1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
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 2.8 Release Notes</title>
11 <div class="doc_title">LLVM 2.8 Release Notes</div>
13 <img align=right src="http://llvm.org/img/DragonSmall.png"
14 width="136" height="136" alt="LLVM Dragon Logo">
17 <li><a href="#intro">Introduction</a></li>
18 <li><a href="#subproj">Sub-project Status Update</a></li>
19 <li><a href="#externalproj">External Projects Using LLVM 2.8</a></li>
20 <li><a href="#whatsnew">What's New in LLVM 2.8?</a></li>
21 <li><a href="GettingStarted.html">Installation Instructions</a></li>
22 <li><a href="#portability">Portability and Supported Platforms</a></li>
23 <li><a href="#knownproblems">Known Problems</a></li>
24 <li><a href="#additionalinfo">Additional Information</a></li>
27 <div class="doc_author">
28 <p>Written by the <a href="http://llvm.org">LLVM Team</a></p>
31 <h1 style="color:red">These are in-progress notes for the upcoming LLVM 2.8
34 <a href="http://llvm.org/releases/2.7/docs/ReleaseNotes.html">LLVM 2.7
35 Release Notes</a>.</h1>
37 <!-- *********************************************************************** -->
38 <div class="doc_section">
39 <a name="intro">Introduction</a>
41 <!-- *********************************************************************** -->
43 <div class="doc_text">
45 <p>This document contains the release notes for the LLVM Compiler
46 Infrastructure, release 2.8. Here we describe the status of LLVM, including
47 major improvements from the previous release and significant known problems.
48 All LLVM releases may be downloaded from the <a
49 href="http://llvm.org/releases/">LLVM releases web site</a>.</p>
51 <p>For more information about LLVM, including information about the latest
52 release, please check out the <a href="http://llvm.org/">main LLVM
53 web site</a>. If you have questions or comments, the <a
54 href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM Developer's
55 Mailing List</a> is a good place to send them.</p>
57 <p>Note that if you are reading this file from a Subversion checkout or the
58 main LLVM web page, this document applies to the <i>next</i> release, not the
59 current one. To see the release notes for a specific release, please see the
60 <a href="http://llvm.org/releases/">releases page</a>.</p>
67 include/llvm/Analysis/LiveValues.h => Dan
68 lib/Transforms/IPO/MergeFunctions.cpp => consider for 2.8.
69 llvm/Analysis/PointerTracking.h => Edwin wants this, consider for 2.8.
72 lib/Transforms/Utils/SSI.cpp -> ABCD depends on it.
76 <!-- Features that need text if they're finished for 2.8:
79 llvm.dbg.value: variable debug info for optimized code
80 loop dependence analysis
83 <!-- for announcement email:
85 Many new papers added to /pubs/
88 <!-- *********************************************************************** -->
89 <div class="doc_section">
90 <a name="subproj">Sub-project Status Update</a>
92 <!-- *********************************************************************** -->
94 <div class="doc_text">
96 The LLVM 2.8 distribution currently consists of code from the core LLVM
97 repository (which roughly includes the LLVM optimizers, code generators
98 and supporting tools), the Clang repository and the llvm-gcc repository. In
99 addition to this code, the LLVM Project includes other sub-projects that are in
100 development. Here we include updates on these subprojects.
106 <!--=========================================================================-->
107 <div class="doc_subsection">
108 <a name="clang">Clang: C/C++/Objective-C Frontend Toolkit</a>
111 <div class="doc_text">
113 <p><a href="http://clang.llvm.org/">Clang</a> is an LLVM front end for the C,
114 C++, and Objective-C languages. Clang aims to provide a better user experience
115 through expressive diagnostics, a high level of conformance to language
116 standards, fast compilation, and low memory use. Like LLVM, Clang provides a
117 modular, library-based architecture that makes it suitable for creating or
118 integrating with other development tools. Clang is considered a
119 production-quality compiler for C and Objective-C on x86 (32- and 64-bit).</p>
121 <p>In the LLVM 2.7 time-frame, the Clang team has made many improvements:</p>
125 <li>C++ Support: Clang is now capable of self-hosting! While still
126 alpha-quality, Clang's C++ support has matured enough to build LLVM and Clang,
127 and C++ is now enabled by default. See the <a
128 href="http://clang.llvm.org/cxx_compatibility.html">Clang C++ compatibility
129 page</a> for common C++ migration issues.</li>
131 <li>Objective-C: Clang now includes experimental support for an updated
132 Objective-C ABI on non-Darwin platforms. This includes support for non-fragile
133 instance variables and accelerated proxies, as well as greater potential for
134 future optimisations. The new ABI is used when compiling with the
135 -fobjc-nonfragile-abi and -fgnu-runtime options. Code compiled with these
136 options may be mixed with code compiled with GCC or clang using the old GNU ABI,
137 but requires the libobjc2 runtime from the GNUstep project.</li>
139 <li>New warnings: Clang contains a number of new warnings, including
140 control-flow warnings (unreachable code, missing return statements in a
141 non-<code>void</code> function, etc.), sign-comparison warnings, and improved
142 format-string warnings.</li>
144 <li>CIndex API and Python bindings: Clang now includes a C API as part of the
145 CIndex library. Although we may make some changes to the API in the future, it
146 is intended to be stable and has been designed for use by external projects. See
148 doxygen <a href="http://clang.llvm.org/doxygen/group__CINDEX.html">CIndex</a>
149 documentation for more details. The CIndex API also includes a preliminary
150 set of Python bindings.</li>
152 <li>ARM Support: Clang now has ABI support for both the Darwin and Linux ARM
153 ABIs. Coupled with many improvements to the LLVM ARM backend, Clang is now
154 suitable for use as a beta quality ARM compiler.</li>
159 <!--=========================================================================-->
160 <div class="doc_subsection">
161 <a name="clangsa">Clang Static Analyzer</a>
164 <div class="doc_text">
166 <p>The <a href="http://clang-analyzer.llvm.org/">Clang Static Analyzer</a>
167 project is an effort to use static source code analysis techniques to
168 automatically find bugs in C and Objective-C programs (and hopefully <a
169 href="http://clang-analyzer.llvm.org/dev_cxx.html">C++ in the
170 future</a>!). The tool is very good at finding bugs that occur on specific
171 paths through code, such as on error conditions.</p>
173 <p>In the LLVM 2.7 time-frame, the analyzer core has made several major and
174 minor improvements, including better support for tracking the fields of
175 structures, initial support (not enabled by default yet) for doing
176 interprocedural (cross-function) analysis, and new checks have been added.
181 <!--=========================================================================-->
182 <div class="doc_subsection">
183 <a name="vmkit">VMKit: JVM/CLI Virtual Machine Implementation</a>
186 <div class="doc_text">
188 The <a href="http://vmkit.llvm.org/">VMKit project</a> is an implementation of
189 a JVM and a CLI Virtual Machine (Microsoft .NET is an
190 implementation of the CLI) using LLVM for static and just-in-time
194 With the release of LLVM 2.7, VMKit has shifted to a great framework for writing
195 virtual machines. VMKit now offers precise and efficient garbage collection with
196 multi-threading support, thanks to the MMTk memory management toolkit, as well
197 as just in time and ahead of time compilation with LLVM. The major changes in
202 <li>Garbage collection: VMKit now uses the MMTk toolkit for garbage collectors.
203 The first collector to be ported is the MarkSweep collector, which is precise,
204 and drastically improves the performance of VMKit.</li>
205 <li>Line number information in the JVM: by using the debug metadata of LLVM, the
206 JVM now supports precise line number information, useful when printing a stack
208 <li>Interface calls in the JVM: we implemented a variant of the Interface Method
209 Table technique for interface calls in the JVM.
216 <!--=========================================================================-->
217 <div class="doc_subsection">
218 <a name="compiler-rt">compiler-rt: Compiler Runtime Library</a>
221 <div class="doc_text">
223 The new LLVM <a href="http://compiler-rt.llvm.org/">compiler-rt project</a>
224 is a simple library that provides an implementation of the low-level
225 target-specific hooks required by code generation and other runtime components.
226 For example, when compiling for a 32-bit target, converting a double to a 64-bit
227 unsigned integer is compiled into a runtime call to the "__fixunsdfdi"
228 function. The compiler-rt library provides highly optimized implementations of
229 this and other low-level routines (some are 3x faster than the equivalent
230 libgcc routines).</p>
233 All of the code in the compiler-rt project is available under the standard LLVM
234 License, a "BSD-style" license. New in LLVM 2.7: compiler_rt now
235 supports ARM targets.</p>
239 <!--=========================================================================-->
240 <div class="doc_subsection">
241 <a name="dragonegg">DragonEgg: llvm-gcc ported to gcc-4.5</a>
244 <div class="doc_text">
246 <a href="http://dragonegg.llvm.org/">DragonEgg</a> is a port of llvm-gcc to
247 gcc-4.5. Unlike llvm-gcc, which makes many intrusive changes to the underlying
248 gcc-4.2 code, dragonegg in theory does not require any gcc-4.5 modifications
249 whatsoever (currently one small patch is needed). This is thanks to the new
250 <a href="http://gcc.gnu.org/wiki/plugins">gcc plugin architecture</a>, which
251 makes it possible to modify the behaviour of gcc at runtime by loading a plugin,
252 which is nothing more than a dynamic library which conforms to the gcc plugin
253 interface. DragonEgg is a gcc plugin that causes the LLVM optimizers to be run
254 instead of the gcc optimizers, and the LLVM code generators instead of the gcc
255 code generators, just like llvm-gcc. To use it, you add
256 "-fplugin=path/dragonegg.so" to the gcc-4.5 command line, and gcc-4.5 magically
257 becomes llvm-gcc-4.5!
261 DragonEgg is still a work in progress. Currently C works very well, while C++,
262 Ada and Fortran work fairly well. All other languages either don't work at all,
263 or only work poorly. For the moment only the x86-32 and x86-64 targets are
264 supported, and only on linux and darwin (darwin needs an additional gcc patch).
268 DragonEgg is a new project which is seeing its first release with llvm-2.7.
274 <!--=========================================================================-->
275 <div class="doc_subsection">
276 <a name="mc">llvm-mc: Machine Code Toolkit</a>
279 <div class="doc_text">
281 The LLVM Machine Code (aka MC) sub-project of LLVM was created to solve a number
282 of problems in the realm of assembly, disassembly, object file format handling,
283 and a number of other related areas that CPU instruction-set level tools work
284 in. It is a sub-project of LLVM which provides it with a number of advantages
285 over other compilers that do not have tightly integrated assembly-level tools.
286 For a gentle introduction, please see the <a
287 href="http://blog.llvm.org/2010/04/intro-to-llvm-mc-project.html">Intro to the
288 LLVM MC Project Blog Post</a>.
291 <p>2.7 includes major parts of the work required by the new MC Project. A few
292 targets have been refactored to support it, and work is underway to support a
293 native assembler in LLVM. This work is not complete in LLVM 2.7, but it has
294 made substantially more progress on LLVM mainline.</p>
296 <p>One minor example of what MC can do is to transcode an AT&T syntax
297 X86 .s file into intel syntax. You can do this with something like:</p>
299 llvm-mc foo.s -output-asm-variant=1 -o foo-intel.s
305 <!-- *********************************************************************** -->
306 <div class="doc_section">
307 <a name="externalproj">External Open Source Projects Using LLVM 2.7</a>
309 <!-- *********************************************************************** -->
311 <div class="doc_text">
313 <p>An exciting aspect of LLVM is that it is used as an enabling technology for
314 a lot of other language and tools projects. This section lists some of the
315 projects that have already been updated to work with LLVM 2.7.</p>
318 <!--=========================================================================-->
319 <div class="doc_subsection">
320 <a name="pure">Pure</a>
323 <div class="doc_text">
325 <a href="http://pure-lang.googlecode.com/">Pure</a>
326 is an algebraic/functional programming language based on term rewriting.
327 Programs are collections of equations which are used to evaluate expressions in
328 a symbolic fashion. Pure offers dynamic typing, eager and lazy evaluation,
329 lexical closures, a hygienic macro system (also based on term rewriting),
330 built-in list and matrix support (including list and matrix comprehensions) and
331 an easy-to-use C interface. The interpreter uses LLVM as a backend to
332 JIT-compile Pure programs to fast native code.</p>
334 <p>Pure versions 0.43 and later have been tested and are known to work with
335 LLVM 2.7 (and continue to work with older LLVM releases >= 2.5).</p>
339 <!--=========================================================================-->
340 <div class="doc_subsection">
341 <a name="RoadsendPHP">Roadsend PHP</a>
344 <div class="doc_text">
346 <a href="http://code.roadsend.com/rphp">Roadsend PHP</a> (rphp) is an open
347 source implementation of the PHP programming
348 language that uses LLVM for its optimizer, JIT and static compiler. This is a
349 reimplementation of an earlier project that is now based on LLVM.
353 <!--=========================================================================-->
354 <div class="doc_subsection">
355 <a name="UnladenSwallow">Unladen Swallow</a>
358 <div class="doc_text">
360 <a href="http://code.google.com/p/unladen-swallow/">Unladen Swallow</a> is a
361 branch of <a href="http://python.org/">Python</a> intended to be fully
362 compatible and significantly faster. It uses LLVM's optimization passes and JIT
367 <!--=========================================================================-->
368 <div class="doc_subsection">
369 <a name="tce">TTA-based Codesign Environment (TCE)</a>
372 <div class="doc_text">
374 <a href="http://tce.cs.tut.fi/">TCE</a> is a toolset for designing
375 application-specific processors (ASP) based on the Transport triggered
376 architecture (TTA). The toolset provides a complete co-design flow from C/C++
377 programs down to synthesizable VHDL and parallel program binaries. Processor
378 customization points include the register files, function units, supported
379 operations, and the interconnection network.</p>
381 <p>TCE uses llvm-gcc/Clang and LLVM for C/C++ language support, target
382 independent optimizations and also for parts of code generation. It generates
383 new LLVM-based code generators "on the fly" for the designed TTA processors and
384 loads them in to the compiler backend as runtime libraries to avoid per-target
385 recompilation of larger parts of the compiler chain.</p>
389 <!--=========================================================================-->
390 <div class="doc_subsection">
391 <a name="safecode">SAFECode Compiler</a>
394 <div class="doc_text">
396 <a href="http://safecode.cs.illinois.edu">SAFECode</a> is a memory safe C
397 compiler built using LLVM. It takes standard, unannotated C code, analyzes the
398 code to ensure that memory accesses and array indexing operations are safe, and
399 instruments the code with run-time checks when safety cannot be proven
404 <!--=========================================================================-->
405 <div class="doc_subsection">
406 <a name="icedtea">IcedTea Java Virtual Machine Implementation</a>
409 <div class="doc_text">
411 <a href="http://icedtea.classpath.org/wiki/Main_Page">IcedTea</a> provides a
412 harness to build OpenJDK using only free software build tools and to provide
413 replacements for the not-yet free parts of OpenJDK. One of the extensions that
414 IcedTea provides is a new JIT compiler named <a
415 href="http://icedtea.classpath.org/wiki/ZeroSharkFaq">Shark</a> which uses LLVM
416 to provide native code generation without introducing processor-dependent
419 <p>Icedtea6 1.8 and later have been tested and are known to work with
420 LLVM 2.7 (and continue to work with older LLVM releases >= 2.6 as well).
424 <!--=========================================================================-->
425 <div class="doc_subsection">
426 <a name="llvm-lua">LLVM-Lua</a>
429 <div class="doc_text">
431 <a href="http://code.google.com/p/llvm-lua/">LLVM-Lua</a> uses LLVM
432 to add JIT and static compiling support to the Lua VM. Lua
433 bytecode is analyzed to remove type checks, then LLVM is used to compile the
434 bytecode down to machine code.
436 <p>LLVM-Lua 1.2.0 have been tested and is known to work with LLVM 2.7.
440 <!--=========================================================================-->
441 <div class="doc_subsection">
442 <a name="MacRuby">MacRuby</a>
445 <div class="doc_text">
447 <a href="http://macruby.org">MacRuby</a> is an implementation of Ruby based on
448 core Mac OS technologies, sponsored by Apple Inc. It uses LLVM at runtime for
449 optimization passes, JIT compilation and exception handling. It also allows
450 static (ahead-of-time) compilation of Ruby code straight to machine code.
452 <p>The upcoming MacRuby 0.6 release works with LLVM 2.7.
456 <!--=========================================================================-->
457 <div class="doc_subsection">
458 <a name="GHC">Glasgow Haskell Compiler (GHC)</a>
461 <div class="doc_text">
463 <a href="http://www.haskell.org/ghc/">GHC</a> is an open source,
464 state-of-the-art programming suite for Haskell, a standard lazy
465 functional programming language. It includes an optimizing static
466 compiler generating good code for a variety of platforms, together
467 with an interactive system for convenient, quick development.</p>
469 <p>In addition to the existing C and native code generators, GHC now
471 href="http://hackage.haskell.org/trac/ghc/wiki/Commentary/Compiler/Backends/LLVM">LLVM
472 code generator</a>. GHC supports LLVM 2.7.</p>
477 <!-- *********************************************************************** -->
478 <div class="doc_section">
479 <a name="whatsnew">What's New in LLVM 2.7?</a>
481 <!-- *********************************************************************** -->
483 <div class="doc_text">
485 <p>This release includes a huge number of bug fixes, performance tweaks and
486 minor improvements. Some of the major improvements and new features are listed
492 <!--=========================================================================-->
493 <div class="doc_subsection">
494 <a name="orgchanges">LLVM Community Changes</a>
497 <div class="doc_text">
499 <p>In addition to changes to the code, between LLVM 2.6 and 2.7, a number of
500 organization changes have happened:
504 <li>LLVM has a new <a href="http://llvm.org/Logo.html">official logo</a>!</li>
506 <li>Ted Kremenek and Doug Gregor have stepped forward as <a
507 href="http://llvm.org/docs/DeveloperPolicy.html#owners">Code Owners</a> of the
508 Clang static analyzer and the Clang frontend, respectively.</li>
510 <li>LLVM now has an <a href="http://blog.llvm.org">official Blog</a> at
511 <a href="http://blog.llvm.org">http://blog.llvm.org</a>. This is a great way
512 to learn about new LLVM-related features as they are implemented. Several
513 features in this release are already explained on the blog.</li>
515 <li>The LLVM web pages are now checked into the SVN server, in the "www",
516 "www-pubs" and "www-releases" SVN modules. Previously they were hidden in a
517 largely inaccessible old CVS server.</li>
519 <li><a href="http://llvm.org">llvm.org</a> is now hosted on a new (and much
520 faster) server. It is still graciously hosted at the University of Illinois
521 of Urbana Champaign.</li>
525 <!--=========================================================================-->
526 <div class="doc_subsection">
527 <a name="majorfeatures">Major New Features</a>
530 <div class="doc_text">
532 <p>LLVM 2.7 includes several major new capabilities:</p>
535 <li>2.7 includes initial support for the <a
536 href="http://en.wikipedia.org/wiki/MicroBlaze">MicroBlaze</a> target.
537 MicroBlaze is a soft processor core designed for Xilinx FPGAs.</li>
539 <li>2.7 includes a new LLVM IR "extensible metadata" feature. This feature
540 supports many different use cases, including allowing front-end authors to
541 encode source level information into LLVM IR, which is consumed by later
542 language-specific passes. This is a great way to do high-level optimizations
543 like devirtualization, type-based alias analysis, etc. See the <a
544 href="http://blog.llvm.org/2010/04/extensible-metadata-in-llvm-ir.html">
545 Extensible Metadata Blog Post</a> for more information.</li>
547 <li>2.7 encodes <a href="SourceLevelDebugging.html">debug information</a>
548 in a completely new way, built on extensible metadata. The new implementation
549 is much more memory efficient and paves the way for improvements to optimized
550 code debugging experience.</li>
552 <li>2.7 now directly supports taking the address of a label and doing an
553 indirect branch through a pointer. This is particularly useful for
554 interpreter loops, and is used to implement the GCC "address of label"
555 extension. For more information, see the <a
556 href="http://blog.llvm.org/2010/01/address-of-label-and-indirect-branches.html">
557 Address of Label and Indirect Branches in LLVM IR Blog Post</a>.
559 <li>2.7 is the first release to start supporting APIs for assembling and
560 disassembling target machine code. These APIs are useful for a variety of
561 low level clients, and are surfaced in the new "enhanced disassembly" API.
562 For more information see the <a
563 href="http://blog.llvm.org/2010/01/x86-disassembler.html">The X86
564 Disassembler Blog Post</a> for more information.</li>
566 <li>2.7 includes major parts of the work required by the new MC Project,
567 see the <a href="#mc">MC update above</a> for more information.</li>
573 <!--=========================================================================-->
574 <div class="doc_subsection">
575 <a name="coreimprovements">LLVM IR and Core Improvements</a>
578 <div class="doc_text">
579 <p>LLVM IR has several new features for better support of new targets and that
580 expose new optimization opportunities:</p>
583 <li>LLVM IR now supports a 16-bit "half float" data type through <a
584 href="LangRef.html#int_fp16">two new intrinsics</a> and APFloat support.</li>
585 <li>LLVM IR supports two new <a href="LangRef.html#fnattrs">function
586 attributes</a>: inlinehint and alignstack(n). The former is a hint to the
587 optimizer that a function was declared 'inline' and thus the inliner should
588 weight it higher when considering inlining it. The later
589 indicates to the code generator that the function diverges from the platform
590 ABI on stack alignment.</li>
591 <li>The new <a href="LangRef.html#int_objectsize">llvm.objectsize</a> intrinsic
592 allows the optimizer to infer the sizes of memory objects in some cases.
593 This intrinsic is used to implement the GCC <tt>__builtin_object_size</tt>
595 <li>LLVM IR now supports marking load and store instructions with <a
596 href="LangRef.html#i_load">"non-temporal" hints</a> (building on the new
597 metadata feature). This hint encourages the code
598 generator to generate non-temporal accesses when possible, which are useful
599 for code that is carefully managing cache behavior. Currently, only the
600 X86 backend provides target support for this feature.</li>
602 <li>LLVM 2.7 has pre-alpha support for <a
603 href="LangRef.html#t_union">unions in LLVM IR</a>.
604 Unfortunately, this support is not really usable in 2.7, so if you're
605 interested in pushing it forward, please help contribute to LLVM mainline.</li>
607 <!-- BELOW COME ACTUAL 2.8 CHANGES -->
609 <li>LLVM 2.8 changes the internal order of operands in <a
610 href="http://llvm.org/doxygen/classllvm_1_1InvokeInst.html"><tt>InvokeInst</tt></a>
611 and <a href="http://llvm.org/doxygen/classllvm_1_1CallInst.html"><tt>CallInst</tt></a>.
612 To be portable across releases, resort to <tt>CallSite</tt> and the
613 high-level accessors, such as <tt>getCalledValue</tt> and <tt>setUnwindDest</tt>.</li>
620 <!--=========================================================================-->
621 <div class="doc_subsection">
622 <a name="optimizer">Optimizer Improvements</a>
625 <div class="doc_text">
627 <p>In addition to a large array of minor performance tweaks and bug fixes, this
628 release includes a few major enhancements and additions to the optimizers:</p>
632 <li>The inliner now merges arrays stack objects in different callees when
633 inlining multiple call sites into one function. This reduces the stack size
634 of the resultant function.</li>
635 <li>The -basicaa alias analysis pass (which is the default) has been improved to
636 be less dependent on "type safe" pointers. It can now look through bitcasts
637 and other constructs more aggressively, allowing better load/store
639 <li>The load elimination optimization in the GVN Pass [<a
640 href="http://blog.llvm.org/2009/12/introduction-to-load-elimination-in-gvn.html">intro
641 blog post</a>] has been substantially improved to be more aggressive about
642 partial redundancy elimination and do more aggressive phi translation. Please
644 href="http://blog.llvm.org/2009/12/advanced-topics-in-redundant-load.html">
645 Advanced Topics in Redundant Load Elimination with a Focus on PHI Translation
646 Blog Post</a> for more details.</li>
647 <li>The module <a href="LangRef.html#datalayout">target data string</a> now
648 includes a notion of 'native' integer data types for the target. This
649 helps mid-level optimizations avoid promoting complex sequences of
650 operations to data types that are not natively supported (e.g. converting
651 i32 operations to i64 on 32-bit chips).</li>
652 <li>The mid-level optimizer is now conservative when operating on a module with
653 no target data. Previously, it would default to SparcV9 settings, which is
654 not what most people expected.</li>
655 <li>Jump threading is now much more aggressive at simplifying correlated
656 conditionals and threading blocks with otherwise complex logic. It has
657 subsumed the old "Conditional Propagation" pass, and -condprop has been
658 removed from LLVM 2.7.</li>
659 <li>The -instcombine pass has been refactored from being one huge file to being
660 a library of its own. Internally, it uses a customized IRBuilder to clean
661 it up and simplify it.</li>
663 <li>The optimal edge profiling pass is reliable and much more complete than in
664 2.6. It can be used with the llvm-prof tool but isn't wired up to the
665 llvm-gcc and clang command line options yet.</li>
667 <li>A new experimental alias analysis implementation, -scev-aa, has been added.
668 It uses LLVM's Scalar Evolution implementation to do symbolic analysis of
669 pointer offset expressions to disambiguate pointers. It can catch a few
670 cases that basicaa cannot, particularly in complex loop nests.</li>
672 <li>The default pass ordering has been tweaked for improved optimization
680 <!--=========================================================================-->
681 <div class="doc_subsection">
682 <a name="executionengine">Interpreter and JIT Improvements</a>
685 <div class="doc_text">
688 <li>The JIT now supports generating debug information and is compatible with
689 the new GDB 7.0 (and later) interfaces for registering dynamically generated
692 <li>The JIT now <a href="http://llvm.org/PR5184">defaults
693 to compiling eagerly</a> to avoid a race condition in the lazy JIT.
694 Clients that still want the lazy JIT can switch it on by calling
695 <tt>ExecutionEngine::DisableLazyCompilation(false)</tt>.</li>
697 <li>It is now possible to create more than one JIT instance in the same process.
698 These JITs can generate machine code in parallel,
699 although <a href="http://llvm.org/docs/ProgrammersManual.html#jitthreading">you
700 still have to obey the other threading restrictions</a>.</li>
706 <!--=========================================================================-->
707 <div class="doc_subsection">
708 <a name="codegen">Target Independent Code Generator Improvements</a>
711 <div class="doc_text">
713 <p>We have put a significant amount of work into the code generator
714 infrastructure, which allows us to implement more aggressive algorithms and make
718 <li>The 'llc -asm-verbose' option (which is now the default) has been enhanced
719 to emit many useful comments to .s files indicating information about spill
720 slots and loop nest structure. This should make it much easier to read and
721 understand assembly files. This is wired up in llvm-gcc and clang to
722 the <tt>-fverbose-asm</tt> option.</li>
724 <li>New LSR with "full strength reduction" mode, which can reduce address
725 register pressure in loops where address generation is important.</li>
727 <li>A new codegen level Common Subexpression Elimination pass (MachineCSE)
728 is available and enabled by default. It catches redundancies exposed by
730 <li>A new pre-register-allocation tail duplication pass is available and enabled
731 by default, it can substantially improve branch prediction quality in some
733 <li>A new sign and zero extension optimization pass (OptimizeExtsPass)
734 is available and enabled by default. This pass can takes advantage
735 architecture features like x86-64 implicit zero extension behavior and
737 <li>The code generator now supports a mode where it attempts to preserve the
738 order of instructions in the input code. This is important for source that
739 is hand scheduled and extremely sensitive to scheduling. It is compatible
740 with the GCC <tt>-fno-schedule-insns</tt> option.</li>
741 <li>The target-independent code generator now supports generating code with
742 arbitrary numbers of result values. Returning more values than was
743 previously supported is handled by returning through a hidden pointer. In
744 2.7, only the X86 and XCore targets have adopted support for this
746 <li>The code generator now supports generating code that follows the
747 <a href="LangRef.html#callingconv">Glasgow Haskell Compiler Calling
748 Convention</a> and ABI.</li>
749 <li>The "<a href="CodeGenerator.html#selectiondag_select">DAG instruction
750 selection</a>" phase of the code generator has been largely rewritten for
751 2.7. Previously, tblgen spit out tons of C++ code which was compiled and
752 linked into the target to do the pattern matching, now it emits a much
753 smaller table which is read by the target-independent code. The primary
754 advantages of this approach is that the size and compile time of various
755 targets is much improved. The X86 code generator shrunk by 1.5MB of code,
757 <li>Almost the entire code generator has switched to emitting code through the
758 MC interfaces instead of printing textually to the .s file. This led to a
759 number of cleanups and speedups. In 2.7, debug an exception handling
760 information does not go through MC yet.</li>
764 <!--=========================================================================-->
765 <div class="doc_subsection">
766 <a name="x86">X86-32 and X86-64 Target Improvements</a>
769 <div class="doc_text">
770 <p>New features of the X86 target include:
774 <li>The X86 backend now optimizes tails calls much more aggressively for
775 functions that use the standard C calling convention.</li>
776 <li>The X86 backend now models scalar SSE registers as subregs of the SSE vector
777 registers, making the code generator more aggressive in cases where scalars
778 and vector types are mixed.</li>
784 <!--=========================================================================-->
785 <div class="doc_subsection">
786 <a name="ARM">ARM Target Improvements</a>
789 <div class="doc_text">
790 <p>New features of the ARM target include:
795 <li>The ARM backend now generates instructions in unified assembly syntax.</li>
797 <li>llvm-gcc now has complete support for the ARM v7 NEON instruction set. This
798 support differs slightly from the GCC implementation. Please see the
800 href="http://blog.llvm.org/2010/04/arm-advanced-simd-neon-intrinsics-and.html">
801 ARM Advanced SIMD (NEON) Intrinsics and Types in LLVM Blog Post</a> for
802 helpful information if migrating code from GCC to LLVM-GCC.</li>
804 <li>The ARM and Thumb code generators now use register scavenging for stack
805 object address materialization. This allows the use of R3 as a general
806 purpose register in Thumb1 code, as it was previous reserved for use in
807 stack address materialization. Secondly, sequential uses of the same
808 value will now re-use the materialized constant.</li>
810 <li>The ARM backend now has good support for ARMv4 targets and has been tested
811 on StrongARM hardware. Previously, LLVM only supported ARMv4T and
814 <li>Atomic builtins are now supported for ARMv6 and ARMv7 (__sync_synchronize,
815 __sync_fetch_and_add, etc.).</li>
822 <!--=========================================================================-->
823 <div class="doc_subsection">
824 <a name="newapis">New Useful APIs</a>
827 <div class="doc_text">
829 <p>This release includes a number of new APIs that are used internally, which
830 may also be useful for external clients.
834 <li>The optimizer uses the new CodeMetrics class to measure the size of code.
835 Various passes (like the inliner, loop unswitcher, etc) all use this to make
836 more accurate estimates of the code size impact of various
838 <li>A new <a href="http://llvm.org/doxygen/InstructionSimplify_8h-source.html">
839 llvm/Analysis/InstructionSimplify.h</a> interface is available for doing
840 symbolic simplification of instructions (e.g. <tt>a+0</tt> -> <tt>a</tt>)
841 without requiring the instruction to exist. This centralizes a lot of
842 ad-hoc symbolic manipulation code scattered in various passes.</li>
843 <li>The optimizer now uses a new <a
844 href="http://llvm.org/doxygen/SSAUpdater_8h-source.html">SSAUpdater</a>
845 class which efficiently supports
846 doing unstructured SSA update operations. This centralized a bunch of code
847 scattered throughout various passes (e.g. jump threading, lcssa,
848 loop rotate, etc) for doing this sort of thing. The code generator has a
849 similar <a href="http://llvm.org/doxygen/MachineSSAUpdater_8h-source.html">
850 MachineSSAUpdater</a> class.</li>
851 <li>The <a href="http://llvm.org/doxygen/Regex_8h-source.html">
852 llvm/Support/Regex.h</a> header exposes a platform independent regular
853 expression API. Building on this, the <a
854 href="TestingGuide.html#FileCheck">FileCheck</a> utility now supports
855 regular exressions.</li>
856 <li>raw_ostream now supports a circular "debug stream" accessed with "dbgs()".
857 By default, this stream works the same way as "errs()", but if you pass
858 <tt>-debug-buffer-size=1000</tt> to opt, the debug stream is capped to a
859 fixed sized circular buffer and the output is printed at the end of the
860 program's execution. This is helpful if you have a long lived compiler
861 process and you're interested in seeing snapshots in time.</li>
867 <!--=========================================================================-->
868 <div class="doc_subsection">
869 <a name="otherimprovements">Other Improvements and New Features</a>
872 <div class="doc_text">
873 <p>Other miscellaneous features include:</p>
876 <li>You can now build LLVM as a big dynamic library (e.g. "libllvm2.7.so"). To
877 get this, configure LLVM with the --enable-shared option.</li>
879 <li>LLVM command line tools now overwrite their output by default. Previously,
880 they would only do this with -f. This makes them more convenient to use, and
881 behave more like standard unix tools.</li>
883 <li>The opt and llc tools now autodetect whether their input is a .ll or .bc
884 file, and automatically do the right thing. This means you don't need to
885 explicitly use the llvm-as tool for most things.</li>
891 <!--=========================================================================-->
892 <div class="doc_subsection">
893 <a name="changes">Major Changes and Removed Features</a>
896 <div class="doc_text">
898 <p>If you're already an LLVM user or developer with out-of-tree changes based
899 on LLVM 2.6, this section lists some "gotchas" that you may run into upgrading
900 from the previous release.</p>
905 The Andersen's alias analysis ("anders-aa") pass, the Predicate Simplifier
906 ("predsimplify") pass, the LoopVR pass, the GVNPRE pass, and the random sampling
907 profiling ("rsprofiling") passes have all been removed. They were not being
908 actively maintained and had substantial problems. If you are interested in
909 these components, you are welcome to ressurect them from SVN, fix the
910 correctness problems, and resubmit them to mainline.</li>
912 <li>LLVM now defaults to building most libraries with RTTI turned off, providing
913 a code size reduction. Packagers who are interested in building LLVM to support
914 plugins that require RTTI information should build with "make REQUIRE_RTTI=1"
915 and should read the new <a href="Packaging.html">Advice on Packaging LLVM</a>
918 <li>The LLVM interpreter now defaults to <em>not</em> using <tt>libffi</tt> even
919 if you have it installed. This makes it more likely that an LLVM built on one
920 system will work when copied to a similar system. To use <tt>libffi</tt>,
921 configure with <tt>--enable-libffi</tt>.</li>
923 <li>Debug information uses a completely different representation, an LLVM 2.6
924 .bc file should work with LLVM 2.7, but debug info won't come forward.</li>
926 <li>The LLVM 2.6 (and earlier) "malloc" and "free" instructions got removed,
927 along with LowerAllocations pass. Now you should just use a call to the
928 malloc and free functions in libc. These calls are optimized as well as
929 the old instructions were.</li>
932 <p>In addition, many APIs have changed in this release. Some of the major LLVM
937 <li>The <tt>add</tt>, <tt>sub</tt>, and <tt>mul</tt> instructions no longer
938 support floating-point operands. The <tt>fadd</tt>, <tt>fsub</tt>, and
939 <tt>fmul</tt> instructions should be used for this purpose instead.</li>
947 <!-- *********************************************************************** -->
948 <div class="doc_section">
949 <a name="portability">Portability and Supported Platforms</a>
951 <!-- *********************************************************************** -->
953 <div class="doc_text">
955 <p>LLVM is known to work on the following platforms:</p>
958 <li>Intel and AMD machines (IA32, X86-64, AMD64, EMT-64) running Red Hat
959 Linux, Fedora Core, FreeBSD and AuroraUX (and probably other unix-like
961 <li>PowerPC and X86-based Mac OS X systems, running 10.4 and above in 32-bit
962 and 64-bit modes.</li>
963 <li>Intel and AMD machines running on Win32 using MinGW libraries (native).</li>
964 <li>Intel and AMD machines running on Win32 with the Cygwin libraries (limited
965 support is available for native builds with Visual C++).</li>
966 <li>Sun x86 and AMD64 machines running Solaris 10, OpenSolaris 0906.</li>
967 <li>Alpha-based machines running Debian GNU/Linux.</li>
970 <p>The core LLVM infrastructure uses GNU autoconf to adapt itself
971 to the machine and operating system on which it is built. However, minor
972 porting may be required to get LLVM to work on new platforms. We welcome your
973 portability patches and reports of successful builds or error messages.</p>
977 <!-- *********************************************************************** -->
978 <div class="doc_section">
979 <a name="knownproblems">Known Problems</a>
981 <!-- *********************************************************************** -->
983 <div class="doc_text">
985 <p>This section contains significant known problems with the LLVM system,
986 listed by component. If you run into a problem, please check the <a
987 href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
988 there isn't already one.</p>
991 <li>LLVM will not correctly compile on Solaris and/or OpenSolaris
992 using the stock GCC 3.x.x series 'out the box',
993 See: <a href="GettingStarted.html#brokengcc">Broken versions of GCC and other tools</a>.
994 However, A <a href="http://pkg.auroraux.org/GCC">Modern GCC Build</a>
995 for x86/x86-64 has been made available from the third party AuroraUX Project
996 that has been meticulously tested for bootstrapping LLVM & Clang.</li>
1001 <!-- ======================================================================= -->
1002 <div class="doc_subsection">
1003 <a name="experimental">Experimental features included with this release</a>
1006 <div class="doc_text">
1008 <p>The following components of this LLVM release are either untested, known to
1009 be broken or unreliable, or are in early development. These components should
1010 not be relied on, and bugs should not be filed against them, but they may be
1011 useful to some people. In particular, if you would like to work on one of these
1012 components, please contact us on the <a
1013 href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev list</a>.</p>
1016 <li>The MSIL, Alpha, SPU, MIPS, PIC16, Blackfin, MSP430, SystemZ and MicroBlaze
1017 backends are experimental.</li>
1018 <li><tt>llc</tt> "<tt>-filetype=asm</tt>" (the default) is the only
1019 supported value for this option. The MachO writer is experimental, and
1020 works much better in mainline SVN.</li>
1025 <!-- ======================================================================= -->
1026 <div class="doc_subsection">
1027 <a name="x86-be">Known problems with the X86 back-end</a>
1030 <div class="doc_text">
1033 <li>The X86 backend does not yet support
1034 all <a href="http://llvm.org/PR879">inline assembly that uses the X86
1035 floating point stack</a>. It supports the 'f' and 't' constraints, but not
1037 <li>The X86 backend generates inefficient floating point code when configured
1038 to generate code for systems that don't have SSE2.</li>
1039 <li>Win64 code generation wasn't widely tested. Everything should work, but we
1040 expect small issues to happen. Also, llvm-gcc cannot build the mingw64
1041 runtime currently due to lack of support for the 'u' inline assembly
1042 constraint and for X87 floating point inline assembly.</li>
1043 <li>The X86-64 backend does not yet support the LLVM IR instruction
1044 <tt>va_arg</tt>. Currently, front-ends support variadic
1045 argument constructs on X86-64 by lowering them manually.</li>
1050 <!-- ======================================================================= -->
1051 <div class="doc_subsection">
1052 <a name="ppc-be">Known problems with the PowerPC back-end</a>
1055 <div class="doc_text">
1058 <li>The Linux PPC32/ABI support needs testing for the interpreter and static
1059 compilation, and lacks support for debug information.</li>
1064 <!-- ======================================================================= -->
1065 <div class="doc_subsection">
1066 <a name="arm-be">Known problems with the ARM back-end</a>
1069 <div class="doc_text">
1072 <li>Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6
1073 processors, thumb programs can crash or produce wrong
1074 results (<a href="http://llvm.org/PR1388">PR1388</a>).</li>
1075 <li>Compilation for ARM Linux OABI (old ABI) is supported but not fully tested.
1081 <!-- ======================================================================= -->
1082 <div class="doc_subsection">
1083 <a name="sparc-be">Known problems with the SPARC back-end</a>
1086 <div class="doc_text">
1089 <li>The SPARC backend only supports the 32-bit SPARC ABI (-m32); it does not
1090 support the 64-bit SPARC ABI (-m64).</li>
1095 <!-- ======================================================================= -->
1096 <div class="doc_subsection">
1097 <a name="mips-be">Known problems with the MIPS back-end</a>
1100 <div class="doc_text">
1103 <li>64-bit MIPS targets are not supported yet.</li>
1108 <!-- ======================================================================= -->
1109 <div class="doc_subsection">
1110 <a name="alpha-be">Known problems with the Alpha back-end</a>
1113 <div class="doc_text">
1117 <li>On 21164s, some rare FP arithmetic sequences which may trap do not have the
1118 appropriate nops inserted to ensure restartability.</li>
1123 <!-- ======================================================================= -->
1124 <div class="doc_subsection">
1125 <a name="c-be">Known problems with the C back-end</a>
1128 <div class="doc_text">
1131 <li><a href="http://llvm.org/PR802">The C backend has only basic support for
1132 inline assembly code</a>.</li>
1133 <li><a href="http://llvm.org/PR1658">The C backend violates the ABI of common
1134 C++ programs</a>, preventing intermixing between C++ compiled by the CBE and
1135 C++ code compiled with <tt>llc</tt> or native compilers.</li>
1136 <li>The C backend does not support all exception handling constructs.</li>
1137 <li>The C backend does not support arbitrary precision integers.</li>
1143 <!-- ======================================================================= -->
1144 <div class="doc_subsection">
1145 <a name="c-fe">Known problems with the llvm-gcc C and C++ front-end</a>
1148 <div class="doc_text">
1150 <p>The only major language feature of GCC not supported by llvm-gcc is
1151 the <tt>__builtin_apply</tt> family of builtins. However, some extensions
1152 are only supported on some targets. For example, trampolines are only
1153 supported on some targets (these are used when you take the address of a
1154 nested function).</p>
1158 <!-- ======================================================================= -->
1159 <div class="doc_subsection">
1160 <a name="fortran-fe">Known problems with the llvm-gcc Fortran front-end</a>
1163 <div class="doc_text">
1165 <li>Fortran support generally works, but there are still several unresolved bugs
1166 in <a href="http://llvm.org/bugs/">Bugzilla</a>. Please see the
1167 tools/gfortran component for details.</li>
1171 <!-- ======================================================================= -->
1172 <div class="doc_subsection">
1173 <a name="ada-fe">Known problems with the llvm-gcc Ada front-end</a>
1176 <div class="doc_text">
1177 The llvm-gcc 4.2 Ada compiler works fairly well; however, this is not a mature
1178 technology, and problems should be expected.
1180 <li>The Ada front-end currently only builds on X86-32. This is mainly due
1181 to lack of trampoline support (pointers to nested functions) on other platforms.
1182 However, it <a href="http://llvm.org/PR2006">also fails to build on X86-64</a>
1183 which does support trampolines.</li>
1184 <li>The Ada front-end <a href="http://llvm.org/PR2007">fails to bootstrap</a>.
1185 This is due to lack of LLVM support for <tt>setjmp</tt>/<tt>longjmp</tt> style
1186 exception handling, which is used internally by the compiler.
1187 Workaround: configure with <tt>--disable-bootstrap</tt>.</li>
1188 <li>The c380004, <a href="http://llvm.org/PR2010">c393010</a>
1189 and <a href="http://llvm.org/PR2421">cxg2021</a> ACATS tests fail
1190 (c380004 also fails with gcc-4.2 mainline).
1191 If the compiler is built with checks disabled then <a href="http://llvm.org/PR2010">c393010</a>
1192 causes the compiler to go into an infinite loop, using up all system memory.</li>
1193 <li>Some GCC specific Ada tests continue to crash the compiler.</li>
1194 <li>The <tt>-E</tt> binder option (exception backtraces)
1195 <a href="http://llvm.org/PR1982">does not work</a> and will result in programs
1196 crashing if an exception is raised. Workaround: do not use <tt>-E</tt>.</li>
1197 <li>Only discrete types <a href="http://llvm.org/PR1981">are allowed to start
1198 or finish at a non-byte offset</a> in a record. Workaround: do not pack records
1199 or use representation clauses that result in a field of a non-discrete type
1200 starting or finishing in the middle of a byte.</li>
1201 <li>The <tt>lli</tt> interpreter <a href="http://llvm.org/PR2009">considers
1202 'main' as generated by the Ada binder to be invalid</a>.
1203 Workaround: hand edit the file to use pointers for <tt>argv</tt> and
1204 <tt>envp</tt> rather than integers.</li>
1205 <li>The <tt>-fstack-check</tt> option <a href="http://llvm.org/PR2008">is
1210 <!-- *********************************************************************** -->
1211 <div class="doc_section">
1212 <a name="additionalinfo">Additional Information</a>
1214 <!-- *********************************************************************** -->
1216 <div class="doc_text">
1218 <p>A wide variety of additional information is available on the <a
1219 href="http://llvm.org">LLVM web page</a>, in particular in the <a
1220 href="http://llvm.org/docs/">documentation</a> section. The web page also
1221 contains versions of the API documentation which is up-to-date with the
1222 Subversion version of the source code.
1223 You can access versions of these documents specific to this release by going
1224 into the "<tt>llvm/doc/</tt>" directory in the LLVM tree.</p>
1226 <p>If you have any questions or comments about LLVM, please feel free to contact
1227 us via the <a href="http://llvm.org/docs/#maillist"> mailing
1232 <!-- *********************************************************************** -->
1236 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1237 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1238 <a href="http://validator.w3.org/check/referer"><img
1239 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1241 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
1242 Last modified: $Date$