docs/ReleaseNotes.html

   1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
   2                       "http://www.w3.org/TR/html4/strict.dtd">
   3 <html>
   4 <head>
   5   <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
   6   <link rel="stylesheet" href="llvm.css" type="text/css">
   7   <title>LLVM 2.6 Release Notes</title>
   8 </head>
   9 <body>
  10
  11 <div class="doc_title">LLVM 2.6 Release Notes</div>
  12
  13 <ol>
  14   <li><a href="#intro">Introduction</a></li>
  15   <li><a href="#subproj">Sub-project Status Update</a></li>
  16   <li><a href="#externalproj">External Projects Using LLVM 2.6</a></li>
  17   <li><a href="#whatsnew">What's New in LLVM 2.6?</a></li>
  18   <li><a href="GettingStarted.html">Installation Instructions</a></li>
  19   <li><a href="#portability">Portability and Supported Platforms</a></li>
  20   <li><a href="#knownproblems">Known Problems</a></li>
  21   <li><a href="#additionalinfo">Additional Information</a></li>
  22 </ol>
  23
  24 <div class="doc_author">
  25   <p>Written by the <a href="http://llvm.org">LLVM Team</a></p>
  26 </div>
  27
  28 <!-- *********************************************************************** -->
  29 <div class="doc_section">
  30   <a name="intro">Introduction</a>
  31 </div>
  32 <!-- *********************************************************************** -->
  33
  34 <div class="doc_text">
  35
  36 <p>This document contains the release notes for the LLVM Compiler
  37 Infrastructure, release 2.6.  Here we describe the status of LLVM, including
  38 major improvements from the previous release and significant known problems.
  39 All LLVM releases may be downloaded from the <a
  40 href="http://llvm.org/releases/">LLVM releases web site</a>.</p>
  41
  42 <p>For more information about LLVM, including information about the latest
  43 release, please check out the <a href="http://llvm.org/">main LLVM
  44 web site</a>.  If you have questions or comments, the <a
  45 href="http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM Developer's Mailing
  46 List</a> is a good place to send them.</p>
  47
  48 <p>Note that if you are reading this file from a Subversion checkout or the
  49 main LLVM web page, this document applies to the <i>next</i> release, not the
  50 current one.  To see the release notes for a specific release, please see the
  51 <a href="http://llvm.org/releases/">releases page</a>.</p>
  52
  53 </div>
  54
  55
  56 <!--
  57 Almost dead code.
  58   include/llvm/Analysis/LiveValues.h => Dan
  59   lib/Transforms/IPO/MergeFunctions.cpp => consider for 2.8.
  60   llvm/Analysis/PointerTracking.h => Edwin wants this, consider for 2.8.
  61 -->
  62
  63
  64 <!-- Unfinished features in 2.6:
  65   gcc plugin.
  66   strong phi elim
  67   variable debug info for optimized code
  68   postalloc scheduler: anti dependence breaking, hazard recognizer?
  69   metadata
  70   loop dependence analysis
  71   ELF Writer?  How stable?
  72   <li>PostRA scheduler improvements, ARM adoption (David Goodwin).</li>
  73   2.7 supports the GDB 7.0 jit interfaces for debug info.
  74   2.7 eliminates ADT/iterator.h
  75  -->
  76
  77  <!-- for announcement email:
  78  Logo web page.
  79  llvm devmtg
  80  compiler_rt
  81  KLEE web page at klee.llvm.org
  82  Many new papers added to /pubs/
  83    Mention gcc plugin.
  84
  85    -->
  86
  87 <!-- *********************************************************************** -->
  88 <div class="doc_section">
  89   <a name="subproj">Sub-project Status Update</a>
  90 </div>
  91 <!-- *********************************************************************** -->
  92
  93 <div class="doc_text">
  94 <p>
  95 The LLVM 2.6 distribution currently consists of code from the core LLVM
  96 repository (which roughly includes the LLVM optimizers, code generators
  97 and supporting tools), the Clang repository and the llvm-gcc repository.  In
  98 addition to this code, the LLVM Project includes other sub-projects that are in
  99 development.  Here we include updates on these subprojects.
 100 </p>
 101
 102 </div>
 103
 104
 105 <!--=========================================================================-->
 106 <div class="doc_subsection">
 107 <a name="clang">Clang: C/C++/Objective-C Frontend Toolkit</a>
 108 </div>
 109
 110 <div class="doc_text">
 111
 112 <p>The <a href="http://clang.llvm.org/">Clang project</a> is an effort to build
 113 a set of new 'LLVM native' front-end technologies for the C family of languages.
 114 LLVM 2.6 is the first release to officially include Clang, and it provides a
 115 production quality C and Objective-C compiler.  If you are interested in <a
 116 href="http://clang.llvm.org/performance.html">fast compiles</a> and
 117 <a href="http://clang.llvm.org/diagnostics.html">good diagnostics</a>, we
 118 encourage you to try it out.  Clang currently compiles typical Objective-C code
 119 3x faster than GCC and compiles C code about 30% faster than GCC at -O0 -g
 120 (which is when the most pressure is on the frontend).</p>
 121
 122 <p>In addition to supporting these languages, C++ support is also <a
 123 href="http://clang.llvm.org/cxx_status.html">well under way</a>, and mainline
 124 Clang is able to parse the libstdc++ 4.2 headers and even codegen simple apps.
 125 If you are interested in Clang C++ support or any other Clang feature, we
 126 strongly encourage you to get involved on the <a
 127 href="http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev">Clang front-end mailing
 128 list</a>.</p>
 129
 130 <p>In the LLVM 2.6 time-frame, the Clang team has made many improvements:</p>
 131
 132 <ul>
 133 <li>C and Objective-C support are now considered production quality.</li>
 134 <li>AuroraUX, FreeBSD and OpenBSD are now supported.</li>
 135 <li>Most of Objective-C 2.0 is now supported with the GNU runtime.</li>
 136 <li>Many many bugs are fixed and lots of features have been added.</li>
 137 </ul>
 138 </div>
 139
 140 <!--=========================================================================-->
 141 <div class="doc_subsection">
 142 <a name="clangsa">Clang Static Analyzer</a>
 143 </div>
 144
 145 <div class="doc_text">
 146
 147 <p>Previously announced in the 2.4 and 2.5 LLVM releases, the Clang project also
 148 includes an early stage static source code analysis tool for <a
 149 href="http://clang.llvm.org/StaticAnalysis.html">automatically finding bugs</a>
 150 in C and Objective-C programs. The tool performs checks to find
 151 bugs that occur on a specific path within a program.</p>
 152
 153 <p>In the LLVM 2.6 time-frame, the analyzer core has undergone several important
 154 improvements and cleanups and now includes a new <em>Checker</em> interface that
 155 is intended to eventually serve as a basis for domain-specific checks. Further,
 156 in addition to generating HTML files for reporting analysis results, the
 157 analyzer can now also emit bug reports in a structured XML format that is
 158 intended to be easily readable by other programs.</p>
 159
 160 <p>The set of checks performed by the static analyzer continues to expand, and
 161 future plans for the tool include full source-level inter-procedural analysis
 162 and deeper checks such as buffer overrun detection. There are many opportunities
 163 to extend and enhance the static analyzer, and anyone interested in working on
 164 this project is encouraged to get involved!</p>
 165
 166 </div>
 167
 168 <!--=========================================================================-->
 169 <div class="doc_subsection">
 170 <a name="vmkit">VMKit: JVM/CLI Virtual Machine Implementation</a>
 171 </div>
 172
 173 <div class="doc_text">
 174 <p>
 175 The <a href="http://vmkit.llvm.org/">VMKit project</a> is an implementation of
 176 a JVM and a CLI Virtual Machine (Microsoft .NET is an
 177 implementation of the CLI) using LLVM for static and just-in-time
 178 compilation.</p>
 179
 180 <p>
 181 VMKit version 0.26 builds with LLVM 2.6 and you can find it on its
 182 <a href="http://vmkit.llvm.org/releases/">web page</a>. The release includes
 183 bug fixes, cleanup and new features. The major changes are:</p>
 184
 185 <ul>
 186
 187 <li>A new llcj tool to generate shared libraries or executables of Java
 188     files.</li>
 189 <li>Cooperative garbage collection. </li>
 190 <li>Fast subtype checking (paper from Click et al [JGI'02]). </li>
 191 <li>Implementation of a two-word header for Java objects instead of the original
 192     three-word header. </li>
 193 <li>Better Java specification-compliance: division by zero checks, stack
 194     overflow checks, finalization and references support. </li>
 195
 196 </ul>
 197 </div>
 198
 199
 200 <!--=========================================================================-->
 201 <div class="doc_subsection">
 202 <a name="compiler-rt">compiler-rt: Compiler Runtime Library</a>
 203 </div>
 204
 205 <div class="doc_text">
 206 <p>
 207 The new LLVM <a href="http://compiler-rt.llvm.org/">compiler-rt project</a>
 208 is a simple library that provides an implementation of the low-level
 209 target-specific hooks required by code generation and other runtime components.
 210 For example, when compiling for a 32-bit target, converting a double to a 64-bit
 211 unsigned integer is compiled into a runtime call to the "__fixunsdfdi"
 212 function. The compiler-rt library provides highly optimized implementations of
 213 this and other low-level routines (some are 3x faster than the equivalent
 214 libgcc routines).</p>
 215
 216 <p>
 217 All of the code in the compiler-rt project is available under the standard LLVM
 218 License, a "BSD-style" license.</p>
 219
 220 </div>
 221
 222 <!--=========================================================================-->
 223 <div class="doc_subsection">
 224 <a name="klee">KLEE: Symbolic Execution and Automatic Test Case Generator</a>
 225 </div>
 226
 227 <div class="doc_text">
 228 <p>
 229 The new LLVM <a href="http://klee.llvm.org/">KLEE project</a> is a symbolic
 230 execution framework for programs in LLVM bitcode form.  KLEE tries to
 231 symbolically evaluate "all" paths through the application and records state
 232 transitions that lead to fault states.  This allows it to construct testcases
 233 that lead to faults and can even be used to verify algorithms.  For more
 234 details, please see the <a
 235 href="http://llvm.org/pubs/2008-12-OSDI-KLEE.html">OSDI 2008 paper</a> about
 236 KLEE.</p>
 237
 238 </div>
 239
 240 <!--=========================================================================-->
 241 <div class="doc_subsection">
 242 <a name="dragonegg">DragonEgg: GCC-4.5 as an LLVM frontend</a>
 243 </div>
 244
 245 <div class="doc_text">
 246 <p>
 247 The goal of <a href="http://dragonegg.llvm.org/">DragonEgg</a> is to make
 248 gcc-4.5 act like llvm-gcc without requiring any gcc modifications whatsoever.
 249 <a href="http://dragonegg.llvm.org/">DragonEgg</a> is a shared library (dragonegg.so)
 250 that is loaded by gcc at runtime.  It uses the new gcc plugin architecture to
 251 disable the GCC optimizers and code generators, and schedule the LLVM optimizers
 252 and code generators (or direct output of LLVM IR) instead.  Currently only Linux
 253 and Darwin are supported, and only on x86-32 and x86-64.  It should be easy to
 254 add additional unix-like architectures and other processor families.  In theory
 255 it should be possible to use <a href="http://dragonegg.llvm.org/">DragonEgg</a>
 256 with any language supported by gcc, however only C and Fortran work well for the
 257 moment.  Ada and C++ work to some extent, while Java, Obj-C and Obj-C++ are so
 258 far entirely untested.  Since gcc-4.5 has not yet been released, neither has
 259 <a href="http://dragonegg.llvm.org/">DragonEgg</a>.  To build
 260 <a href="http://dragonegg.llvm.org/">DragonEgg</a> you will need to check out the
 261 development versions of <a href="http://gcc.gnu.org/svn.html/"> gcc</a>,
 262 <a href="http://llvm.org/docs/GettingStarted.html#checkout">llvm</a> and
 263 <a href="http://dragonegg.llvm.org/">DragonEgg</a> from their respective
 264 subversion repositories, and follow the instructions in the
 265 <a href="http://dragonegg.llvm.org/">DragonEgg</a> README.
 266 </p>
 267
 268 </div>
 269
 270
 271 <!--=========================================================================-->
 272 <div class="doc_subsection">
 273 <a name="mc">llvm-mc: Machine Code Toolkit</a>
 274 </div>
 275
 276 <div class="doc_text">
 277 <p>
 278 The LLVM Machine Code (MC) Toolkit project is a (very early) effort to build
 279 better tools for dealing with machine code, object file formats, etc.  The idea
 280 is to be able to generate most of the target specific details of assemblers and
 281 disassemblers from existing LLVM target .td files (with suitable enhancements),
 282 and to build infrastructure for reading and writing common object file formats.
 283 One of the first deliverables is to build a full assembler and integrate it into
 284 the compiler, which is predicted to substantially reduce compile time in some
 285 scenarios.
 286 </p>
 287
 288 <p>In the LLVM 2.6 timeframe, the MC framework has grown to the point where it
 289 can reliably parse and pretty print (with some encoding information) a
 290 darwin/x86 .s file successfully, and has the very early phases of a Mach-O
 291 assembler in progress.  Beyond the MC framework itself, major refactoring of the
 292 LLVM code generator has started.  The idea is to make the code generator reason
 293 about the code it is producing in a much more semantic way, rather than a
 294 textual way.  For example, the code generator now uses MCSection objects to
 295 represent section assignments, instead of text strings that print to .section
 296 directives.</p>
 297
 298 <p>MC is an early and ongoing project that will hopefully continue to lead to
 299 many improvements in the code generator and build infrastructure useful for many
 300 other situations.
 301 </p>
 302
 303 </div>
 304
 305
 306 <!-- *********************************************************************** -->
 307 <div class="doc_section">
 308   <a name="externalproj">External Open Source Projects Using LLVM 2.6</a>
 309 </div>
 310 <!-- *********************************************************************** -->
 311
 312 <div class="doc_text">
 313
 314 <p>An exciting aspect of LLVM is that it is used as an enabling technology for
 315    a lot of other language and tools projects.  This section lists some of the
 316    projects that have already been updated to work with LLVM 2.6.</p>
 317 </div>
 318
 319
 320 <!--=========================================================================-->
 321 <div class="doc_subsection">
 322 <a name="Rubinius">Rubinius</a>
 323 </div>
 324
 325 <div class="doc_text">
 326 <p><a href="http://github.com/evanphx/rubinius">Rubinius</a> is an environment
 327 for running Ruby code which strives to write as much of the core class
 328 implementation in Ruby as possible. Combined with a bytecode interpreting VM, it
 329 uses LLVM to optimize and compile ruby code down to machine code. Techniques
 330 such as type feedback, method inlining, and uncommon traps are all used to
 331 remove dynamism from ruby execution and increase performance.</p>
 332
 333 <p>Since LLVM 2.5, Rubinius has made several major leaps forward, implementing
 334 a counter based JIT, type feedback and speculative method inlining.
 335 </p>
 336
 337 </div>
 338
 339 <!--=========================================================================-->
 340 <div class="doc_subsection">
 341 <a name="macruby">MacRuby</a>
 342 </div>
 343
 344 <div class="doc_text">
 345
 346 <p>
 347 <a href="http://macruby.org">MacRuby</a> is an implementation of Ruby on top of
 348 core Mac OS X technologies, such as the Objective-C common runtime and garbage
 349 collector and the CoreFoundation framework. It is principally developed by
 350 Apple and aims at enabling the creation of full-fledged Mac OS X applications.
 351 </p>
 352
 353 <p>
 354 MacRuby uses LLVM for optimization passes, JIT and AOT compilation of Ruby
 355 expressions. It also uses zero-cost DWARF exceptions to implement Ruby exception
 356 handling.</p>
 357
 358 </div>
 359
 360
 361 <!--=========================================================================-->
 362 <div class="doc_subsection">
 363 <a name="pure">Pure</a>
 364 </div>
 365
 366 <div class="doc_text">
 367 <p>
 368 <a href="http://pure-lang.googlecode.com/">Pure</a>
 369 is an algebraic/functional programming language based on term rewriting.
 370 Programs are collections of equations which are used to evaluate expressions in
 371 a symbolic fashion. Pure offers dynamic typing, eager and lazy evaluation,
 372 lexical closures, a hygienic macro system (also based on term rewriting),
 373 built-in list and matrix support (including list and matrix comprehensions) and
 374 an easy-to-use C interface. The interpreter uses LLVM as a backend to
 375  JIT-compile Pure programs to fast native code.</p>
 376
 377 <p>Pure versions 0.31 and later have been tested and are known to work with
 378 LLVM 2.6 (and continue to work with older LLVM releases >= 2.3 as well).
 379 </p>
 380 </div>
 381
 382
 383 <!--=========================================================================-->
 384 <div class="doc_subsection">
 385 <a name="ldc">LLVM D Compiler</a>
 386 </div>
 387
 388 <div class="doc_text">
 389 <p>
 390 <a href="http://www.dsource.org/projects/ldc">LDC</a> is an implementation of
 391 the D Programming Language using the LLVM optimizer and code generator.
 392 The LDC project works great with the LLVM 2.6 release.  General improvements in
 393 this
 394 cycle have included new inline asm constraint handling, better debug info
 395 support, general bug fixes and better x86-64 support.  This has allowed
 396 some major improvements in LDC, getting it much closer to being as
 397 fully featured as the original DMD compiler from DigitalMars.
 398 </p>
 399 </div>
 400
 401 <!--=========================================================================-->
 402 <div class="doc_subsection">
 403 <a name="RoadsendPHP">Roadsend PHP</a>
 404 </div>
 405
 406 <div class="doc_text">
 407 <p>
 408 <a href="http://code.roadsend.com/rphp">Roadsend PHP</a> (rphp) is an open
 409 source implementation of the PHP programming
 410 language that uses LLVM for its optimizer, JIT and static compiler. This is a
 411 reimplementation of an earlier project that is now based on LLVM.</p>
 412 </div>
 413
 414 <!--=========================================================================-->
 415 <div class="doc_subsection">
 416 <a name="UnladenSwallow">Unladen Swallow</a>
 417 </div>
 418
 419 <div class="doc_text">
 420 <p>
 421 <a href="http://code.google.com/p/unladen-swallow/">Unladen Swallow</a> is a
 422 branch of <a href="http://python.org/">Python</a> intended to be fully
 423 compatible and significantly faster.  It uses LLVM's optimization passes and JIT
 424 compiler.</p>
 425 </div>
 426
 427 <!--=========================================================================-->
 428 <div class="doc_subsection">
 429 <a name="llvm-lua">llvm-lua</a>
 430 </div>
 431
 432 <div class="doc_text">
 433 <p>
 434 <a href="http://code.google.com/p/llvm-lua/">LLVM-Lua</a> uses LLVM to add JIT
 435 and static compiling support to the Lua VM.  Lua bytecode is analyzed to
 436 remove type checks, then LLVM is used to compile the bytecode down to machine
 437 code.</p>
 438 </div>
 439
 440 <!--=========================================================================-->
 441 <div class="doc_subsection">
 442 <a name="icedtea">IcedTea Java Virtual Machine Implementation</a>
 443 </div>
 444
 445 <div class="doc_text">
 446 <p>
 447 <a href="http://icedtea.classpath.org/wiki/Main_Page">IcedTea</a> provides a
 448 harness to build OpenJDK using only free software build tools and to provide
 449 replacements for the not-yet free parts of OpenJDK.  One of the extensions that
 450 IcedTea provides is a new JIT compiler named <a
 451 href="http://icedtea.classpath.org/wiki/ZeroSharkFaq">Shark</a> which uses LLVM
 452 to provide native code generation without introducing processor-dependent
 453 code.
 454 </p>
 455 </div>
 456
 457
 458
 459 <!-- *********************************************************************** -->
 460 <div class="doc_section">
 461   <a name="whatsnew">What's New in LLVM 2.6?</a>
 462 </div>
 463 <!-- *********************************************************************** -->
 464
 465 <div class="doc_text">
 466
 467 <p>This release includes a huge number of bug fixes, performance tweaks and
 468 minor improvements.  Some of the major improvements and new features are listed
 469 in this section.
 470 </p>
 471
 472 </div>
 473
 474 <!--=========================================================================-->
 475 <div class="doc_subsection">
 476 <a name="majorfeatures">Major New Features</a>
 477 </div>
 478
 479 <div class="doc_text">
 480
 481 <p>LLVM 2.6 includes several major new capabilities:</p>
 482
 483 <ul>
 484 <li>New <a href="#compiler-rt">compiler-rt</a>, <A href="#klee">KLEE</a>
 485     and <a href="#mc">machine code toolkit</a> sub-projects.</li>
 486 <li>Debug information now includes line numbers when optimizations are enabled.
 487     This allows statistical sampling tools like OProfile and Shark to map
 488     samples back to source lines.</li>
 489 <li>LLVM now includes new experimental backends to support the MSP430, SystemZ
 490     and BlackFin architectures.</li>
 491 <li>LLVM supports a new <a href="GoldPlugin.html">Gold Linker Plugin</a> which
 492     enables support for <a href="LinkTimeOptimization.html">transparent
 493     link-time optimization</a> on ELF targets when used with the Gold binutils
 494     linker.</li>
 495 <li>LLVM now supports doing optimization and code generation on multiple
 496     threads.  Please see the <a href="ProgrammersManual.html#threading">LLVM
 497     Programmer's Manual</a> for more information.</li>
 498 <li>LLVM now has experimental support for <a
 499     href="http://nondot.org/~sabre/LLVMNotes/EmbeddedMetadata.txt">embedded
 500     metadata</a> in LLVM IR, though the implementation is not guaranteed to be
 501     final and the .bc file format may change in future releases.  Debug info
 502     does not yet use this format in LLVM 2.6.</li>
 503 </ul>
 504
 505 </div>
 506
 507 <!--=========================================================================-->
 508 <div class="doc_subsection">
 509 <a name="coreimprovements">LLVM IR and Core Improvements</a>
 510 </div>
 511
 512 <div class="doc_text">
 513 <p>LLVM IR has several new features for better support of new targets and that
 514 expose new optimization opportunities:</p>
 515
 516 <ul>
 517 <li>The <a href="LangRef.html#i_add">add</a>, <a
 518     href="LangRef.html#i_sub">sub</a> and <a href="LangRef.html#i_mul">mul</a>
 519     instructions have been split into integer and floating point versions (like
 520     divide and remainder), introducing new <a
 521     href="LangRef.html#i_fadd">fadd</a>, <a href="LangRef.html#i_fsub">fsub</a>,
 522     and <a href="LangRef.html#i_fmul">fmul</a> instructions.</li>
 523 <li>The <a href="LangRef.html#i_add">add</a>, <a
 524     href="LangRef.html#i_sub">sub</a> and <a href="LangRef.html#i_mul">mul</a>
 525     instructions now support optional "nsw" and "nuw" bits which indicate that
 526     the operation is guaranteed to not overflow (in the signed or
 527     unsigned case, respectively).  This gives the optimizer more information and
 528     can be used for things like C signed integer values, which are undefined on
 529     overflow.</li>
 530 <li>The <a href="LangRef.html#i_sdiv">sdiv</a> instruction now supports an
 531     optional "exact" flag which indicates that the result of the division is
 532     guaranteed to have a remainder of zero.  This is useful for optimizing pointer
 533     subtraction in C.</li>
 534 <li>The <a href="LangRef.html#i_getelementptr">getelementptr</a> instruction now
 535     supports arbitrary integer index values for array/pointer indices.  This
 536     allows for better code generation on 16-bit pointer targets like PIC16.</li>
 537 <li>The <a href="LangRef.html#i_getelementptr">getelementptr</a> instruction now
 538     supports an "inbounds" optimization hint that tells the optimizer that the
 539     pointer is guaranteed to be within its allocated object.</li>
 540 <li>LLVM now support a series of new linkage types for global values which allow
 541     for better optimization and new capabilities:
 542     <ul>
 543     <li><a href="LangRef.html#linkage_linkonce">linkonce_odr</a> and
 544         <a href="LangRef.html#linkage_weak">weak_odr</a> have the same linkage
 545         semantics as the non-"odr" linkage types.  The difference is that these
 546         linkage types indicate that all definitions of the specified function
 547         are guaranteed to have the same semantics.  This allows inlining
 548         templates functions in C++ but not inlining weak functions in C,
 549         which previously both got the same linkage type.</li>
 550     <li><a href="LangRef.html#linkage_available_externally">available_externally
 551         </a> is a new linkage type that gives the optimizer visibility into the
 552         definition of a function (allowing inlining and side effect analysis)
 553         but that does not cause code to be generated.  This allows better
 554         optimization of "GNU inline" functions, extern templates, etc.</li>
 555     <li><a href="LangRef.html#linkage_linker_private">linker_private</a> is a
 556         new linkage type (which is only useful on Mac OS X) that is used for
 557         some metadata generation and other obscure things.</li>
 558     </ul></li>
 559 <li>Finally, target-specific intrinsics can now return multiple values, which
 560     is useful for modeling target operations with multiple results.</li>
 561 </ul>
 562
 563 </div>
 564
 565 <!--=========================================================================-->
 566 <div class="doc_subsection">
 567 <a name="optimizer">Optimizer Improvements</a>
 568 </div>
 569
 570 <div class="doc_text">
 571
 572 <p>In addition to a large array of minor performance tweaks and bug fixes, this
 573 release includes a few major enhancements and additions to the optimizers:</p>
 574
 575 <ul>
 576
 577 <li>The <a href="Passes.html#scalarrepl">Scalar Replacement of Aggregates</a>
 578     pass has many improvements that allow it to better promote vector unions,
 579     variables which are memset, and much more strange code that can happen to
 580     do bitfield accesses to register operations.  An interesting change is that
 581     it now produces "unusual" integer sizes (like i1704) in some cases and lets
 582     other optimizers clean things up.</li>
 583 <li>The <a href="Passes.html#loop-reduce">Loop Strength Reduction</a> pass now
 584     promotes small integer induction variables to 64-bit on 64-bit targets,
 585     which provides a major performance boost for much numerical code.  It also
 586     promotes shorts to int on 32-bit hosts, etc.  LSR now also analyzes pointer
 587     expressions (e.g. getelementptrs), as well as integers.</li>
 588 <li>The <a href="Passes.html#gvn">GVN</a> pass now eliminates partial
 589     redundancies of loads in simple cases.</li>
 590 <li>The <a href="Passes.html#inline">Inliner</a> now reuses stack space when
 591     inlining similar arrays from multiple callees into one caller.</li>
 592 <li>LLVM includes a new experimental Static Single Information (SSI)
 593     construction pass.</li>
 594
 595 </ul>
 596
 597 </div>
 598
 599
 600 <!--=========================================================================-->
 601 <div class="doc_subsection">
 602 <a name="executionengine">Interpreter and JIT Improvements</a>
 603 </div>
 604
 605 <div class="doc_text">
 606
 607 <ul>
 608 <li>LLVM has a new "EngineBuilder" class which makes it more obvious how to
 609     set up and configure an ExecutionEngine (a JIT or interpreter).</li>
 610 <li>The JIT now supports generating more than 16M of code.</li>
 611 <li>When configured with <tt>--with-oprofile</tt>, the JIT can now inform
 612      OProfile about JIT'd code, allowing OProfile to get line number and function
 613      name information for JIT'd functions.</li>
 614 <li>When "libffi" is available, the LLVM interpreter now uses it, which supports
 615     calling almost arbitrary external (natively compiled) functions.</li>
 616 <li>Clients of the JIT can now register a 'JITEventListener' object to receive
 617     callbacks when the JIT emits or frees machine code. The OProfile support
 618     uses this mechanism.</li>
 619 </ul>
 620
 621 </div>
 622
 623 <!--=========================================================================-->
 624 <div class="doc_subsection">
 625 <a name="codegen">Target Independent Code Generator Improvements</a>
 626 </div>
 627
 628 <div class="doc_text">
 629
 630 <p>We have put a significant amount of work into the code generator
 631 infrastructure, which allows us to implement more aggressive algorithms and make
 632 it run faster:</p>
 633
 634 <ul>
 635
 636 <li>The <tt>llc -asm-verbose</tt> option (exposed from llvm-gcc as <tt>-dA</tt>
 637     and clang as <tt>-fverbose-asm</tt> or <tt>-dA</tt>) now adds a lot of
 638     useful information in comments to
 639     the generated .s file.  This information includes location information (if
 640     built with <tt>-g</tt>) and loop nest information.</li>
 641 <li>The code generator now supports a new MachineVerifier pass which is useful
 642     for finding bugs in targets and codegen passes.</li>
 643 <li>The Machine LICM is now enabled by default.  It hoists instructions out of
 644     loops (such as constant pool loads, loads from read-only stubs, vector
 645     constant synthesization code, etc.) and is currently configured to only do
 646     so when the hoisted operation can be rematerialized.</li>
 647 <li>The Machine Sinking pass is now enabled by default.  This pass moves
 648     side-effect free operations down the CFG so that they are executed on fewer
 649     paths through a function.</li>
 650 <li>The code generator now performs "stack slot coloring" of register spills,
 651     which allows spill slots to be reused.  This leads to smaller stack frames
 652     in cases where there are lots of register spills.</li>
 653 <li>The register allocator has many improvements to take better advantage of
 654     commutable operations, various spiller peephole optimizations, and can now
 655     coalesce cross-register-class copies.</li>
 656 <li>Tblgen now supports multiclass inheritance and a number of new string and
 657     list operations like <tt>!(subst)</tt>, <tt>!(foreach)</tt>, <tt>!car</tt>,
 658     <tt>!cdr</tt>, <tt>!null</tt>, <tt>!if</tt>, <tt>!cast</tt>.
 659     These make the .td files more expressive and allow more aggressive factoring
 660     of duplication across instruction patterns.</li>
 661 <li>Target-specific intrinsics can now be added without having to hack VMCore to
 662     add them.  This makes it easier to maintain out-of-tree targets.</li>
 663 <li>The instruction selector is better at propagating information about values
 664     (such as whether they are sign/zero extended etc.) across basic block
 665     boundaries.</li>
 666 <li>The SelectionDAG datastructure has new nodes for representing buildvector
 667     and <a href="http://llvm.org/PR2957">vector shuffle</a> operations.  This
 668     makes operations and pattern matching more efficient and easier to get
 669     right.</li>
 670 <li>The Prolog/Epilog Insertion Pass now has experimental support for performing
 671     the "shrink wrapping" optimization, which moves spills and reloads around in
 672     the CFG to avoid doing saves on paths that don't need them.</li>
 673 <li>LLVM includes new experimental support for writing ELF .o files directly
 674     from the compiler.  It works well for many simple C testcases, but doesn't
 675     support exception handling, debug info, inline assembly, etc.</li>
 676 <li>Targets can now specify register allocation hints through
 677     <tt>MachineRegisterInfo::setRegAllocationHint</tt>. A regalloc hint consists
 678     of hint type and physical register number. A hint type of zero specifies a
 679     register allocation preference. Other hint type values are target specific
 680     which are resolved by <tt>TargetRegisterInfo::ResolveRegAllocHint</tt>. An
 681     example is the ARM target which uses register hints to request that the
 682     register allocator provide an even / odd register pair to two virtual
 683     registers.</li>
 684 </ul>
 685 </div>
 686
 687 <!--=========================================================================-->
 688 <div class="doc_subsection">
 689 <a name="x86">X86-32 and X86-64 Target Improvements</a>
 690 </div>
 691
 692 <div class="doc_text">
 693 <p>New features of the X86 target include:
 694 </p>
 695
 696 <ul>
 697
 698 <li>SSE 4.2 builtins are now supported.</li>
 699 <li>GCC-compatible soft float modes are now supported, which are typically used
 700     by OS kernels.</li>
 701 <li>X86-64 now models implicit zero extensions better, which allows the code
 702     generator to remove a lot of redundant zexts.  It also models the 8-bit "H"
 703     registers as subregs, which allows them to be used in some tricky
 704     situations.</li>
 705 <li>X86-64 now supports the "local exec" and "initial exec" thread local storage
 706     model.</li>
 707 <li>The vector forms of the <a href="LangRef.html#i_icmp">icmp</a> and <a
 708     href="LangRef.html#i_fcmp">fcmp</a> instructions now select to efficient
 709     SSE operations.</li>
 710 <li>Support for the win64 calling conventions have improved.  The primary
 711     missing feature is support for varargs function definitions.  It seems to
 712     work well for many win64 JIT purposes.</li>
 713 <li>The X86 backend has preliminary support for <a
 714     href="CodeGenerator.html#x86_memory">mapping address spaces to segment
 715     register references</a>.  This allows you to write GS or FS relative memory
 716     accesses directly in LLVM IR for cases where you know exactly what you're
 717     doing (such as in an OS kernel).  There are some known problems with this
 718     support, but it works in simple cases.</li>
 719 <li>The X86 code generator has been refactored to move all global variable
 720     reference logic to one place
 721     (<tt>X86Subtarget::ClassifyGlobalReference</tt>) which
 722     makes it easier to reason about.</li>
 723
 724 </ul>
 725
 726 </div>
 727
 728 <!--=========================================================================-->
 729 <div class="doc_subsection">
 730 <a name="pic16">PIC16 Target Improvements</a>
 731 </div>
 732
 733 <div class="doc_text">
 734 <p>New features of the PIC16 target include:
 735 </p>
 736
 737 <ul>
 738 <li>Support for floating-point, indirect function calls, and
 739     passing/returning aggregate types to functions.
 740 <li>The code generator is able to generate debug info into output COFF files.
 741 <li>Support for placing an object into a specific section or at a specific
 742     address in memory.</li>
 743 </ul>
 744
 745 <p>Things not yet supported:</p>
 746
 747 <ul>
 748 <li>Variable arguments.</li>
 749 <li>Interrupts/programs.</li>
 750 </ul>
 751
 752 </div>
 753
 754 <!--=========================================================================-->
 755 <div class="doc_subsection">
 756 <a name="ARM">ARM Target Improvements</a>
 757 </div>
 758
 759 <div class="doc_text">
 760 <p>New features of the ARM target include:
 761 </p>
 762
 763 <ul>
 764
 765 <li>Preliminary support for processors, such as the Cortex-A8 and Cortex-A9,
 766 that implement version v7-A of the ARM architecture.  The ARM backend now
 767 supports both the Thumb2 and Advanced SIMD (Neon) instruction sets.</li>
 768
 769 <li>The AAPCS-VFP "hard float" calling conventions are also supported with the
 770 <tt>-float-abi=hard</tt> flag.</li>
 771
 772 <li>The ARM calling convention code is now tblgen generated instead of resorting
 773     to C++ code.</li>
 774 </ul>
 775
 776 <p>These features are still somewhat experimental
 777 and subject to change. The Neon intrinsics, in particular, may change in future
 778 releases of LLVM.  ARMv7 support has progressed a lot on top of tree since 2.6
 779 branched.</p>
 780
 781
 782 </div>
 783
 784 <!--=========================================================================-->
 785 <div class="doc_subsection">
 786 <a name="OtherTarget">Other Target Specific Improvements</a>
 787 </div>
 788
 789 <div class="doc_text">
 790 <p>New features of other targets include:
 791 </p>
 792
 793 <ul>
 794 <li>Mips now supports O32 Calling Convention.</li>
 795 <li>Many improvements to the 32-bit PowerPC SVR4 ABI (used on powerpc-linux)
 796     support, lots of bugs fixed.</li>
 797 <li>Added support for the 64-bit PowerPC SVR4 ABI (used on powerpc64-linux).
 798     Needs more testing.</li>
 799 </ul>
 800
 801 </div>
 802
 803 <!--=========================================================================-->
 804 <div class="doc_subsection">
 805 <a name="newapis">New Useful APIs</a>
 806 </div>
 807
 808 <div class="doc_text">
 809
 810 <p>This release includes a number of new APIs that are used internally, which
 811    may also be useful for external clients.
 812 </p>
 813
 814 <ul>
 815 <li>New <a href="http://llvm.org/doxygen/PrettyStackTrace_8h-source.html">
 816     <tt>PrettyStackTrace</tt> class</a> allows crashes of llvm tools (and applications
 817     that integrate them) to provide more detailed indication of what the
 818     compiler was doing at the time of the crash (e.g. running a pass).
 819     At the top level for each LLVM tool, it includes the command line arguments.
 820     </li>
 821 <li>New <a href="http://llvm.org/doxygen/StringRef_8h-source.html">StringRef</a>
 822     and <a href="http://llvm.org/doxygen/Twine_8h-source.html">Twine</a> classes
 823     make operations on character ranges and
 824     string concatenation to be more efficient.  <tt>StringRef</tt> is just a <tt>const
 825     char*</tt> with a length, <tt>Twine</tt> is a light-weight rope.</li>
 826 <li>LLVM has new <tt>WeakVH</tt>, <tt>AssertingVH</tt> and <tt>CallbackVH</tt>
 827     classes, which make it easier to write LLVM IR transformations.  <tt>WeakVH</tt>
 828     is automatically drops to null when the referenced <tt>Value</tt> is deleted,
 829     and is updated across a <tt>replaceAllUsesWith</tt> operation.
 830     <tt>AssertingVH</tt> aborts the program if the
 831     referenced value is destroyed while it is being referenced.  <tt>CallbackVH</tt>
 832     is a customizable class for handling value references.  See <a
 833     href="http://llvm.org/doxygen/ValueHandle_8h-source.html">ValueHandle.h</a>
 834     for more information.</li>
 835 <li>The new '<a href="http://llvm.org/doxygen/Triple_8h-source.html">Triple
 836     </a>' class centralizes a lot of logic that reasons about target
 837     triples.</li>
 838 <li>The new '<a href="http://llvm.org/doxygen/ErrorHandling_8h-source.html">
 839     llvm_report_error()</a>' set of APIs allows tools to embed the LLVM
 840     optimizer and backend and recover from previously unrecoverable errors.</li>
 841 <li>LLVM has new abstractions for <a
 842     href="http://llvm.org/doxygen/Atomic_8h-source.html">atomic operations</a>
 843     and <a href="http://llvm.org/doxygen/RWMutex_8h-source.html">reader/writer
 844     locks</a>.</li>
 845 <li>LLVM has new <a href="http://llvm.org/doxygen/SourceMgr_8h-source.html">
 846     <tt>SourceMgr</tt> and <tt>SMLoc</tt> classes</a> which implement caret
 847     diagnostics and basic include stack processing for simple parsers. It is
 848     used by tablegen, llvm-mc, the .ll parser and FileCheck.</li>
 849 </ul>
 850
 851
 852 </div>
 853
 854 <!--=========================================================================-->
 855 <div class="doc_subsection">
 856 <a name="otherimprovements">Other Improvements and New Features</a>
 857 </div>
 858
 859 <div class="doc_text">
 860 <p>Other miscellaneous features include:</p>
 861
 862 <ul>
 863 <li>LLVM now includes a new internal '<a
 864     href="http://llvm.org/cmds/FileCheck.html">FileCheck</a>' tool which allows
 865     writing much more accurate regression tests that run faster.  Please see the
 866     <a href="TestingGuide.html#FileCheck">FileCheck section of the Testing
 867     Guide</a> for more information.</li>
 868 <li>LLVM profile information support has been significantly improved to produce
 869 correct use counts, and has support for edge profiling with reduced runtime
 870 overhead.  Combined, the generated profile information is both more correct and
 871 imposes about half as much overhead (2.6. from 12% to 6% overhead on SPEC
 872 CPU2000).</li>
 873 <li>The C bindings (in the llvm/include/llvm-c directory) include many newly
 874     supported APIs.</li>
 875 <li>LLVM 2.6 includes a brand new experimental LLVM bindings to the Ada2005
 876     programming language.</li>
 877
 878 <li>The LLVMC driver has several new features:
 879   <ul>
 880   <li>Dynamic plugins now work on Windows.</li>
 881   <li>New option property: init. Makes possible to provide default values for
 882       options defined in plugins (interface to <tt>cl::init</tt>).</li>
 883   <li>New example: Skeleton, shows how to create a standalone LLVMC-based
 884       driver.</li>
 885   <li>New example: mcc16, a driver for the PIC16 toolchain.</li>
 886   </ul>
 887 </li>
 888
 889 </ul>
 890
 891 </div>
 892
 893
 894 <!--=========================================================================-->
 895 <div class="doc_subsection">
 896 <a name="changes">Major Changes and Removed Features</a>
 897 </div>
 898
 899 <div class="doc_text">
 900
 901 <p>If you're already an LLVM user or developer with out-of-tree changes based
 902 on LLVM 2.5, this section lists some "gotchas" that you may run into upgrading
 903 from the previous release.</p>
 904
 905 <ul>
 906 <li>The Itanium (IA64) backend has been removed.  It was not actively supported
 907     and had bitrotted.</li>
 908 <li>The BigBlock register allocator has been removed, it had also bitrotted.</li>
 909 <li>The C Backend (<tt>-march=c</tt>) is no longer considered part of the LLVM release
 910 criteria.  We still want it to work, but no one is maintaining it and it lacks
 911 support for arbitrary precision integers and other important IR features.</li>
 912
 913 <li>All LLVM tools now default to overwriting their output file, behaving more
 914     like standard unix tools.  Previously, this only happened with the '<tt>-f</tt>'
 915     option.</li>
 916 <li>LLVM build now builds all libraries as .a files instead of some
 917   libraries as relinked .o files.  This requires some APIs like
 918   InitializeAllTargets.h.
 919   </li>
 920 </ul>
 921
 922
 923 <p>In addition, many APIs have changed in this release.  Some of the major LLVM
 924 API changes are:</p>
 925
 926 <ul>
 927 <li>All uses of <tt>hash_set</tt> and <tt>hash_map</tt> have been removed from
 928     the LLVM tree and the wrapper headers have been removed.</li>
 929 <li>The llvm/Streams.h and <tt>DOUT</tt> member of Debug.h have been removed.  The
 930     <tt>llvm::Ostream</tt> class has been completely removed and replaced with
 931     uses of <tt>raw_ostream</tt>.</li>
 932 <li>LLVM's global uniquing tables for <tt>Type</tt>s and <tt>Constant</tt>s have
 933     been privatized into members of an <tt>LLVMContext</tt>.  A number of APIs
 934     now take an <tt>LLVMContext</tt> as a parameter.  To smooth the transition
 935     for clients that will only ever use a single context, the new
 936     <tt>getGlobalContext()</tt> API can be used to access a default global
 937     context which can be passed in any and all cases where a context is
 938     required.
 939 <li>The <tt>getABITypeSize</tt> methods are now called <tt>getAllocSize</tt>.</li>
 940 <li>The <tt>Add</tt>, <tt>Sub</tt> and <tt>Mul</tt> operators are no longer
 941     overloaded for floating-point types. Floating-point addition, subtraction
 942     and multiplication are now represented with new operators <tt>FAdd</tt>,
 943     <tt>FSub</tt> and <tt>FMul</tt>. In the <tt>IRBuilder</tt> API,
 944     <tt>CreateAdd</tt>, <tt>CreateSub</tt>, <tt>CreateMul</tt> and
 945     <tt>CreateNeg</tt> should only be used for integer arithmetic now;
 946     <tt>CreateFAdd</tt>, <tt>CreateFSub</tt>, <tt>CreateFMul</tt> and
 947     <tt>CreateFNeg</tt> should now be used for floating-point arithmetic.</li>
 948 <li>The <tt>DynamicLibrary</tt> class can no longer be constructed, its functionality has
 949     moved to static member functions.</li>
 950 <li><tt>raw_fd_ostream</tt>'s constructor for opening a given filename now
 951     takes an extra <tt>Force</tt> argument. If <tt>Force</tt> is set to
 952     <tt>false</tt>, an error will be reported if a file with the given name
 953     already exists. If <tt>Force</tt> is set to <tt>true</tt>, the file will
 954     be silently truncated (which is the behavior before this flag was
 955     added).</li>
 956 <li><tt>SCEVHandle</tt> no longer exists, because reference counting is no
 957     longer done for <tt>SCEV*</tt> objects, instead <tt>const SCEV*</tt>
 958     should be used.</li>
 959
 960 <li>Many APIs, notably <tt>llvm::Value</tt>, now use the <tt>StringRef</tt>
 961 and <tt>Twine</tt> classes instead of passing <tt>const char*</tt>
 962 or <tt>std::string</tt>, as described in
 963 the <a href="ProgrammersManual.html#string_apis">Programmer's Manual</a>. Most
 964 clients should be unaffected by this transition, unless they are used to
 965 <tt>Value::getName()</tt> returning a string. Here are some tips on updating to
 966 2.6:
 967   <ul>
 968     <li><tt>getNameStr()</tt> is still available, and matches the old
 969       behavior. Replacing <tt>getName()</tt> calls with this is an safe option,
 970       although more efficient alternatives are now possible.</li>
 971
 972     <li>If you were just relying on <tt>getName()</tt> being able to be sent to
 973       a <tt>std::ostream</tt>, consider migrating
 974       to <tt>llvm::raw_ostream</tt>.</li>
 975
 976     <li>If you were using <tt>getName().c_str()</tt> to get a <tt>const
 977         char*</tt> pointer to the name, you can use <tt>getName().data()</tt>.
 978         Note that this string (as before), may not be the entire name if the
 979         name contains embedded null characters.</li>
 980
 981     <li>If you were using <tt>operator +</tt> on the result of <tt>getName()</tt> and
 982       treating the result as an <tt>std::string</tt>, you can either
 983       use <tt>Twine::str</tt> to get the result as an <tt>std::string</tt>, or
 984       could move to a <tt>Twine</tt> based design.</li>
 985
 986     <li><tt>isName()</tt> should be replaced with comparison
 987       against <tt>getName()</tt> (this is now efficient).
 988   </ul>
 989 </li>
 990
 991 <li>The registration interfaces for backend Targets has changed (what was
 992 previously <tt>TargetMachineRegistry</tt>). For backend authors, see the <a
 993 href="WritingAnLLVMBackend.html#TargetRegistration">Writing An LLVM Backend</a>
 994 guide. For clients, the notable API changes are:
 995   <ul>
 996     <li><tt>TargetMachineRegistry</tt> has been renamed
 997       to <tt>TargetRegistry</tt>.</li>
 998
 999     <li>Clients should move to using the <tt>TargetRegistry::lookupTarget()</tt>
1000       function to find targets.</li>
1001   </ul>
1002 </li>
1003 </ul>
1004
1005 </div>
1006
1007
1008
1009 <!-- *********************************************************************** -->
1010 <div class="doc_section">
1011   <a name="portability">Portability and Supported Platforms</a>
1012 </div>
1013 <!-- *********************************************************************** -->
1014
1015 <div class="doc_text">
1016
1017 <p>LLVM is known to work on the following platforms:</p>
1018
1019 <ul>
1020 <li>Intel and AMD machines (IA32, X86-64, AMD64, EMT-64) running Red Hat
1021     Linux, Fedora Core, FreeBSD and AuroraUX (and probably other unix-like
1022     systems).</li>
1023 <li>PowerPC and X86-based Mac OS X systems, running 10.3 and above in 32-bit
1024     and 64-bit modes.</li>
1025 <li>Intel and AMD machines running on Win32 using MinGW libraries (native).</li>
1026 <li>Intel and AMD machines running on Win32 with the Cygwin libraries (limited
1027     support is available for native builds with Visual C++).</li>
1028 <li>Sun x86 and AMD64 machines running Solaris 10, OpenSolaris 0906.</li>
1029 <li>Alpha-based machines running Debian GNU/Linux.</li>
1030 </ul>
1031
1032 <p>The core LLVM infrastructure uses GNU autoconf to adapt itself
1033 to the machine and operating system on which it is built.  However, minor
1034 porting may be required to get LLVM to work on new platforms.  We welcome your
1035 portability patches and reports of successful builds or error messages.</p>
1036
1037 </div>
1038
1039 <!-- *********************************************************************** -->
1040 <div class="doc_section">
1041   <a name="knownproblems">Known Problems</a>
1042 </div>
1043 <!-- *********************************************************************** -->
1044
1045 <div class="doc_text">
1046
1047 <p>This section contains significant known problems with the LLVM system,
1048 listed by component.  If you run into a problem, please check the <a
1049 href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
1050 there isn't already one.</p>
1051
1052 <ul>
1053 <li>The llvm-gcc bootstrap will fail with some versions of binutils (e.g. 2.15)
1054     with a message of "<tt><a href="http://llvm.org/PR5004">Error: can not do 8
1055     byte pc-relative relocation</a></tt>" when building C++ code.  We intend to
1056     fix this on mainline, but a workaround for 2.6 is to upgrade to binutils
1057     2.17 or later.</li>
1058
1059 <li>LLVM will not correctly compile on Solaris and/or OpenSolaris
1060 using the stock GCC 3.x.x series 'out the box',
1061 See: <a href="GettingStarted.html#brokengcc">Broken versions of GCC and other tools</a>.
1062 However, A <a href="http://pkg.auroraux.org/GCC">Modern GCC Build</a>
1063 for x86/x86-64 has been made available from the third party AuroraUX Project
1064 that has been meticulously tested for bootstrapping LLVM &amp; Clang.</li>
1065 </ul>
1066
1067 </div>
1068
1069 <!-- ======================================================================= -->
1070 <div class="doc_subsection">
1071   <a name="experimental">Experimental features included with this release</a>
1072 </div>
1073
1074 <div class="doc_text">
1075
1076 <p>The following components of this LLVM release are either untested, known to
1077 be broken or unreliable, or are in early development.  These components should
1078 not be relied on, and bugs should not be filed against them, but they may be
1079 useful to some people.  In particular, if you would like to work on one of these
1080 components, please contact us on the <a
1081 href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev list</a>.</p>
1082
1083 <ul>
1084 <li>The MSIL, Alpha, SPU, MIPS, PIC16, Blackfin, MSP430 and SystemZ backends are
1085     experimental.</li>
1086 <li>The <tt>llc</tt> "<tt>-filetype=asm</tt>" (the default) is the only
1087     supported value for this option.  The ELF writer is experimental.</li>
1088 <li>The implementation of Andersen's Alias Analysis has many known bugs.</li>
1089 </ul>
1090
1091 </div>
1092
1093 <!-- ======================================================================= -->
1094 <div class="doc_subsection">
1095   <a name="x86-be">Known problems with the X86 back-end</a>
1096 </div>
1097
1098 <div class="doc_text">
1099
1100 <ul>
1101   <li>The X86 backend does not yet support
1102     all <a href="http://llvm.org/PR879">inline assembly that uses the X86
1103     floating point stack</a>.  It supports the 'f' and 't' constraints, but not
1104     'u'.</li>
1105   <li>The X86 backend generates inefficient floating point code when configured
1106     to generate code for systems that don't have SSE2.</li>
1107   <li>Win64 code generation wasn't widely tested. Everything should work, but we
1108     expect small issues to happen. Also, llvm-gcc cannot build the mingw64
1109     runtime currently due
1110     to <a href="http://llvm.org/PR2255">several</a>
1111     <a href="http://llvm.org/PR2257">bugs</a> and due to lack of support for
1112     the
1113     'u' inline assembly constraint and for X87 floating point inline assembly.</li>
1114   <li>The X86-64 backend does not yet support the LLVM IR instruction
1115       <tt>va_arg</tt>. Currently, the llvm-gcc and front-ends support variadic
1116       argument constructs on X86-64 by lowering them manually.</li>
1117 </ul>
1118
1119 </div>
1120
1121 <!-- ======================================================================= -->
1122 <div class="doc_subsection">
1123   <a name="ppc-be">Known problems with the PowerPC back-end</a>
1124 </div>
1125
1126 <div class="doc_text">
1127
1128 <ul>
1129 <li>The Linux PPC32/ABI support needs testing for the interpreter and static
1130 compilation, and lacks support for debug information.</li>
1131 </ul>
1132
1133 </div>
1134
1135 <!-- ======================================================================= -->
1136 <div class="doc_subsection">
1137   <a name="arm-be">Known problems with the ARM back-end</a>
1138 </div>
1139
1140 <div class="doc_text">
1141
1142 <ul>
1143 <li>Support for the Advanced SIMD (Neon) instruction set is still incomplete
1144 and not well tested.  Some features may not work at all, and the code quality
1145 may be poor in some cases.</li>
1146 <li>Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6
1147 processors, thumb programs can crash or produce wrong
1148 results (<a href="http://llvm.org/PR1388">PR1388</a>).</li>
1149 <li>Compilation for ARM Linux OABI (old ABI) is supported but not fully tested.
1150 </li>
1151 </ul>
1152
1153 </div>
1154
1155 <!-- ======================================================================= -->
1156 <div class="doc_subsection">
1157   <a name="sparc-be">Known problems with the SPARC back-end</a>
1158 </div>
1159
1160 <div class="doc_text">
1161
1162 <ul>
1163 <li>The SPARC backend only supports the 32-bit SPARC ABI (-m32); it does not
1164     support the 64-bit SPARC ABI (-m64).</li>
1165 </ul>
1166
1167 </div>
1168
1169 <!-- ======================================================================= -->
1170 <div class="doc_subsection">
1171   <a name="mips-be">Known problems with the MIPS back-end</a>
1172 </div>
1173
1174 <div class="doc_text">
1175
1176 <ul>
1177 <li>64-bit MIPS targets are not supported yet.</li>
1178 </ul>
1179
1180 </div>
1181
1182 <!-- ======================================================================= -->
1183 <div class="doc_subsection">
1184   <a name="alpha-be">Known problems with the Alpha back-end</a>
1185 </div>
1186
1187 <div class="doc_text">
1188
1189 <ul>
1190
1191 <li>On 21164s, some rare FP arithmetic sequences which may trap do not have the
1192 appropriate nops inserted to ensure restartability.</li>
1193
1194 </ul>
1195 </div>
1196
1197 <!-- ======================================================================= -->
1198 <div class="doc_subsection">
1199   <a name="c-be">Known problems with the C back-end</a>
1200 </div>
1201
1202 <div class="doc_text">
1203
1204 <ul>
1205 <li><a href="http://llvm.org/PR802">The C backend has only basic support for
1206     inline assembly code</a>.</li>
1207 <li><a href="http://llvm.org/PR1658">The C backend violates the ABI of common
1208     C++ programs</a>, preventing intermixing between C++ compiled by the CBE and
1209     C++ code compiled with <tt>llc</tt> or native compilers.</li>
1210 <li>The C backend does not support all exception handling constructs.</li>
1211 <li>The C backend does not support arbitrary precision integers.</li>
1212 </ul>
1213
1214 </div>
1215
1216
1217 <!-- ======================================================================= -->
1218 <div class="doc_subsection">
1219   <a name="c-fe">Known problems with the llvm-gcc C front-end</a>
1220 </div>
1221
1222 <div class="doc_text">
1223
1224 <p>The only major language feature of GCC not supported by llvm-gcc is
1225     the <tt>__builtin_apply</tt> family of builtins.   However, some extensions
1226     are only supported on some targets.  For example, trampolines are only
1227     supported on some targets (these are used when you take the address of a
1228     nested function).</p>
1229
1230 <p>If you run into GCC extensions which are not supported, please let us know.
1231 </p>
1232
1233 </div>
1234
1235 <!-- ======================================================================= -->
1236 <div class="doc_subsection">
1237   <a name="c++-fe">Known problems with the llvm-gcc C++ front-end</a>
1238 </div>
1239
1240 <div class="doc_text">
1241
1242 <p>The C++ front-end is considered to be fully
1243 tested and works for a number of non-trivial programs, including LLVM
1244 itself, Qt, Mozilla, etc.</p>
1245
1246 <ul>
1247 <li>Exception handling works well on the X86 and PowerPC targets. Currently
1248   only Linux and Darwin targets are supported (both 32 and 64 bit).</li>
1249 </ul>
1250
1251 </div>
1252
1253 <!-- ======================================================================= -->
1254 <div class="doc_subsection">
1255   <a name="fortran-fe">Known problems with the llvm-gcc Fortran front-end</a>
1256 </div>
1257
1258 <div class="doc_text">
1259 <ul>
1260 <li>Fortran support generally works, but there are still several unresolved bugs
1261     in <a href="http://llvm.org/bugs/">Bugzilla</a>.  Please see the
1262     tools/gfortran component for details.</li>
1263 </ul>
1264 </div>
1265
1266 <!-- ======================================================================= -->
1267 <div class="doc_subsection">
1268   <a name="ada-fe">Known problems with the llvm-gcc Ada front-end</a>
1269 </div>
1270
1271 <div class="doc_text">
1272 The llvm-gcc 4.2 Ada compiler works fairly well; however, this is not a mature
1273 technology, and problems should be expected.
1274 <ul>
1275 <li>The Ada front-end currently only builds on X86-32.  This is mainly due
1276 to lack of trampoline support (pointers to nested functions) on other platforms.
1277 However, it <a href="http://llvm.org/PR2006">also fails to build on X86-64</a>
1278 which does support trampolines.</li>
1279 <li>The Ada front-end <a href="http://llvm.org/PR2007">fails to bootstrap</a>.
1280 This is due to lack of LLVM support for <tt>setjmp</tt>/<tt>longjmp</tt> style
1281 exception handling, which is used internally by the compiler.
1282 Workaround: configure with <tt>--disable-bootstrap</tt>.</li>
1283 <li>The c380004, <a href="http://llvm.org/PR2010">c393010</a>
1284 and <a href="http://llvm.org/PR2421">cxg2021</a> ACATS tests fail
1285 (c380004 also fails with gcc-4.2 mainline).
1286 If the compiler is built with checks disabled then <a href="http://llvm.org/PR2010">c393010</a>
1287 causes the compiler to go into an infinite loop, using up all system memory.</li>
1288 <li>Some GCC specific Ada tests continue to crash the compiler.</li>
1289 <li>The <tt>-E</tt> binder option (exception backtraces)
1290 <a href="http://llvm.org/PR1982">does not work</a> and will result in programs
1291 crashing if an exception is raised.  Workaround: do not use <tt>-E</tt>.</li>
1292 <li>Only discrete types <a href="http://llvm.org/PR1981">are allowed to start
1293 or finish at a non-byte offset</a> in a record.  Workaround: do not pack records
1294 or use representation clauses that result in a field of a non-discrete type
1295 starting or finishing in the middle of a byte.</li>
1296 <li>The <tt>lli</tt> interpreter <a href="http://llvm.org/PR2009">considers
1297 'main' as generated by the Ada binder to be invalid</a>.
1298 Workaround: hand edit the file to use pointers for <tt>argv</tt> and
1299 <tt>envp</tt> rather than integers.</li>
1300 <li>The <tt>-fstack-check</tt> option <a href="http://llvm.org/PR2008">is
1301 ignored</a>.</li>
1302 </ul>
1303 </div>
1304
1305 <!-- ======================================================================= -->
1306 <div class="doc_subsection">
1307         <a name="ocaml-bindings">Known problems with the O'Caml bindings</a>
1308 </div>
1309
1310 <div class="doc_text">
1311
1312 <p>The <tt>Llvm.Linkage</tt> module is broken, and has incorrect values. Only
1313 <tt>Llvm.Linkage.External</tt>, <tt>Llvm.Linkage.Available_externally</tt>, and
1314 <tt>Llvm.Linkage.Link_once</tt> will be correct. If you need any of the other linkage
1315 modes, you'll have to write an external C library in order to expose the
1316 functionality. This has been fixed in the trunk.</p>
1317 </div>
1318
1319 <!-- *********************************************************************** -->
1320 <div class="doc_section">
1321   <a name="additionalinfo">Additional Information</a>
1322 </div>
1323 <!-- *********************************************************************** -->
1324
1325 <div class="doc_text">
1326
1327 <p>A wide variety of additional information is available on the <a
1328 href="http://llvm.org">LLVM web page</a>, in particular in the <a
1329 href="http://llvm.org/docs/">documentation</a> section.  The web page also
1330 contains versions of the API documentation which is up-to-date with the
1331 Subversion version of the source code.
1332 You can access versions of these documents specific to this release by going
1333 into the "<tt>llvm/doc/</tt>" directory in the LLVM tree.</p>
1334
1335 <p>If you have any questions or comments about LLVM, please feel free to contact
1336 us via the <a href="http://llvm.org/docs/#maillist"> mailing
1337 lists</a>.</p>
1338
1339 </div>
1340
1341 <!-- *********************************************************************** -->
1342
1343 <hr>
1344 <address>
1345   <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1346   src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1347   <a href="http://validator.w3.org/check/referer"><img
1348   src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1349
1350   <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
1351   Last modified: $Date$
1352 </address>
1353
1354 </body>
1355 </html>