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 3.0 Release Notes</title>
11 <h1>LLVM 3.0 Release Notes</h1>
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 3.0</a></li>
20 <li><a href="#whatsnew">What's New in LLVM 3.0?</a></li>
21 <li><a href="GettingStarted.html">Installation Instructions</a></li>
22 <li><a href="#knownproblems">Known Problems</a></li>
23 <li><a href="#additionalinfo">Additional Information</a></li>
26 <div class="doc_author">
27 <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 3.0
34 <a href="http://llvm.org/releases/2.9/docs/ReleaseNotes.html">LLVM 2.9
35 Release Notes</a>.</h1>
38 <!-- *********************************************************************** -->
40 <a name="intro">Introduction</a>
42 <!-- *********************************************************************** -->
46 <p>This document contains the release notes for the LLVM Compiler
47 Infrastructure, release 3.0. Here we describe the status of LLVM, including
48 major improvements from the previous release, improvements in various
49 subprojects of LLVM, and some of the current users of the code.
50 All LLVM releases may be downloaded from
51 the <a href="http://llvm.org/releases/">LLVM releases web site</a>.</p>
53 <p>For more information about LLVM, including information about the latest
54 release, please check out the <a href="http://llvm.org/">main LLVM web
55 site</a>. If you have questions or comments,
56 the <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM
57 Developer's Mailing List</a> is a good place to send them.</p>
59 <p>Note that if you are reading this file from a Subversion checkout or the main
60 LLVM web page, this document applies to the <i>next</i> release, not the
61 current one. To see the release notes for a specific release, please see the
62 <a href="http://llvm.org/releases/">releases page</a>.</p>
67 <!-- *********************************************************************** -->
69 <a name="subproj">Sub-project Status Update</a>
71 <!-- *********************************************************************** -->
75 <p>The LLVM 3.0 distribution currently consists of code from the core LLVM
76 repository (which roughly includes the LLVM optimizers, code generators and
77 supporting tools), and the Clang repository. In
78 addition to this code, the LLVM Project includes other sub-projects that are
79 in development. Here we include updates on these subprojects.</p>
81 <!--=========================================================================-->
83 <a name="clang">Clang: C/C++/Objective-C Frontend Toolkit</a>
88 <p><a href="http://clang.llvm.org/">Clang</a> is an LLVM front end for the C,
89 C++, and Objective-C languages. Clang aims to provide a better user
90 experience through expressive diagnostics, a high level of conformance to
91 language standards, fast compilation, and low memory use. Like LLVM, Clang
92 provides a modular, library-based architecture that makes it suitable for
93 creating or integrating with other development tools. Clang is considered a
94 production-quality compiler for C, Objective-C, C++ and Objective-C++ on x86
95 (32- and 64-bit), and for Darwin/ARM targets.</p>
97 <p>In the LLVM 3.0 time-frame, the Clang team has made many improvements:</p>
100 <li>Greatly improved support for building C++ applications, with greater
101 stability and better diagnostics.</li>
103 <li><a href="http://clang.llvm.org/cxx_status.html">Improved support</a> for
104 the <a href="http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_detail.htm?csnumber=50372">C++
105 2011</a> standard (aka "C++'0x"), including implementations of non-static data member
106 initializers, alias templates, delegating constructors, range-based
107 for loops, and implicitly-generated move constructors and move assignment
108 operators, among others.</li>
110 <li>Implemented support for some features of the upcoming C1x standard,
111 including static assertions and generic selections.</li>
113 <li>Better detection of include and linking paths for system headers and
114 libraries, especially for Linux distributions.</li>
116 <li>Several improvements to Objective-C support, including:
119 <li><a href="http://clang.llvm.org/docs/AutomaticReferenceCounting.html">
120 Automatic Reference Counting</a> (ARC) and an improved memory model
121 cleanly separating object and C memory.</li>
123 <li>A migration tool for moving manual retain/release code to ARC</li>
125 <li>Better support for data hiding, allowing instance variables to be
126 declared in implementation contexts or class extensions</li>
127 <li>Weak linking support for Objective-C classes</li>
128 <li>Improved static type checking by inferring the return type of methods
129 such as +alloc and -init.</li>
132 Some new Objective-C features require either the Mac OS X 10.7 / iOS 5
133 Objective-C runtime, or version 1.6 or later of the GNUstep Objective-C
134 runtime version.</li>
136 <li>Implemented a number of optimizations in <tt>libclang</tt>, the Clang C
137 interface, to improve the performance of code completion and the mapping
138 from source locations to abstract syntax tree nodes.</li>
142 <p>If Clang rejects your code but another compiler accepts it, please take a
143 look at the <a href="http://clang.llvm.org/compatibility.html">language
144 compatibility</a> guide to make sure this is not intentional or a known
149 <!--=========================================================================-->
151 <a name="dragonegg">DragonEgg: GCC front-ends, LLVM back-end</a>
155 <p><a href="http://dragonegg.llvm.org/">DragonEgg</a> is a
156 <a href="http://gcc.gnu.org/wiki/plugins">gcc plugin</a> that replaces GCC's
157 optimizers and code generators with LLVM's. It works with gcc-4.5 or gcc-4.6,
158 targets the x86-32 and x86-64 processor families, and has been successfully
159 used on the Darwin, FreeBSD, KFreeBSD, Linux and OpenBSD platforms. It fully
160 supports Ada, C, C++ and Fortran. It has partial support for Go, Java, Obj-C
163 <p>The 3.0 release has the following notable changes:</p>
166 <li>GCC version 4.6 is now fully supported.</li>
168 <li>Patching and building GCC is no longer required: the plugin should work
169 with your system GCC (version 4.5 or 4.6; on Debian/Ubuntu systems the
170 gcc-4.5-plugin-dev or gcc-4.6-plugin-dev package is also needed).</li>
172 <li>The <tt>-fplugin-arg-dragonegg-enable-gcc-optzns</tt> option, which runs
173 GCC's optimizers as well as LLVM's, now works much better. This is the
174 option to use if you want ultimate performance! It is still experimental
175 though: it may cause the plugin to crash.</li>
177 <li>The type and constant conversion logic has been almost entirely rewritten,
178 fixing a multitude of obscure bugs.</li>
184 <!--=========================================================================-->
186 <a name="compiler-rt">compiler-rt: Compiler Runtime Library</a>
191 <p>The new LLVM <a href="http://compiler-rt.llvm.org/">compiler-rt project</a>
192 is a simple library that provides an implementation of the low-level
193 target-specific hooks required by code generation and other runtime
194 components. For example, when compiling for a 32-bit target, converting a
195 double to a 64-bit unsigned integer is compiled into a runtime call to the
196 "__fixunsdfdi" function. The compiler-rt library provides highly optimized
197 implementations of this and other low-level routines (some are 3x faster than
198 the equivalent libgcc routines).</p>
200 <p>In the LLVM 3.0 timeframe, the target specific ARM code has converted to
201 "unified" assembly syntax, and several new functions have been added to the
206 <!--=========================================================================-->
208 <a name="lldb">LLDB: Low Level Debugger</a>
213 <p>LLDB is a ground-up implementation of a command line debugger, as well as a
214 debugger API that can be used from other applications. LLDB makes use of the
215 Clang parser to provide high-fidelity expression parsing (particularly for
216 C++) and uses the LLVM JIT for target support.</p>
218 <p>LLDB has advanced by leaps and bounds in the 3.0 timeframe. It is
219 dramatically more stable and useful, and includes both a
220 new <a href="http://lldb.llvm.org/tutorial.html">tutorial</a> and
221 a <a href="http://lldb.llvm.org/lldb-gdb.html">side-by-side comparison with
226 <!--=========================================================================-->
228 <a name="libc++">libc++: C++ Standard Library</a>
233 <p>Like compiler_rt, libc++ is now <a href="DeveloperPolicy.html#license">dual
234 licensed</a> under the MIT and UIUC license, allowing it to be used more
237 <p>Libc++ has been ported to FreeBSD and imported into the base system. It is
238 planned to be the default STL implementation for FreeBSD 10.</p>
242 <!--=========================================================================-->
244 <a name="vmkit">VMKit</a>
249 <p>The <a href="http://vmkit.llvm.org/">VMKit project</a> is an
250 implementation of a Java Virtual Machine (Java VM or JVM) that uses LLVM for
251 static and just-in-time compilation.
253 <p>In the LLVM 3.0 time-frame, VMKit has had significant improvements on both
254 runtime and startup performance:</p>
257 <li>Precompilation: by compiling ahead of time a small subset of Java's core
258 library, the startup performance have been highly optimized to the point that
259 running a 'Hello World' program takes less than 30 milliseconds.</li>
261 <li>Customization: by customizing virtual methods for individual classes,
262 the VM can statically determine the target of a virtual call, and decide to
265 <li>Inlining: the VM does more inlining than it did before, by allowing more
266 bytecode instructions to be inlined, and thanks to customization. It also
267 inlines GC barriers, and object allocations.</li>
269 <li>New exception model: the generated code for a method that does not do
270 any try/catch is not penalized anymore by the eventuality of calling a
271 method that throws an exception. Instead, the method that throws the
272 exception jumps directly to the method that could catch it.</li>
278 <!--=========================================================================-->
280 <a name="LLBrowse">LLBrowse: IR Browser</a>
285 <p><a href="http://llvm.org/svn/llvm-project/llbrowse/trunk/doc/LLBrowse.html">
286 LLBrowse</a> is an interactive viewer for LLVM modules. It can load any LLVM
287 module and displays its contents as an expandable tree view, facilitating an
288 easy way to inspect types, functions, global variables, or metadata nodes. It
289 is fully cross-platform, being based on the popular wxWidgets GUI
295 <!--=========================================================================-->
298 <a name="klee">KLEE: A Symbolic Execution Virtual Machine</a>
303 <a href="http://klee.llvm.org/">KLEE</a> is a symbolic execution framework for
304 programs in LLVM bitcode form. KLEE tries to symbolically evaluate "all" paths
305 through the application and records state transitions that lead to fault
306 states. This allows it to construct testcases that lead to faults and can even
307 be used to verify some algorithms.
315 <!-- *********************************************************************** -->
317 <a name="externalproj">External Open Source Projects Using LLVM 3.0</a>
319 <!-- *********************************************************************** -->
323 <p>An exciting aspect of LLVM is that it is used as an enabling technology for
324 a lot of other language and tools projects. This section lists some of the
325 projects that have already been updated to work with LLVM 3.0.</p>
327 <!--=========================================================================-->
328 <h3>AddressSanitizer</h3>
332 <p><a href="http://code.google.com/p/address-sanitizer/">AddressSanitizer</a>
333 uses compiler instrumentation and a specialized malloc library to find C/C++
334 bugs such as use-after-free and out-of-bound accesses to heap, stack, and
335 globals. The key feature of the tool is speed: the average slowdown
336 introduced by AddressSanitizer is less than 2x.</p>
340 <!--=========================================================================-->
345 <p><a href="http://www.clamav.net">Clam AntiVirus</a> is an open source (GPL)
346 anti-virus toolkit for UNIX, designed especially for e-mail scanning on mail
349 <p>Since version 0.96 it
350 has <a href="http://vrt-sourcefire.blogspot.com/2010/09/introduction-to-clamavs-low-level.html">bytecode
351 signatures</a> that allow writing detections for complex malware.</p>
353 <p>It uses LLVM's JIT to speed up the execution of bytecode on X86, X86-64,
354 PPC32/64, falling back to its own interpreter otherwise. The git version was
355 updated to work with LLVM 3.0.</p>
359 <!--=========================================================================-->
360 <h3>clang_complete for VIM</h3>
364 <p><a href="https://github.com/Rip-Rip/clang_complete">clang_complete</a> is a
365 VIM plugin, that provides accurate C/C++ autocompletion using the clang front
366 end. The development version of clang complete, can directly use libclang
367 which can maintain a cache to speed up auto completion.</p>
371 <!--=========================================================================-->
376 <p><a href="https://bitbucket.org/dwilliamson/clreflect">clReflect</a> is a C++
377 parser that uses clang/LLVM to derive a light-weight reflection database
378 suitable for use in game development. It comes with a very simple runtime
379 library for loading and querying the database, requiring no external
380 dependencies (including CRT), and an additional utility library for object
381 management and serialisation.</p>
385 <!--=========================================================================-->
386 <h3>Cling C++ Interpreter</h3>
390 <p><a href="http://cern.ch/cling">Cling</a> is an interactive compiler interface
391 (aka C++ interpreter). It uses LLVM's JIT and clang; it currently supports
392 C++ and C. It has a prompt interface, runs source files, calls into shared
393 libraries, prints the value of expressions, even does runtime lookup of
394 identifiers (dynamic scopes). And it just behaves like one would expect from
399 <!--=========================================================================-->
400 <h3>Crack Programming Language</h3>
404 <p><a href="http://code.google.com/p/crack-language/">Crack</a> aims to provide
405 the ease of development of a scripting language with the performance of a
406 compiled language. The language derives concepts from C++, Java and Python,
407 incorporating object-oriented programming, operator overloading and strong
412 <!--=========================================================================-->
417 <p><a href="http://eerolanguage.org/">Eero</a> is a fully
418 header-and-binary-compatible dialect of Objective-C 2.0, implemented with a
419 patched version of the Clang/LLVM compiler. It features a streamlined syntax,
420 Python-like indentation, and new operators, for improved readability and
421 reduced code clutter. It also has new features such as limited forms of
422 operator overloading and namespaces, and strict (type-and-operator-safe)
423 enumerations. It is inspired by languages such as Smalltalk, Python, and
428 <!--=========================================================================-->
429 <h3>FAUST Real-Time Audio Signal Processing Language</h3>
433 <p><a href="http://faust.grame.fr/">FAUST</a> is a compiled language for
434 real-time audio signal processing. The name FAUST stands for Functional
435 AUdio STream. Its programming model combines two approaches: functional
436 programming and block diagram composition. In addition with the C, C++, Java
437 output formats, the Faust compiler can now generate LLVM bitcode, and works
443 <!--=========================================================================-->
444 <h3>Glasgow Haskell Compiler (GHC)</h3>
448 <p>GHC is an open source, state-of-the-art programming suite for Haskell, a
449 standard lazy functional programming language. It includes an optimizing
450 static compiler generating good code for a variety of platforms, together
451 with an interactive system for convenient, quick development.</p>
453 <p>GHC 7.0 and onwards include an LLVM code generator, supporting LLVM 2.8 and
454 later. Since LLVM 2.9, GHC now includes experimental support for the ARM
455 platform with LLVM 3.0.</p>
459 <!--=========================================================================-->
464 <p><a href="http://botwars.tk/gwscript/">gwXscript</a> is an object oriented,
465 aspect oriented programming language which can create both executables (ELF,
466 EXE) and shared libraries (DLL, SO, DYNLIB). The compiler is implemented in
467 its own language and translates scripts into LLVM-IR which can be optimized
468 and translated into native code by the LLVM framework. Source code in
469 gwScript contains definitions that expand the namespaces. So you can build
470 your project and simply 'plug out' features by removing a file. The remaining
471 project does not leave scars since you directly separate concerns by the
472 'template' feature of gwX. It is also possible to add new features to a
473 project by just adding files and without editing the original project. This
474 language is used for example to create games or content management systems
475 that should be extendable.</p>
477 <p>gwXscript is strongly typed and offers comfort with its native types string,
478 hash and array. You can easily write new libraries in gwXscript or native
479 code. gwXscript is type safe and users should not be able to crash your
480 program or execute malicious code except code that is eating CPU time.</p>
484 <!--=========================================================================-->
485 <h3>include-what-you-use</h3>
489 <p><a href="http://code.google.com/p/include-what-you-use">include-what-you-use</a>
490 is a tool to ensure that a file directly <code>#include</code>s
491 all <code>.h</code> files that provide a symbol that the file uses. It also
492 removes superfluous <code>#include</code>s from source files.</p>
496 <!--=========================================================================-->
497 <h3>ispc: The Intel SPMD Program Compiler</h3>
501 <p><a href="http://ispc.github.com">ispc</a> is a compiler for "single program,
502 multiple data" (SPMD) programs. It compiles a C-based SPMD programming
503 language to run on the SIMD units of CPUs; it often delivers 5-6x speedups on
504 a single core of a CPU with an 8-wide SIMD unit compared to serial code,
505 while still providing a clean and easy-to-understand programming model. For
506 an introduction to the language and its performance,
507 see <a href="http://ispc.github.com/example.html">the walkthrough</a> of a short
508 example program. ispc is licensed under the BSD license.</p>
512 <!--=========================================================================-->
513 <h3>The Julia Programming Language</h3>
517 <p><a href="http://github.com/JuliaLang/julia">Julia</a> is a high-level,
518 high-performance dynamic language for technical
519 computing. It provides a sophisticated compiler, distributed parallel
520 execution, numerical accuracy, and an extensive mathematical function
521 library. The compiler uses type inference to generate fast code
522 without any type declarations, and uses LLVM's optimization passes and
523 JIT compiler. The language is designed around multiple dispatch,
524 giving programs a large degree of flexibility. It is ready for use on many
525 kinds of problems.</p>
528 <!--=========================================================================-->
529 <h3>LanguageKit and Pragmatic Smalltalk</h3>
533 <p><a href="http://etoileos.com/etoile/features/languagekit/">LanguageKit</a> is
534 a framework for implementing dynamic languages sharing an object model with
535 Objective-C. It provides static and JIT compilation using LLVM along with
536 its own interpreter. Pragmatic Smalltalk is a dialect of Smalltalk, built on
537 top of LanguageKit, that interfaces directly with Objective-C, sharing the
538 same object representation and message sending behaviour. These projects are
539 developed as part of the Étoilé desktop environment.</p>
543 <!--=========================================================================-->
548 <p><a href="http://lua-av.mat.ucsb.edu/blog/">LuaAV</a> is a real-time
549 audiovisual scripting environment based around the Lua language and a
550 collection of libraries for sound, graphics, and other media protocols. LuaAV
551 uses LLVM and Clang to JIT compile efficient user-defined audio synthesis
552 routines specified in a declarative syntax.</p>
556 <!--=========================================================================-->
561 <p>An open source, cross-platform implementation of C# and the CLR that is
562 binary compatible with Microsoft.NET. Has an optional, dynamically-loaded
563 LLVM code generation backend in Mini, the JIT compiler.</p>
565 <p>Note that we use a Git mirror of LLVM with some patches. See:
566 https://github.com/mono/llvm</p>
570 <!--=========================================================================-->
575 <p><a href="http://polly.grosser.es">Polly</a> is an advanced data-locality
576 optimizer and automatic parallelizer. It uses an advanced, mathematical
577 model to calculate detailed data dependency information which it uses to
578 optimize the loop structure of a program. Polly can speed up sequential code
579 by improving memory locality and consequently the cache use. Furthermore,
580 Polly is able to expose different kind of parallelism which it exploits by
581 introducing (basic) OpenMP and SIMD code. A mid-term goal of Polly is to
582 automatically create optimized GPU code.</p>
586 <!--=========================================================================-->
587 <h3>Portable OpenCL (pocl)</h3>
591 <p>Portable OpenCL is an open source implementation of the OpenCL standard which
592 can be easily adapted for new targets. One of the goals of the project is
593 improving performance portability of OpenCL programs, avoiding the need for
594 target-dependent manual optimizations. A "native" target is included, which
595 allows running OpenCL kernels on the host (CPU).</p>
599 <!--=========================================================================-->
603 <p><a href="http://pure-lang.googlecode.com/">Pure</a> is an
604 algebraic/functional programming language based on term rewriting. Programs
605 are collections of equations which are used to evaluate expressions in a
606 symbolic fashion. The interpreter uses LLVM as a backend to JIT-compile Pure
607 programs to fast native code. Pure offers dynamic typing, eager and lazy
608 evaluation, lexical closures, a hygienic macro system (also based on term
609 rewriting), built-in list and matrix support (including list and matrix
610 comprehensions) and an easy-to-use interface to C and other programming
611 languages (including the ability to load LLVM bitcode modules, and inline C,
612 C++, Fortran and Faust code in Pure programs if the corresponding LLVM-enabled
613 compilers are installed).</p>
615 <p>Pure version 0.48 has been tested and is known to work with LLVM 3.0
616 (and continues to work with older LLVM releases >= 2.5).</p>
620 <!--=========================================================================-->
621 <h3>Renderscript</h3>
625 <p><a href="http://developer.android.com/guide/topics/renderscript/index.html">Renderscript</a>
626 is Android's advanced 3D graphics rendering and compute API. It provides a
627 portable C99-based language with extensions to facilitate common use cases
628 for enhancing graphics and thread level parallelism. The Renderscript
629 compiler frontend is based on Clang/LLVM. It emits a portable bitcode format
630 for the actual compiled script code, as well as reflects a Java interface for
631 developers to control the execution of the compiled bitcode. Executable
632 machine code is then generated from this bitcode by an LLVM backend on the
633 device. Renderscript is thus able to provide a mechanism by which Android
634 developers can improve performance of their applications while retaining
639 <!--=========================================================================-->
644 <p><a href="http://safecode.cs.illinois.edu">SAFECode</a> is a memory safe C/C++
645 compiler built using LLVM. It takes standard, unannotated C/C++ code,
646 analyzes the code to ensure that memory accesses and array indexing
647 operations are safe, and instruments the code with run-time checks when
648 safety cannot be proven statically. SAFECode can be used as a debugging aid
649 (like Valgrind) to find and repair memory safety bugs. It can also be used
650 to protect code from security attacks at run-time.</p>
654 <!--=========================================================================-->
655 <h3>The Stupid D Compiler (SDC)</h3>
659 <p><a href="https://github.com/bhelyer/SDC">The Stupid D Compiler</a> is a
660 project seeking to write a self-hosting compiler for the D programming
661 language without using the frontend of the reference compiler (DMD).</p>
665 <!--=========================================================================-->
666 <h3>TTA-based Co-design Environment (TCE)</h3>
670 <p>TCE is a toolset for designing application-specific processors (ASP) based on
671 the Transport triggered architecture (TTA). The toolset provides a complete
672 co-design flow from C/C++ programs down to synthesizable VHDL and parallel
673 program binaries. Processor customization points include the register files,
674 function units, supported operations, and the interconnection network.</p>
676 <p>TCE uses Clang and LLVM for C/C++ language support, target independent
677 optimizations and also for parts of code generation. It generates new
678 LLVM-based code generators "on the fly" for the designed TTA processors and
679 loads them in to the compiler backend as runtime libraries to avoid
680 per-target recompilation of larger parts of the compiler chain.</p>
684 <!--=========================================================================-->
685 <h3>Tart Programming Language</h3>
689 <p><a href="http://code.google.com/p/tart/">Tart</a> is a general-purpose,
690 strongly typed programming language designed for application
691 developers. Strongly inspired by Python and C#, Tart focuses on practical
692 solutions for the professional software developer, while avoiding the clutter
693 and boilerplate of legacy languages like Java and C++. Although Tart is still
694 in development, the current implementation supports many features expected of
695 a modern programming language, such as garbage collection, powerful
696 bidirectional type inference, a greatly simplified syntax for template
697 metaprogramming, closures and function literals, reflection, operator
698 overloading, explicit mutability and immutability, and much more. Tart is
699 flexible enough to accommodate a broad range of programming styles and
700 philosophies, while maintaining a strong commitment to simplicity, minimalism
701 and elegance in design.</p>
705 <!--=========================================================================-->
706 <h3>ThreadSanitizer</h3>
710 <p><a href="http://code.google.com/p/data-race-test/">ThreadSanitizer</a> is a
711 data race detector for (mostly) C and C++ code, available for Linux, Mac OS
712 and Windows. On different systems, we use binary instrumentation frameworks
713 (Valgrind and Pin) as frontends that generate the program events for the race
714 detection algorithm. On Linux, there's an option of using LLVM-based
715 compile-time instrumentation.</p>
721 <!-- *********************************************************************** -->
723 <a name="whatsnew">What's New in LLVM 3.0?</a>
725 <!-- *********************************************************************** -->
729 <p>This release includes a huge number of bug fixes, performance tweaks and
730 minor improvements. Some of the major improvements and new features are
731 listed in this section.</p>
733 <!--=========================================================================-->
735 <a name="majorfeatures">Major New Features</a>
740 <!-- Features that need text if they're finished for 3.1:
744 loop dependence analysis
745 CorrelatedValuePropagation
746 lib/Transforms/IPO/MergeFunctions.cpp => consider for 3.1.
749 <p><b>llvm-gcc is gone</b>. LLVM's configure script doesn't depend on llvm-gcc anymore, clean layering.</p>
751 <p>LLVM 3.0 includes several major new capabilities:</p>
754 Analysis/RegionInfo.h + Dom Frontiers
755 SparseBitVector: used in LiveVar.
756 llvm/lib/Archive - replace with lib object?
760 Type system rewrite: http://blog.llvm.org/2011/11/llvm-30-type-system-rewrite.html
761 Better performance for Neon code in clang due to SRoA improvements.
762 New regalloc on by default. Lin scan going away in 3.1
763 PGO / builtin_expect improvements (summary needed)
765 AVX support, assembler, compiler and disassembler.
766 IndVar improvements: andy
767 PTX backend improvements: Justin
768 llvm-rtdyld & MC JIT: JimG
769 InstAliases now automatically used in the asmprinter where they are shorter.
770 Integrated assembler on by default for arm/thumb?
771 PostOrder Dominator frontiers were removed.
772 Line Profiling / gcov support
773 EH and debug information produced with CFI directives, yielding smaller executables: http://blog.mozilla.com/respindola/2011/05/12/cfi-directives/
774 X86-64 generates smaller and faster code at -O0 (fast isel improvements)
775 Better code generation for Cortex-A9
776 Many APIs take ArrayRef's now.
777 Pass manager extension API.
778 ARM inline asm constraints implemented.
779 LangRef.html#fnattrs uwtable attribute for asynch unwind tables.
780 better performance for indirect gotos.
781 llvm.prefetch now takes a 4th argument that specifies whether the prefetch happens to the icache or dcache.
782 New PackedVector, TinyPtrVector class (see Programmer's Manual)
783 New nonlazybind function attribute.
784 ARC language specific optimizer (Transforms/ObjCARC) a decent example of language-specific transformation.
785 LLVM 3.0 removes support for reading LLVM 2.8 and earlier files. Aim to maintain compatibility all the way back to 3.0 "forever".
787 New llvm.expect intrinsic.
788 Table generated MC expansion logic for pseudo instructions that expand to multiple MC instructions through the PseudoInstExpansion class. (JimG)
789 New llvm.fma intrinsic.
791 Euro dev meeting and main one too.
792 New atomics instructions, "#i_fence" instruction, cmpxchg, atomicrmw too. What target support (X86/ARM)? Also 'atomic load/store'. See Atomics.html
793 X86: inline assembler supports .code32 and .code64.
794 Exception handling rewrite: new landingpad and resume instruction. Unwind gone.
795 LowerSetJmp pass removed, unused.
796 llvm-objdump / dwarf parser library / llvm-dwarfdump (d0k)
797 object file parsing stuff and llvm-size (mspencer)
799 Old arm disassembler replaced with a new one based on autogenerated encoding information from ARM .td files.
800 Frontend tests removed from llvm/test/Frontend* (was this completed for 3.0?)
801 Segmented stack support (X86 only?) Rafael and Sanjoy Das: docs/SegmentedStacks.html should be in CodeGen.html status table?
802 X86 backend support for NaCl (David Meyer / Nick L)
803 Codegen now supports vector "select" operations on vector comparisons, turning
804 them into various optimized code sequences (e.g. using the SSE4/AVX "blend"
806 #line directives in integrated assembler
807 SSE domain fixing code enabled for AVX (Bruno/Jakob). Domain fixing pass is
808 now target independent (ExecutionDepsFix pass). (Jakob)
809 X86 backend synthesizes horizontal add/sub instructions from generic code.
810 returns_twice attribute (rafael)
811 Tablegen has been split into a library, clang tblgen pieces now live in clang.
812 The llvm version is now named llvm-tblgen instead of tblgen.
813 X86: Tons of encoding improvements and new instructions (e.g. Atom, Ivy Bridge,
814 and BMI instructions)
815 added to assembler and disassembler (Craig Topper)
816 data layout string can encode the natural alignment of the target's stack for better optimization (LangRef.html#datalayout)
829 <!--=========================================================================-->
831 <a name="coreimprovements">LLVM IR and Core Improvements</a>
836 <p>LLVM IR has several new features for better support of new targets and that
837 expose new optimization opportunities:</p>
839 <p>One of the biggest changes is that 3.0 has a new exception handling
840 system. The old system used LLVM intrinsics to convey the exception handling
841 information to the code generator. It worked in most cases, but not
842 all. Inlining was especially difficult to get right. Also, the intrinsics
843 could be moved away from the <code>invoke</code> instruction, making it hard
844 to recover that information.</p>
846 <p>The new EH system makes exception handling a first-class member of the IR. It
847 adds two new instructions:</p>
850 <li><a href="LangRef.html#i_landingpad"><code>landingpad</code></a> —
851 this instruction defines a landing pad basic block. It contains all of the
852 information that's needed by the code generator. It's also required to be
853 the first non-PHI instruction in the landing pad. In addition, a landing
854 pad may be jumped to only by the unwind edge of an <code>invoke</code>
857 <li><a href="LangRef.html#i_resume"><code>resume</code></a> — this
858 instruction causes the current exception to resume traveling up the
859 stack. It replaces the <code>@llvm.eh.resume</code> intrinsic.</li>
862 <p>Converting from the old EH API to the new EH API is rather simple, because a
863 lot of complexity has been removed. The two intrinsics,
864 <code>@llvm.eh.exception</code> and <code>@llvm.eh.selector</code> have been
865 superseded by the <code>landingpad</code> instruction. Instead of generating
866 a call to <code>@llvm.eh.exception</code> and <code>@llvm.eh.selector</code>:
868 <div class="doc_code">
870 Function *ExcIntr = Intrinsic::getDeclaration(TheModule,
871 Intrinsic::eh_exception);
872 Function *SlctrIntr = Intrinsic::getDeclaration(TheModule,
873 Intrinsic::eh_selector);
875 // The exception pointer.
876 Value *ExnPtr = Builder.CreateCall(ExcIntr, "exc_ptr");
878 std::vector<Value*> Args;
879 Args.push_back(ExnPtr);
880 Args.push_back(Builder.CreateBitCast(Personality,
881 Type::getInt8PtrTy(Context)));
883 <i>// Add selector clauses to Args.</i>
885 // The selector call.
886 Builder.CreateCall(SlctrIntr, Args, "exc_sel");
890 <p>You should instead generate a <code>landingpad</code> instruction, that
891 returns an exception object and selector value:</p>
893 <div class="doc_code">
895 LandingPadInst *LPadInst =
896 Builder.CreateLandingPad(StructType::get(Int8PtrTy, Int32Ty, NULL),
899 Value *LPadExn = Builder.CreateExtractValue(LPadInst, 0);
900 Builder.CreateStore(LPadExn, getExceptionSlot());
902 Value *LPadSel = Builder.CreateExtractValue(LPadInst, 1);
903 Builder.CreateStore(LPadSel, getEHSelectorSlot());
907 <p>It's now trivial to add the individual clauses to the <code>landingpad</code>
910 <div class="doc_code">
912 <i><b>// Adding a catch clause</b></i>
913 Constant *TypeInfo = getTypeInfo();
914 LPadInst->addClause(TypeInfo);
916 <i><b>// Adding a C++ catch-all</b></i>
917 LPadInst->addClause(Constant::getNullValue(Builder.getInt8PtrTy()));
919 <i><b>// Adding a cleanup</b></i>
920 LPadInst->setCleanup(true);
922 <i><b>// Adding a filter clause</b></i>
923 std::vector<Constant*> TypeInfos;
924 Constant *TypeInfo = getFilterTypeInfo();
925 TypeInfos.push_back(Builder.CreateBitCast(TypeInfo, Builder.getInt8PtrTy()));
927 ArrayType *FilterTy = ArrayType::get(Int8PtrTy, TypeInfos.size());
928 LPadInst->addClause(ConstantArray::get(FilterTy, TypeInfos));
932 <p>Converting from using the <code>@llvm.eh.resume</code> intrinsic to
933 the <code>resume</code> instruction is trivial. It takes the exception
934 pointer and exception selector values returned by
935 the <code>landingpad</code> instruction:</p>
937 <div class="doc_code">
939 Type *UnwindDataTy = StructType::get(Builder.getInt8PtrTy(),
940 Builder.getInt32Ty(), NULL);
941 Value *UnwindData = UndefValue::get(UnwindDataTy);
942 Value *ExcPtr = Builder.CreateLoad(getExceptionObjSlot());
943 Value *ExcSel = Builder.CreateLoad(getExceptionSelSlot());
944 UnwindData = Builder.CreateInsertValue(UnwindData, ExcPtr, 0, "exc_ptr");
945 UnwindData = Builder.CreateInsertValue(UnwindData, ExcSel, 1, "exc_sel");
946 Builder.CreateResume(UnwindData);
952 <!--=========================================================================-->
954 <a name="loopoptimization">Loop Optimization Improvements</a>
958 <p>The induction variable simplification pass in 3.0 only modifies
959 induction variables when profitable. Sign and zero extension
960 elimination, linear function test replacement, loop unrolling, and
961 other simplifications that require induction variable analysis have
962 been generalized so they no longer require loops to be rewritten in a
963 typically suboptimal form prior to optimization. This new design
964 preserves more IR level information, avoids undoing earlier loop
965 optimizations (particularly hand-optimized loops), and no longer
966 strongly depends on the code generator rewriting loops a second time
967 in a now optimal form--an intractable problem.</p>
969 <p>The original behavior can be restored with -mllvm -enable-iv-rewrite;
970 however, support for this mode will be short lived. As such, bug
971 reports should be filed for any significant performance regressions
972 when moving from -mllvm -enable-iv-rewrite to the 3.0 default mode.</p>
975 <!--=========================================================================-->
977 <a name="optimizer">Optimizer Improvements</a>
982 <p>In addition to a large array of minor performance tweaks and bug fixes, this
983 release includes a few major enhancements and additions to the
987 <li>Information about <a href="BranchWeightMetadata.html">branch probability</a>
988 and basic block frequency is now available within LLVM, based on a
989 combination of static branch prediction heuristics and
990 <code>__builtin_expect</code> calls. That information is currently used for
991 register spill placement and if-conversion, with additional optimizations
992 planned for future releases. The same framework is intended for eventual
993 use with profile-guided optimization.</li>
1000 <!--=========================================================================-->
1002 <a name="mc">MC Level Improvements</a>
1007 <p>The LLVM Machine Code (aka MC) subsystem was created to solve a number of
1008 problems in the realm of assembly, disassembly, object file format handling,
1009 and a number of other related areas that CPU instruction-set level tools work
1013 <li>The ELF object streamers are much more full featured.</li>
1014 <li>Target dependent relocation handling has been refactored into the Targets.</li>
1015 <li>Early stage MC-JIT infrastructure has been implemented.</li>
1018 <p>The MC-JIT is a major new feature for MC, and will eventually grow to replace
1019 the current JIT implementation. It emits object files direct to memory and
1020 uses a runtime dynamic linker to resolve references and drive lazy compilation.
1021 The MC-JIT enables much greater code reuse between the JIT and the static
1022 compiler and provides better integration with the platform ABI as a result.</p>
1024 <p>For more information, please see
1025 the <a href="http://blog.llvm.org/2010/04/intro-to-llvm-mc-project.html">Intro
1026 to the LLVM MC Project Blog Post</a>.</p>
1030 <!--=========================================================================-->
1032 <a name="codegen">Target Independent Code Generator Improvements</a>
1037 <p>We have put a significant amount of work into the code generator
1038 infrastructure, which allows us to implement more aggressive algorithms and
1039 make it run faster:</p>
1048 <!--=========================================================================-->
1050 <a name="x86">X86-32 and X86-64 Target Improvements</a>
1055 <p>New features and major changes in the X86 target include:</p>
1058 <li>The X86 backend, assembler and disassembler now completely support AVX.
1059 To enable it pass <code>-mavx</code> to the compiler.</li>
1061 <li>The X86 backend now supports
1062 all <a href="http://llvm.org/PR879">inline assembly that uses the X86
1063 floating point stack</a>.</li>
1065 <li>The CRC32 intrinsics have been renamed. The intrinsics were previously
1066 <code>@llvm.x86.sse42.crc32.[8|16|32]</code>
1067 and <code>@llvm.x86.sse42.crc64.[8|64]</code>. They have been renamed to
1068 <code>@llvm.x86.sse42.crc32.32.[8|16|32]</code> and
1069 <code>@llvm.x86.sse42.crc32.64.[8|64]</code>.</li>
1075 <!--=========================================================================-->
1077 <a name="ARM">ARM Target Improvements</a>
1082 <p>New features of the ARM target include:</p>
1085 <li>Reworked Set Jump Long Jump EH Lowering,</li>
1086 <li>improved support for Cortex-M series processors, and</li>
1087 <li>beta quality integrated assembler support.</li>
1092 <!--=========================================================================-->
1094 <a name="MIPS">MIPS Target Improvements</a>
1099 <p>New features and major changes in the MIPS target include:</p>
1102 <li>Most MIPS32r1 and r2 instructions are now supported.</li>
1103 <li>LE/BE MIPS32r1/r2 has been tested extensively.</li>
1104 <li>O32 ABI has been fully tested.</li>
1105 <li>MIPS backend has migrated to using the MC infrastructure for assembly printing. Initial support for direct object code emission has been implemented too.</li>
1106 <li>Delay slot filler has been updated. Now it tries to fill delay slots with useful instructions instead of always filling them with NOPs.</li>
1107 <li>Support for old-style JIT is complete.</li>
1108 <li>Support for old architectures (MIPS1 and MIPS2) has been removed.</li>
1109 <li>Initial support for MIPS64 has been added.</li>
1113 <!--=========================================================================-->
1115 <a name="PTX">PTX Target Improvements</a>
1121 The PTX back-end is still experimental, but is fairly usable for compute kernels
1122 in LLVM 3.0. Most scalar arithmetic is implemented, as well as intrinsics to
1123 access the special PTX registers and sync instructions. The major missing
1124 pieces are texture/sampler support and some vector operations.</p>
1126 <p>That said, the backend is already being used for domain-specific languages
1127 and works well with the <a href="http://www.pcc.me.uk/~peter/libclc/">libclc
1128 library</a> to supply OpenCL built-ins. With it, you can use Clang to compile
1129 OpenCL code into PTX and execute it by loading the resulting PTX as a binary
1130 blob using the nVidia OpenCL library. It has been tested with several OpenCL
1131 programs, including some from the nVidia GPU Computing SDK, and the performance
1132 is on par with the nVidia compiler.</p>
1136 <!--=========================================================================-->
1138 <a name="OtherTS">Other Target Specific Improvements</a>
1143 <p>PPC32/ELF va_arg was implemented.</p>
1144 <p>PPC32 initial support for .o file writing was implemented.</p>
1145 <p>MicroBlaze scheduling itineraries were added that model the
1146 3-stage and the 5-stage pipeline architectures. The 3-stage
1147 pipeline model can be selected with <code>-mcpu=mblaze3</code>
1148 and the 5-stage pipeline model can be selected with
1149 <code>-mcpu=mblaze5</code>.</p>
1159 <!--=========================================================================-->
1161 <a name="changes">Major Changes and Removed Features</a>
1166 <p>If you're already an LLVM user or developer with out-of-tree changes based on
1167 LLVM 2.9, this section lists some "gotchas" that you may run into upgrading
1168 from the previous release.</p>
1171 <li>The <code>LLVMC</code> meta compiler driver was removed.</li>
1172 <li>The <code>LowerSetJmp</code> pass wasn't used effectively by any
1173 target and has been removed.</li>
1174 <li>The old <code>TailDup</code> pass was not used in the standard pipeline
1175 and was unable to update ssa form, so it has been removed.
1176 <li>The syntax of volatile loads and stores in IR has been changed to
1177 "<code>load volatile</code>"/"<code>store volatile</code>". The old
1178 syntax ("<code>volatile load</code>"/"<code>volatile store</code>")
1179 is still accepted, but is now considered deprecated.</li>
1180 <li>The old atomic intrinsics (<code>llvm.memory.barrier</code> and
1181 <code>llvm.atomic.*</code>) are now gone. Please use the new atomic
1182 instructions, described in the <a href="Atomics.html">atomics guide</a>.
1185 <h4>Windows (32-bit)</h4>
1189 <li>On Win32(MinGW32 and MSVC), Windows 2000 will not be supported.
1190 Windows XP or higher is required.</li>
1197 <!--=========================================================================-->
1199 <a name="api_changes">Internal API Changes</a>
1204 <p>In addition, many APIs have changed in this release. Some of the major
1205 LLVM API changes are:</p>
1208 <li>The biggest and most pervasive change is that llvm::Types are no longer
1209 returned or accepted as 'const' values. Instead, just pass around
1210 non-const Types.</li>
1212 <li><code>PHINode::reserveOperandSpace</code> has been removed. Instead, you
1213 must specify how many operands to reserve space for when you create the
1214 PHINode, by passing an extra argument
1215 into <code>PHINode::Create</code>.</li>
1217 <li>PHINodes no longer store their incoming BasicBlocks as operands. Instead,
1218 the list of incoming BasicBlocks is stored separately, and can be accessed
1219 with new functions <code>PHINode::block_begin</code>
1220 and <code>PHINode::block_end</code>.</li>
1222 <li>Various functions now take an <code>ArrayRef</code> instead of either a
1223 pair of pointers (or iterators) to the beginning and end of a range, or a
1224 pointer and a length. Others now return an <code>ArrayRef</code> instead
1225 of a reference to a <code>SmallVector</code>
1226 or <code>std::vector</code>. These include:
1228 <!-- Please keep this list sorted. -->
1229 <li><code>CallInst::Create</code></li>
1230 <li><code>ComputeLinearIndex</code> (in <code>llvm/CodeGen/Analysis.h</code>)</li>
1231 <li><code>ConstantArray::get</code></li>
1232 <li><code>ConstantExpr::getExtractElement</code></li>
1233 <li><code>ConstantExpr::getGetElementPtr</code></li>
1234 <li><code>ConstantExpr::getInBoundsGetElementPtr</code></li>
1235 <li><code>ConstantExpr::getIndices</code></li>
1236 <li><code>ConstantExpr::getInsertElement</code></li>
1237 <li><code>ConstantExpr::getWithOperands</code></li>
1238 <li><code>ConstantFoldCall</code> (in <code>llvm/Analysis/ConstantFolding.h</code>)</li>
1239 <li><code>ConstantFoldInstOperands</code> (in <code>llvm/Analysis/ConstantFolding.h</code>)</li>
1240 <li><code>ConstantVector::get</code></li>
1241 <li><code>DIBuilder::createComplexVariable</code></li>
1242 <li><code>DIBuilder::getOrCreateArray</code></li>
1243 <li><code>ExtractValueInst::Create</code></li>
1244 <li><code>ExtractValueInst::getIndexedType</code></li>
1245 <li><code>ExtractValueInst::getIndices</code></li>
1246 <li><code>FindInsertedValue</code> (in <code>llvm/Analysis/ValueTracking.h</code>)</li>
1247 <li><code>gep_type_begin</code> (in <code>llvm/Support/GetElementPtrTypeIterator.h</code>)</li>
1248 <li><code>gep_type_end</code> (in <code>llvm/Support/GetElementPtrTypeIterator.h</code>)</li>
1249 <li><code>GetElementPtrInst::Create</code></li>
1250 <li><code>GetElementPtrInst::CreateInBounds</code></li>
1251 <li><code>GetElementPtrInst::getIndexedType</code></li>
1252 <li><code>InsertValueInst::Create</code></li>
1253 <li><code>InsertValueInst::getIndices</code></li>
1254 <li><code>InvokeInst::Create</code></li>
1255 <li><code>IRBuilder::CreateCall</code></li>
1256 <li><code>IRBuilder::CreateExtractValue</code></li>
1257 <li><code>IRBuilder::CreateGEP</code></li>
1258 <li><code>IRBuilder::CreateInBoundsGEP</code></li>
1259 <li><code>IRBuilder::CreateInsertValue</code></li>
1260 <li><code>IRBuilder::CreateInvoke</code></li>
1261 <li><code>MDNode::get</code></li>
1262 <li><code>MDNode::getIfExists</code></li>
1263 <li><code>MDNode::getTemporary</code></li>
1264 <li><code>MDNode::getWhenValsUnresolved</code></li>
1265 <li><code>SimplifyGEPInst</code> (in <code>llvm/Analysis/InstructionSimplify.h</code>)</li>
1266 <li><code>TargetData::getIndexedOffset</code></li>
1269 <li>All forms of <code>StringMap::getOrCreateValue</code> have been remove
1270 except for the one which takes a <code>StringRef</code>.</li>
1272 <li>The <code>LLVMBuildUnwind</code> function from the C API was removed. The
1273 LLVM <code>unwind</code> instruction has been deprecated for a long time
1274 and isn't used by the current front-ends. So this was removed during the
1275 exception handling rewrite.</li>
1277 <li>The <code>LLVMAddLowerSetJmpPass</code> function from the C API was
1278 removed because the <code>LowerSetJmp</code> pass was removed.</li>
1280 <li>The <code>DIBuilder</code> interface used by front ends to encode
1281 debugging information in the LLVM IR now expects clients to
1282 use <code>DIBuilder::finalize()</code> at the end of translation unit to
1283 complete debugging information encoding.</li>
1285 <li>The way the type system works has been
1286 rewritten: <code>PATypeHolder</code> and <code>OpaqueType</code> are gone,
1287 and all APIs deal with <code>Type*</code> instead of <code>const
1288 Type*</code>. If you need to create recursive structures, then create a
1289 named structure, and use <code>setBody()</code> when all its elements are
1290 built. Type merging and refining is gone too: named structures are not
1291 merged with other structures, even if their layout is identical. (of
1292 course anonymous structures are still uniqued by layout).</li>
1294 <li>TargetSelect.h moved to Support/ from Target/</li>
1296 <li>UpgradeIntrinsicCall no longer upgrades pre-2.9 intrinsic calls (for
1297 example <code>llvm.memset.i32</code>).</li>
1299 <li>It is mandatory to initialize all out-of-tree passes too and their dependencies now with
1300 <code>INITIALIZE_PASS{BEGIN,END,}</code>
1301 and <code>INITIALIZE_{PASS,AG}_DEPENDENCY</code>.</li>
1303 <li>The interface for MemDepResult in MemoryDependenceAnalysis has been
1304 enhanced with new return types Unknown and NonFuncLocal, in addition to
1305 the existing types Clobber, Def, and NonLocal.</li>
1312 <!-- *********************************************************************** -->
1314 <a name="knownproblems">Known Problems</a>
1316 <!-- *********************************************************************** -->
1320 <p>This section contains significant known problems with the LLVM system, listed
1321 by component. If you run into a problem, please check
1322 the <a href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
1323 there isn't already one.</p>
1325 <!-- ======================================================================= -->
1327 <a name="experimental">Experimental features included with this release</a>
1332 <p>The following components of this LLVM release are either untested, known to
1333 be broken or unreliable, or are in early development. These components
1334 should not be relied on, and bugs should not be filed against them, but they
1335 may be useful to some people. In particular, if you would like to work on
1336 one of these components, please contact us on
1337 the <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev
1341 <li>The Alpha, Blackfin, CellSPU, MicroBlaze, MSP430, MIPS, PTX, SystemZ and
1342 XCore backends are experimental.</li>
1344 <li><tt>llc</tt> "<tt>-filetype=obj</tt>" is experimental on all targets other
1345 than darwin and ELF X86 systems.</li>
1350 <!-- ======================================================================= -->
1352 <a name="x86-be">Known problems with the X86 back-end</a>
1358 <li>The X86-64 backend <a href="http://llvm.org/PR1740">does not yet support
1359 the <tt>va_arg</tt> LLVM IR instruction</a>. Currently, front-ends support
1360 variadic argument constructs on X86-64 by lowering them manually.</li>
1365 <!-- ======================================================================= -->
1367 <a name="ppc-be">Known problems with the PowerPC back-end</a>
1373 <li>The PPC32/ELF support lacks PIC support.</li>
1378 <!-- ======================================================================= -->
1380 <a name="arm-be">Known problems with the ARM back-end</a>
1386 <li>Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6
1387 processors, thumb programs can crash or produce wrong results
1388 (<a href="http://llvm.org/PR1388">PR1388</a>).</li>
1390 <li>Compilation for ARM Linux OABI (old ABI) is supported but not fully
1396 <!-- ======================================================================= -->
1398 <a name="sparc-be">Known problems with the SPARC back-end</a>
1404 <li>The SPARC backend only supports the 32-bit SPARC ABI (-m32); it does not
1405 support the 64-bit SPARC ABI (-m64).</li>
1410 <!-- ======================================================================= -->
1412 <a name="mips-be">Known problems with the MIPS back-end</a>
1418 <li>64-bit MIPS targets are not supported yet.</li>
1423 <!-- ======================================================================= -->
1425 <a name="alpha-be">Known problems with the Alpha back-end</a>
1431 <li>On 21164s, some rare FP arithmetic sequences which may trap do not have
1432 the appropriate nops inserted to ensure restartability.</li>
1437 <!-- ======================================================================= -->
1439 <a name="c-be">Known problems with the C back-end</a>
1444 <p>The C backend has numerous problems and is not being actively maintained.
1445 Depending on it for anything serious is not advised.</p>
1448 <li><a href="http://llvm.org/PR802">The C backend has only basic support for
1449 inline assembly code</a>.</li>
1451 <li><a href="http://llvm.org/PR1658">The C backend violates the ABI of common
1452 C++ programs</a>, preventing intermixing between C++ compiled by the CBE
1453 and C++ code compiled with <tt>llc</tt> or native compilers.</li>
1455 <li>The C backend does not support all exception handling constructs.</li>
1457 <li>The C backend does not support arbitrary precision integers.</li>
1464 <!-- *********************************************************************** -->
1466 <a name="additionalinfo">Additional Information</a>
1468 <!-- *********************************************************************** -->
1472 <p>A wide variety of additional information is available on
1473 the <a href="http://llvm.org/">LLVM web page</a>, in particular in
1474 the <a href="http://llvm.org/docs/">documentation</a> section. The web page
1475 also contains versions of the API documentation which is up-to-date with the
1476 Subversion version of the source code. You can access versions of these
1477 documents specific to this release by going into the "<tt>llvm/doc/</tt>"
1478 directory in the LLVM tree.</p>
1480 <p>If you have any questions or comments about LLVM, please feel free to contact
1481 us via the <a href="http://llvm.org/docs/#maillist"> mailing lists</a>.</p>
1485 <!-- *********************************************************************** -->
1489 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1490 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1491 <a href="http://validator.w3.org/check/referer"><img
1492 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1494 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
1495 Last modified: $Date$