots in the function. 1327 AP.EmitInt32(MD.live_size(PI)); 1328 AP.EOL("live root count"); 1329 1330 // And for each live root... 1331 for (GCFunctionInfo::live_iterator LI = MD.live_begin(PI), 1332 LE = MD.live_end(PI); 1333 LI != LE; ++LI) { 1334 // Print its offset within the stack frame. 1335 AP.EmitInt32(LI->StackOffset); 1336 AP.EOL("stack offset"); 1337 } 1338 } 1339 } 1340} 1341</pre></blockquote> 1342 1343</div> 1344 1345</div> 1346 1347<!-- *********************************************************************** --> 1348<h2> 1349 <a name="references">References</a> 1350</h2> 1351<!-- *********************************************************************** --> 1352 1353<div> 1354 1355<p><a name="appel89">[Appel89]</a> Runtime Tags Aren't Necessary. Andrew 1356W. Appel. Lisp and Symbolic Computation 19(7):703-705, July 1989.</p> 1357 1358<p><a name="goldberg91">[Goldberg91]</a> Tag-free garbage collection for 1359strongly typed programming languages. Benjamin Goldberg. ACM SIGPLAN 1360PLDI'91.</p> 1361 1362<p><a name="tolmach94">[Tolmach94]</a> Tag-free garbage collection using 1363explicit type parameters. Andrew Tolmach. Proceedings of the 1994 ACM 1364conference on LISP and functional programming.</p> 1365 1366<p><a name="henderson02">[Henderson2002]</a> <a 1367href="http://citeseer.ist.psu.edu/henderson02accurate.html"> 1368Accurate Garbage Collection in an Uncooperative Environment</a>. 1369Fergus Henderson. International Symposium on Memory Management 2002.</p> 1370 1371</div> 1372 1373 1374<!-- *********************************************************************** --> 1375 1376<hr> 1377<address> 1378 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img 1379 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a> 1380 <a href="http://validator.w3.org/check/referer"><img 1381 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a> 1382 1383 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br> 1384 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br> 1385 Last modified: $Date$ 1386</address> 1387 1388</body> 1389</html> 1390