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>
14 <img style="float:right" src="http://llvm.org/img/DragonSmall.png"
15 width="136" height="136" alt="LLVM Dragon Logo">
19 <li><a href="#intro">Introduction</a></li>
20 <li><a href="#subproj">Sub-project Status Update</a></li>
21 <li><a href="#externalproj">External Projects Using LLVM 3.0</a></li>
22 <li><a href="#whatsnew">What's New in LLVM 3.0?</a></li>
23 <li><a href="GettingStarted.html">Installation Instructions</a></li>
24 <li><a href="#knownproblems">Known Problems</a></li>
25 <li><a href="#additionalinfo">Additional Information</a></li>
28 <div class="doc_author">
29 <p>Written by the <a href="http://llvm.org/">LLVM Team</a></p>
33 <h1 style="color:red">These are in-progress notes for the upcoming LLVM 3.0
36 <a href="http://llvm.org/releases/2.9/docs/ReleaseNotes.html">LLVM 2.9
37 Release Notes</a>.</h1>
40 <!-- *********************************************************************** -->
42 <a name="intro">Introduction</a>
44 <!-- *********************************************************************** -->
48 <p>This document contains the release notes for the LLVM Compiler
49 Infrastructure, release 3.0. Here we describe the status of LLVM, including
50 major improvements from the previous release, improvements in various
51 subprojects of LLVM, and some of the current users of the code.
52 All LLVM releases may be downloaded from
53 the <a href="http://llvm.org/releases/">LLVM releases web site</a>.</p>
55 <p>For more information about LLVM, including information about the latest
56 release, please check out the <a href="http://llvm.org/">main LLVM web
57 site</a>. If you have questions or comments,
58 the <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM
59 Developer's Mailing List</a> is a good place to send them.</p>
61 <p>Note that if you are reading this file from a Subversion checkout or the main
62 LLVM web page, this document applies to the <i>next</i> release, not the
63 current one. To see the release notes for a specific release, please see the
64 <a href="http://llvm.org/releases/">releases page</a>.</p>
69 <!-- *********************************************************************** -->
71 <a name="subproj">Sub-project Status Update</a>
73 <!-- *********************************************************************** -->
77 <p>The LLVM 3.0 distribution currently consists of code from the core LLVM
78 repository (which roughly includes the LLVM optimizers, code generators and
79 supporting tools), and the Clang repository. In
80 addition to this code, the LLVM Project includes other sub-projects that are
81 in development. Here we include updates on these subprojects.</p>
83 <!--=========================================================================-->
85 <a name="clang">Clang: C/C++/Objective-C Frontend Toolkit</a>
90 <p><a href="http://clang.llvm.org/">Clang</a> is an LLVM front end for the C,
91 C++, and Objective-C languages. Clang aims to provide a better user
92 experience through expressive diagnostics, a high level of conformance to
93 language standards, fast compilation, and low memory use. Like LLVM, Clang
94 provides a modular, library-based architecture that makes it suitable for
95 creating or integrating with other development tools. Clang is considered a
96 production-quality compiler for C, Objective-C, C++ and Objective-C++ on x86
97 (32- and 64-bit), and for Darwin/ARM targets.</p>
99 <p>In the LLVM 3.0 time-frame, the Clang team has made many improvements:</p>
101 <li>Greatly improved support for building C++ applications, with greater
102 stability and better diagnostics.</li>
104 <li><a href="http://clang.llvm.org/cxx_status.html">Improved support</a> for
105 the <a href="http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_detail.htm?csnumber=50372">C++
106 2011</a> standard (aka "C++'0x"), including implementations of non-static data member
107 initializers, alias templates, delegating constructors, range-based
108 for loops, and implicitly-generated move constructors and move assignment
109 operators, among others.</li>
111 <li>Implemented support for some features of the upcoming C1x standard,
112 including static assertions and generic selections.</li>
114 <li>Better detection of include and linking paths for system headers and
115 libraries, especially for Linux distributions.</li>
117 <li>Several improvements to Objective-C support, including:
120 <li><a href="http://clang.llvm.org/docs/AutomaticReferenceCounting.html">
121 Automatic Reference Counting</a> (ARC) and an improved memory model
122 cleanly separating object and C memory.</li>
124 <li>A migration tool for moving manual retain/release code to ARC</li>
126 <li>Better support for data hiding, allowing instance variables to be
127 declared in implementation contexts or class extensions</li>
128 <li>Weak linking support for Objective-C classes</li>
129 <li>Improved static type checking by inferring the return type of methods
130 such as +alloc and -init.</li>
133 Some new Objective-C features require either the Mac OS X 10.7 / iOS 5
134 Objective-C runtime, or version 1.6 or later of the GNUstep Objective-C
135 runtime version.</li>
137 <li>Implemented a number of optimizations in <tt>libclang</tt>, the Clang C
138 interface, to improve the performance of code completion and the mapping
139 from source locations to abstract syntax tree nodes.</li>
141 <p>For more details about the changes to Clang since the 2.9 release, see the
142 <a href="http://clang.llvm.org/docs/ReleaseNotes.html">Clang release notes</a>
146 <p>If Clang rejects your code but another compiler accepts it, please take a
147 look at the <a href="http://clang.llvm.org/compatibility.html">language
148 compatibility</a> guide to make sure this is not intentional or a known
153 <!--=========================================================================-->
155 <a name="dragonegg">DragonEgg: GCC front-ends, LLVM back-end</a>
159 <p><a href="http://dragonegg.llvm.org/">DragonEgg</a> is a
160 <a href="http://gcc.gnu.org/wiki/plugins">gcc plugin</a> that replaces GCC's
161 optimizers and code generators with LLVM's. It works with gcc-4.5 or gcc-4.6,
162 targets the x86-32 and x86-64 processor families, and has been successfully
163 used on the Darwin, FreeBSD, KFreeBSD, Linux and OpenBSD platforms. It fully
164 supports Ada, C, C++ and Fortran. It has partial support for Go, Java, Obj-C
167 <p>The 3.0 release has the following notable changes:</p>
170 <li>GCC version 4.6 is now fully supported.</li>
172 <li>Patching and building GCC is no longer required: the plugin should work
173 with your system GCC (version 4.5 or 4.6; on Debian/Ubuntu systems the
174 gcc-4.5-plugin-dev or gcc-4.6-plugin-dev package is also needed).</li>
176 <li>The <tt>-fplugin-arg-dragonegg-enable-gcc-optzns</tt> option, which runs
177 GCC's optimizers as well as LLVM's, now works much better. This is the
178 option to use if you want ultimate performance! It is still experimental
179 though: it may cause the plugin to crash. Setting the optimization level
180 to <tt>-O4</tt> when using this option will optimize even harder, though
181 this usually doesn't result in any improvement over <tt>-O3</tt>.</li>
183 <li>The type and constant conversion logic has been almost entirely rewritten,
184 fixing a multitude of obscure bugs.</li>
190 <!--=========================================================================-->
192 <a name="compiler-rt">compiler-rt: Compiler Runtime Library</a>
197 <p>The new LLVM <a href="http://compiler-rt.llvm.org/">compiler-rt project</a>
198 is a simple library that provides an implementation of the low-level
199 target-specific hooks required by code generation and other runtime
200 components. For example, when compiling for a 32-bit target, converting a
201 double to a 64-bit unsigned integer is compiled into a runtime call to the
202 "__fixunsdfdi" function. The compiler-rt library provides highly optimized
203 implementations of this and other low-level routines (some are 3x faster than
204 the equivalent libgcc routines).</p>
206 <p>In the LLVM 3.0 timeframe, the target specific ARM code has converted to
207 "unified" assembly syntax, and several new functions have been added to the
212 <!--=========================================================================-->
214 <a name="lldb">LLDB: Low Level Debugger</a>
219 <p>LLDB is a ground-up implementation of a command line debugger, as well as a
220 debugger API that can be used from other applications. LLDB makes use of the
221 Clang parser to provide high-fidelity expression parsing (particularly for
222 C++) and uses the LLVM JIT for target support.</p>
224 <p>LLDB has advanced by leaps and bounds in the 3.0 timeframe. It is
225 dramatically more stable and useful, and includes both a
226 new <a href="http://lldb.llvm.org/tutorial.html">tutorial</a> and
227 a <a href="http://lldb.llvm.org/lldb-gdb.html">side-by-side comparison with
232 <!--=========================================================================-->
234 <a name="libc++">libc++: C++ Standard Library</a>
239 <p>Like compiler_rt, libc++ is now <a href="DeveloperPolicy.html#license">dual
240 licensed</a> under the MIT and UIUC license, allowing it to be used more
243 <p>Libc++ has been ported to FreeBSD and imported into the base system. It is
244 planned to be the default STL implementation for FreeBSD 10.</p>
248 <!--=========================================================================-->
250 <a name="vmkit">VMKit</a>
255 <p>The <a href="http://vmkit.llvm.org/">VMKit project</a> is an
256 implementation of a Java Virtual Machine (Java VM or JVM) that uses LLVM for
257 static and just-in-time compilation.
259 <p>In the LLVM 3.0 time-frame, VMKit has had significant improvements on both
260 runtime and startup performance:</p>
263 <li>Precompilation: by compiling ahead of time a small subset of Java's core
264 library, the startup performance have been highly optimized to the point that
265 running a 'Hello World' program takes less than 30 milliseconds.</li>
267 <li>Customization: by customizing virtual methods for individual classes,
268 the VM can statically determine the target of a virtual call, and decide to
271 <li>Inlining: the VM does more inlining than it did before, by allowing more
272 bytecode instructions to be inlined, and thanks to customization. It also
273 inlines GC barriers, and object allocations.</li>
275 <li>New exception model: the generated code for a method that does not do
276 any try/catch is not penalized anymore by the eventuality of calling a
277 method that throws an exception. Instead, the method that throws the
278 exception jumps directly to the method that could catch it.</li>
284 <!--=========================================================================-->
286 <a name="LLBrowse">LLBrowse: IR Browser</a>
291 <p><a href="http://llvm.org/svn/llvm-project/llbrowse/trunk/doc/LLBrowse.html">
292 LLBrowse</a> is an interactive viewer for LLVM modules. It can load any LLVM
293 module and displays its contents as an expandable tree view, facilitating an
294 easy way to inspect types, functions, global variables, or metadata nodes. It
295 is fully cross-platform, being based on the popular wxWidgets GUI
301 <!--=========================================================================-->
304 <a name="klee">KLEE: A Symbolic Execution Virtual Machine</a>
309 <a href="http://klee.llvm.org/">KLEE</a> is a symbolic execution framework for
310 programs in LLVM bitcode form. KLEE tries to symbolically evaluate "all" paths
311 through the application and records state transitions that lead to fault
312 states. This allows it to construct testcases that lead to faults and can even
313 be used to verify some algorithms.
321 <!-- *********************************************************************** -->
323 <a name="externalproj">External Open Source Projects Using LLVM 3.0</a>
325 <!-- *********************************************************************** -->
329 <p>An exciting aspect of LLVM is that it is used as an enabling technology for
330 a lot of other language and tools projects. This section lists some of the
331 projects that have already been updated to work with LLVM 3.0.</p>
333 <!--=========================================================================-->
334 <h3>AddressSanitizer</h3>
338 <p><a href="http://code.google.com/p/address-sanitizer/">AddressSanitizer</a>
339 uses compiler instrumentation and a specialized malloc library to find C/C++
340 bugs such as use-after-free and out-of-bound accesses to heap, stack, and
341 globals. The key feature of the tool is speed: the average slowdown
342 introduced by AddressSanitizer is less than 2x.</p>
346 <!--=========================================================================-->
351 <p><a href="http://www.clamav.net">Clam AntiVirus</a> is an open source (GPL)
352 anti-virus toolkit for UNIX, designed especially for e-mail scanning on mail
355 <p>Since version 0.96 it
356 has <a href="http://vrt-sourcefire.blogspot.com/2010/09/introduction-to-clamavs-low-level.html">bytecode
357 signatures</a> that allow writing detections for complex malware.
358 It uses LLVM's JIT to speed up the execution of bytecode on X86, X86-64,
359 PPC32/64, falling back to its own interpreter otherwise. The git version was
360 updated to work with LLVM 3.0.</p>
364 <!--=========================================================================-->
365 <h3>clang_complete for VIM</h3>
369 <p><a href="https://github.com/Rip-Rip/clang_complete">clang_complete</a> is a
370 VIM plugin, that provides accurate C/C++ autocompletion using the clang front
371 end. The development version of clang complete, can directly use libclang
372 which can maintain a cache to speed up auto completion.</p>
376 <!--=========================================================================-->
381 <p><a href="https://bitbucket.org/dwilliamson/clreflect">clReflect</a> is a C++
382 parser that uses clang/LLVM to derive a light-weight reflection database
383 suitable for use in game development. It comes with a very simple runtime
384 library for loading and querying the database, requiring no external
385 dependencies (including CRT), and an additional utility library for object
386 management and serialisation.</p>
390 <!--=========================================================================-->
391 <h3>Cling C++ Interpreter</h3>
395 <p><a href="http://cern.ch/cling">Cling</a> is an interactive compiler interface
396 (aka C++ interpreter). It supports C++ and C, and uses LLVM's JIT and the
397 Clang parser. It has a prompt interface, runs source files, calls into shared
398 libraries, prints the value of expressions, even does runtime lookup of
399 identifiers (dynamic scopes). And it just behaves like one would expect from
404 <!--=========================================================================-->
405 <h3>Crack Programming Language</h3>
409 <p><a href="http://code.google.com/p/crack-language/">Crack</a> aims to provide
410 the ease of development of a scripting language with the performance of a
411 compiled language. The language derives concepts from C++, Java and Python,
412 incorporating object-oriented programming, operator overloading and strong
417 <!--=========================================================================-->
422 <p><a href="http://eerolanguage.org/">Eero</a> is a fully
423 header-and-binary-compatible dialect of Objective-C 2.0, implemented with a
424 patched version of the Clang/LLVM compiler. It features a streamlined syntax,
425 Python-like indentation, and new operators, for improved readability and
426 reduced code clutter. It also has new features such as limited forms of
427 operator overloading and namespaces, and strict (type-and-operator-safe)
428 enumerations. It is inspired by languages such as Smalltalk, Python, and
433 <!--=========================================================================-->
434 <h3>FAUST Real-Time Audio Signal Processing Language</h3>
438 <p><a href="http://faust.grame.fr/">FAUST</a> is a compiled language for
439 real-time audio signal processing. The name FAUST stands for Functional
440 AUdio STream. Its programming model combines two approaches: functional
441 programming and block diagram composition. In addition with the C, C++, Java
442 output formats, the Faust compiler can now generate LLVM bitcode, and works
448 <!--=========================================================================-->
449 <h3>Glasgow Haskell Compiler (GHC)</h3>
453 <p>GHC is an open source, state-of-the-art programming suite for Haskell, a
454 standard lazy functional programming language. It includes an optimizing
455 static compiler generating good code for a variety of platforms, together
456 with an interactive system for convenient, quick development.</p>
458 <p>GHC 7.0 and onwards include an LLVM code generator, supporting LLVM 2.8 and
459 later. Since LLVM 2.9, GHC now includes experimental support for the ARM
460 platform with LLVM 3.0.</p>
464 <!--=========================================================================-->
469 <p><a href="http://botwars.tk/gwscript/">gwXscript</a> is an object oriented,
470 aspect oriented programming language which can create both executables (ELF,
471 EXE) and shared libraries (DLL, SO, DYNLIB). The compiler is implemented in
472 its own language and translates scripts into LLVM-IR which can be optimized
473 and translated into native code by the LLVM framework. Source code in
474 gwScript contains definitions that expand the namespaces. So you can build
475 your project and simply 'plug out' features by removing a file. The remaining
476 project does not leave scars since you directly separate concerns by the
477 'template' feature of gwX. It is also possible to add new features to a
478 project by just adding files and without editing the original project. This
479 language is used for example to create games or content management systems
480 that should be extendable.</p>
482 <p>gwXscript is strongly typed and offers comfort with its native types string,
483 hash and array. You can easily write new libraries in gwXscript or native
484 code. gwXscript is type safe and users should not be able to crash your
485 program or execute malicious code except code that is eating CPU time.</p>
489 <!--=========================================================================-->
490 <h3>include-what-you-use</h3>
494 <p><a href="http://code.google.com/p/include-what-you-use">include-what-you-use</a>
495 is a tool to ensure that a file directly <code>#include</code>s
496 all <code>.h</code> files that provide a symbol that the file uses. It also
497 removes superfluous <code>#include</code>s from source files.</p>
501 <!--=========================================================================-->
502 <h3>ispc: The Intel SPMD Program Compiler</h3>
506 <p><a href="http://ispc.github.com">ispc</a> is a compiler for "single program,
507 multiple data" (SPMD) programs. It compiles a C-based SPMD programming
508 language to run on the SIMD units of CPUs; it often delivers 5-6x speedups on
509 a single core of a CPU with an 8-wide SIMD unit compared to serial code,
510 while still providing a clean and easy-to-understand programming model. For
511 an introduction to the language and its performance,
512 see <a href="http://ispc.github.com/example.html">the walkthrough</a> of a short
513 example program. ispc is licensed under the BSD license.</p>
517 <!--=========================================================================-->
518 <h3>The Julia Programming Language</h3>
522 <p><a href="http://github.com/JuliaLang/julia">Julia</a> is a high-level,
523 high-performance dynamic language for technical
524 computing. It provides a sophisticated compiler, distributed parallel
525 execution, numerical accuracy, and an extensive mathematical function
526 library. The compiler uses type inference to generate fast code
527 without any type declarations, and uses LLVM's optimization passes and
528 JIT compiler. The language is designed around multiple dispatch,
529 giving programs a large degree of flexibility. It is ready for use on many
530 kinds of problems.</p>
533 <!--=========================================================================-->
534 <h3>LanguageKit and Pragmatic Smalltalk</h3>
538 <p><a href="http://etoileos.com/etoile/features/languagekit/">LanguageKit</a> is
539 a framework for implementing dynamic languages sharing an object model with
540 Objective-C. It provides static and JIT compilation using LLVM along with
541 its own interpreter. Pragmatic Smalltalk is a dialect of Smalltalk, built on
542 top of LanguageKit, that interfaces directly with Objective-C, sharing the
543 same object representation and message sending behaviour. These projects are
544 developed as part of the Étoilé desktop environment.</p>
548 <!--=========================================================================-->
553 <p><a href="http://lua-av.mat.ucsb.edu/blog/">LuaAV</a> is a real-time
554 audiovisual scripting environment based around the Lua language and a
555 collection of libraries for sound, graphics, and other media protocols. LuaAV
556 uses LLVM and Clang to JIT compile efficient user-defined audio synthesis
557 routines specified in a declarative syntax.</p>
561 <!--=========================================================================-->
566 <p>An open source, cross-platform implementation of C# and the CLR that is
567 binary compatible with Microsoft.NET. Has an optional, dynamically-loaded
568 LLVM code generation backend in Mini, the JIT compiler.</p>
570 <p>Note that we use a Git mirror of LLVM <a
571 href="https://github.com/mono/llvm">with some patches</a>.</p>
575 <!--=========================================================================-->
580 <p><a href="http://polly.grosser.es">Polly</a> is an advanced data-locality
581 optimizer and automatic parallelizer. It uses an advanced, mathematical
582 model to calculate detailed data dependency information which it uses to
583 optimize the loop structure of a program. Polly can speed up sequential code
584 by improving memory locality and consequently the cache use. Furthermore,
585 Polly is able to expose different kind of parallelism which it exploits by
586 introducing (basic) OpenMP and SIMD code. A mid-term goal of Polly is to
587 automatically create optimized GPU code.</p>
591 <!--=========================================================================-->
592 <h3>Portable OpenCL (pocl)</h3>
596 <p>Portable OpenCL is an open source implementation of the OpenCL standard which
597 can be easily adapted for new targets. One of the goals of the project is
598 improving performance portability of OpenCL programs, avoiding the need for
599 target-dependent manual optimizations. A "native" target is included, which
600 allows running OpenCL kernels on the host (CPU).</p>
604 <!--=========================================================================-->
608 <p><a href="http://pure-lang.googlecode.com/">Pure</a> is an
609 algebraic/functional programming language based on term rewriting. Programs
610 are collections of equations which are used to evaluate expressions in a
611 symbolic fashion. The interpreter uses LLVM as a backend to JIT-compile Pure
612 programs to fast native code. Pure offers dynamic typing, eager and lazy
613 evaluation, lexical closures, a hygienic macro system (also based on term
614 rewriting), built-in list and matrix support (including list and matrix
615 comprehensions) and an easy-to-use interface to C and other programming
616 languages (including the ability to load LLVM bitcode modules, and inline C,
617 C++, Fortran and Faust code in Pure programs if the corresponding LLVM-enabled
618 compilers are installed).</p>
620 <p>Pure version 0.48 has been tested and is known to work with LLVM 3.0
621 (and continues to work with older LLVM releases >= 2.5).</p>
625 <!--=========================================================================-->
626 <h3>Renderscript</h3>
630 <p><a href="http://developer.android.com/guide/topics/renderscript/index.html">Renderscript</a>
631 is Android's advanced 3D graphics rendering and compute API. It provides a
632 portable C99-based language with extensions to facilitate common use cases
633 for enhancing graphics and thread level parallelism. The Renderscript
634 compiler frontend is based on Clang/LLVM. It emits a portable bitcode format
635 for the actual compiled script code, as well as reflects a Java interface for
636 developers to control the execution of the compiled bitcode. Executable
637 machine code is then generated from this bitcode by an LLVM backend on the
638 device. Renderscript is thus able to provide a mechanism by which Android
639 developers can improve performance of their applications while retaining
644 <!--=========================================================================-->
649 <p><a href="http://safecode.cs.illinois.edu">SAFECode</a> is a memory safe C/C++
650 compiler built using LLVM. It takes standard, unannotated C/C++ code,
651 analyzes the code to ensure that memory accesses and array indexing
652 operations are safe, and instruments the code with run-time checks when
653 safety cannot be proven statically. SAFECode can be used as a debugging aid
654 (like Valgrind) to find and repair memory safety bugs. It can also be used
655 to protect code from security attacks at run-time.</p>
659 <!--=========================================================================-->
660 <h3>The Stupid D Compiler (SDC)</h3>
664 <p><a href="https://github.com/bhelyer/SDC">The Stupid D Compiler</a> is a
665 project seeking to write a self-hosting compiler for the D programming
666 language without using the frontend of the reference compiler (DMD).</p>
670 <!--=========================================================================-->
671 <h3>TTA-based Co-design Environment (TCE)</h3>
675 <p>TCE is a toolset for designing application-specific processors (ASP) based on
676 the Transport triggered architecture (TTA). The toolset provides a complete
677 co-design flow from C/C++ programs down to synthesizable VHDL and parallel
678 program binaries. Processor customization points include the register files,
679 function units, supported operations, and the interconnection network.</p>
681 <p>TCE uses Clang and LLVM for C/C++ language support, target independent
682 optimizations and also for parts of code generation. It generates new
683 LLVM-based code generators "on the fly" for the designed TTA processors and
684 loads them in to the compiler backend as runtime libraries to avoid
685 per-target recompilation of larger parts of the compiler chain.</p>
689 <!--=========================================================================-->
690 <h3>Tart Programming Language</h3>
694 <p><a href="http://code.google.com/p/tart/">Tart</a> is a general-purpose,
695 strongly typed programming language designed for application
696 developers. Strongly inspired by Python and C#, Tart focuses on practical
697 solutions for the professional software developer, while avoiding the clutter
698 and boilerplate of legacy languages like Java and C++. Although Tart is still
699 in development, the current implementation supports many features expected of
700 a modern programming language, such as garbage collection, powerful
701 bidirectional type inference, a greatly simplified syntax for template
702 metaprogramming, closures and function literals, reflection, operator
703 overloading, explicit mutability and immutability, and much more. Tart is
704 flexible enough to accommodate a broad range of programming styles and
705 philosophies, while maintaining a strong commitment to simplicity, minimalism
706 and elegance in design.</p>
710 <!--=========================================================================-->
711 <h3>ThreadSanitizer</h3>
715 <p><a href="http://code.google.com/p/data-race-test/">ThreadSanitizer</a> is a
716 data race detector for (mostly) C and C++ code, available for Linux, Mac OS
717 and Windows. On different systems, we use binary instrumentation frameworks
718 (Valgrind and Pin) as frontends that generate the program events for the race
719 detection algorithm. On Linux, there's an option of using LLVM-based
720 compile-time instrumentation.</p>
726 <!-- *********************************************************************** -->
728 <a name="whatsnew">What's New in LLVM 3.0?</a>
730 <!-- *********************************************************************** -->
734 <p>This release includes a huge number of bug fixes, performance tweaks and
735 minor improvements. Some of the major improvements and new features are
736 listed in this section.</p>
738 <!--=========================================================================-->
740 <a name="majorfeatures">Major New Features</a>
745 <!-- Features that need text if they're finished for 3.1:
749 loop dependence analysis
750 CorrelatedValuePropagation
751 lib/Transforms/IPO/MergeFunctions.cpp => consider for 3.1.
752 Integrated assembler on by default for arm/thumb?
757 Analysis/RegionInfo.h + Dom Frontiers
758 SparseBitVector: used in LiveVar.
759 llvm/lib/Archive - replace with lib object?
762 <p>LLVM 3.0 includes several major changes and big features:</p>
765 <li>llvm-gcc is no longer supported, and not included in the release. We
766 recommend switching to <a
767 href="http://clang.llvm.org/">Clang</a> or <a
768 href="http://dragonegg.llvm.org/">DragonEgg</a>.</li>
770 <li>The linear scan register allocator has been replaced with a new "greedy"
771 register allocator, enabling live range splitting and many other
772 optimizations that lead to better code quality. Please see its <a
773 href="http://blog.llvm.org/2011/09/greedy-register-allocation-in-llvm-30.html">blog post</a> or its talk at the <a
774 href="http://llvm.org/devmtg/2011-11/">Developer Meeting</a>
775 for more information.</li>
776 <li>LLVM IR now includes full support for <a href="Atomics.html">atomics
777 memory operations</a> intended to support the C++'11 and C'1x memory models.
778 This includes <a href="LangRef.html#memoryops">atomic load and store,
779 compare and exchange, and read/modify/write instructions</a> as well as a
780 full set of <a href="LangRef.html#ordering">memory ordering constraints</a>.
781 Please see the <a href="Atomics.html">Atomics Guide</a> for more
784 <li>The LLVM IR exception handling representation has been redesigned and
785 reimplemented, making it more elegant, fixing a huge number of bugs, and
786 enabling inlining and other optimizations. Please see its <a href=
787 "http://blog.llvm.org/2011/11/llvm-30-exception-handling-redesign.html">blog
788 post</a> and the <a href="ExceptionHandling.html">Exception Handling
789 documentation</a> for more information.</li>
790 <li>The LLVM IR Type system has been redesigned and reimplemented, making it
791 faster and solving some long-standing problems.
793 href="http://blog.llvm.org/2011/11/llvm-30-type-system-rewrite.html">blog
794 post</a> for more information.</li>
796 <li>The MIPS backend has made major leaps in this release, going from an
797 experimental target to being virtually production quality and supporting a
798 wide variety of MIPS subtargets. See the <a href="#MIPS">MIPS section</a>
799 below for more information.</li>
801 <li>The optimizer and code generator now supports gprof and gcov-style coverage
802 and profiling information, and includes a new llvm-cov tool (but also works
803 with gcov). Clang exposes coverage and profiling through GCC-compatible
804 command line options.</li>
810 <!--=========================================================================-->
812 <a name="coreimprovements">LLVM IR and Core Improvements</a>
817 <p>LLVM IR has several new features for better support of new targets and that
818 expose new optimization opportunities:</p>
821 <li><a href="Atomics.html">Atomic memory accesses and memory ordering</a> are
822 now directly expressible in the IR.</li>
823 <li>A new <a href="LangRef.html#int_fma">llvm.fma intrinsic</a> directly
824 represents floating point multiply accumulate operations without an
825 intermediate rounding stage.</li>
826 <li>A new <a href="LangRef.html#int_expect">llvm.expect intrinsic</a> allows a
827 frontend to express expected control flow (and the
828 <a href="BranchWeightMetadata.html#builtin_expect">
829 <code>__builtin_expect</code></a> from GNU C).</li>
830 <li>The <a href="LangRef.html#int_prefetch">llvm.prefetch intrinsic</a> now
831 takes a 4th argument that specifies whether the prefetch happens from the
832 icache or dcache.</li>
833 <li>The new <a href="LangRef.html#uwtable">uwtable function attribute</a>
834 allows a frontend to control emission of unwind tables.</li>
835 <li>The new <a href="LangRef.html#fnattrs">nonlazybind function
836 attribute</a> allow optimization of Global Offset Table (GOT) accesses.</li>
837 <li>The new <a href="LangRef.html#returns_twice">returns_twice attribute</a>
838 allows better modeling of functions like setjmp.</li>
839 <li>The <a href="LangRef.html#datalayout">target datalayout</a> string can now
840 encode the natural alignment of the target's stack for better optimization.
845 <!--=========================================================================-->
847 <a name="optimizer">Optimizer Improvements</a>
852 <p>In addition to many minor performance tweaks and bug fixes, this
853 release includes a few major enhancements and additions to the
857 <li>The pass manager now has an extension API that allows front-ends and plugins
858 to insert their own optimizations in the well-known places in the standard
859 pass optimization pipeline.</li>
861 <li>Information about <a href="BranchWeightMetadata.html">branch probability</a>
862 and basic block frequency is now available within LLVM, based on a
863 combination of static branch prediction heuristics and
864 <code>__builtin_expect</code> calls. That information is currently used for
865 register spill placement and if-conversion, with additional optimizations
866 planned for future releases. The same framework is intended for eventual
867 use with profile-guided optimization.</li>
869 <li>The "-indvars" induction variable simplification pass only modifies
870 induction variables when profitable. Sign and zero extension
871 elimination, linear function test replacement, loop unrolling, and
872 other simplifications that require induction variable analysis have
873 been generalized so they no longer require loops to be rewritten into
874 canonical form prior to optimization. This new design
875 preserves more IR level information, avoids undoing earlier loop
876 optimizations (particularly hand-optimized loops), and no longer
877 requires the code generator to reconstruct loops into an optimal form -
878 an intractable problem.</li>
880 <li>LLVM now includes a pass to optimize retain/release calls for the
881 <a href="http://clang.llvm.org/docs/AutomaticReferenceCounting.html">Automatic
882 Reference Counting</a> (ARC) Objective-C language feature (in
883 lib/Transforms/Scalar/ObjCARC.cpp). It is a decent example of implementing
884 a source-language-specific optimization in LLVM.</li>
890 <!--=========================================================================-->
892 <a name="mc">MC Level Improvements</a>
897 <p>The LLVM Machine Code (aka MC) subsystem was created to solve a number of
898 problems in the realm of assembly, disassembly, object file format handling,
899 and a number of other related areas that CPU instruction-set level tools work
900 in. For more information, please see
901 the <a href="http://blog.llvm.org/2010/04/intro-to-llvm-mc-project.html">Intro
902 to the LLVM MC Project Blog Post</a>.</p>
905 <li>The MC layer has undergone significant refactoring to eliminate layering
906 violations that caused it to pull in the LLVM compiler backend code.</li>
907 <li>The ELF object file writers are much more full featured.</li>
908 <li>The integrated assembler now supports #line directives.</li>
909 <li>An early implementation of a JIT built on top of the MC framework (known
910 as MC-JIT) has been implemented and will eventually replace the old JIT.
911 It emits object files direct to memory and uses a runtime dynamic linker to
912 resolve references and drive lazy compilation. The MC-JIT enables much
913 greater code reuse between the JIT and the static compiler and provides
914 better integration with the platform ABI as a result.
916 <li>The assembly printer now makes uses of assemblers instruction aliases
917 (InstAliases) to print simplified mneumonics when possible.</li>
918 <li>TableGen can now autogenerate MC expansion logic for pseudo
919 instructions that expand to multiple MC instructions (through the
920 PseudoInstExpansion class).</li>
921 <li>A new llvm-dwarfdump tool provides a start of a drop-in
922 replacement for the corresponding tool that use LLVM libraries. As part of
923 this, LLVM has the beginnings of a dwarf parsing library.</li>
924 <li>llvm-objdump has more output including, symbol by symbol disassembly,
925 inline relocations, section headers, symbol tables, and section contents.
926 Support for archive files has also been added.</li>
927 <li>llvm-nm has gained support for archives of binary files.</li>
928 <li>llvm-size has been added. This tool prints out section sizes.</li>
933 <!--=========================================================================-->
935 <a name="codegen">Target Independent Code Generator Improvements</a>
940 <p>We have put a significant amount of work into the code generator
941 infrastructure, which allows us to implement more aggressive algorithms and
942 make it run faster:</p>
945 <li>LLVM can now produce code that works with libgcc
946 to <a href="SegmentedStacks.html">dynamically allocate stack
947 segments</a>, as opposed to allocating a worst-case chunk of
948 virtual memory for each thread.</li>
949 <li>LLVM generates substantially better code for indirect gotos due to a new
950 tail duplication pass, which can be a substantial performance win for
951 interpreter loops that use them.</li>
952 <li>Exception handling and debug frame information is now emitted with CFI
953 directives. This lets the assembler produce more compact info as it knows
954 the final offsets, yielding <a href="http://blog.mozilla.com/respindola/2011/05/12/cfi-directives/">much smaller executables</a> for some C++ applications.
955 If the system assembler doesn't support it, MC exands the directives when
956 the integrated assembler is not used.
959 <li>The code generator now supports vector "select" operations on vector
960 comparisons, turning them into various optimized code sequences (e.g.
961 using the SSE4/AVX "blend" instructions).</li>
962 <li>The SSE execution domain fix pass and the ARM NEON move fix pass have been
963 merged to a target independent execution dependency fix pass. This pass is
964 used to select alternative equivalent opcodes in a way that minimizes
965 execution domain crossings. Closely connected instructions are moved to
966 the same execution domain when possible. Targets can override the
967 <code>getExecutionDomain</code> and <code>setExecutionDomain</code> hooks
968 to use the pass.</li>
972 <!--=========================================================================-->
974 <a name="x86">X86-32 and X86-64 Target Improvements</a>
979 <p>New features and major changes in the X86 target include:</p>
982 <li>The X86 backend, assembler and disassembler now have full support for AVX 1.
983 To enable it pass <code>-mavx</code> to the compiler. AVX2 implementation is
984 underway on mainline.</li>
985 <li>The integrated assembler and disassembler now support a broad range of new
986 instructions including Atom, Ivy Bridge, <a
987 href="http://en.wikipedia.org/wiki/SSE4a">SSE4a/BMI</a> instructions, <a
988 href="http://en.wikipedia.org/wiki/RdRand">rdrand</a> and many others.</li>
989 <li>The X86 backend now fully supports the <a href="http://llvm.org/PR879">X87
990 floating point stack inline assembly constraints</a>.</li>
991 <li>The integrated assembler now supports the <tt>.code32</tt> and
992 <tt>.code64</tt> directives to switch between 32-bit and 64-bit
994 <li>The X86 backend now synthesizes horizontal add/sub instructions from generic
995 vector code when the appropriate instructions are enabled.</li>
996 <li>The X86-64 backend generates smaller and faster code at -O0 due to
997 improvements in fast instruction selection.</li>
998 <li><a href="http://code.google.com/p/nativeclient/">Native Client</a>
999 subtarget support has been added.</li>
1001 <li>The CRC32 intrinsics have been renamed. The intrinsics were previously
1002 <code>@llvm.x86.sse42.crc32.[8|16|32]</code>
1003 and <code>@llvm.x86.sse42.crc64.[8|64]</code>. They have been renamed to
1004 <code>@llvm.x86.sse42.crc32.32.[8|16|32]</code> and
1005 <code>@llvm.x86.sse42.crc32.64.[8|64]</code>.</li>
1010 <!--=========================================================================-->
1012 <a name="ARM">ARM Target Improvements</a>
1017 <p>New features of the ARM target include:</p>
1020 <li>The ARM backend generates much faster code for Cortex-A9 chips.</li>
1021 <li>The ARM backend has improved support for Cortex-M series processors.</li>
1022 <li>The ARM inline assembly constraints have been implemented and are now fully
1024 <li>NEON code produced by Clang often runs much faster due to improvements in
1025 the Scalar Replacement of Aggregates pass.</li>
1026 <li>The old ARM disassembler is replaced with a new one based on autogenerated
1027 encoding information from ARM .td files.</li>
1028 <li>The integrated assembler has made major leaps forward, but is still beta quality in LLVM 3.0.</li>
1033 <!--=========================================================================-->
1035 <a name="MIPS">MIPS Target Improvements</a>
1040 <p>This release has seen major new work on just about every aspect of the MIPS
1041 backend. Some of the major new features include:</p>
1044 <li>Most MIPS32r1 and r2 instructions are now supported.</li>
1045 <li>LE/BE MIPS32r1/r2 has been tested extensively.</li>
1046 <li>O32 ABI has been fully tested.</li>
1047 <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>
1048 <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>
1049 <li>Support for old-style JIT is complete.</li>
1050 <li>Support for old architectures (MIPS1 and MIPS2) has been removed.</li>
1051 <li>Initial support for MIPS64 has been added.</li>
1055 <!--=========================================================================-->
1057 <a name="PTX">PTX Target Improvements</a>
1063 The PTX back-end is still experimental, but is fairly usable for compute kernels
1064 in LLVM 3.0. Most scalar arithmetic is implemented, as well as intrinsics to
1065 access the special PTX registers and sync instructions. The major missing
1066 pieces are texture/sampler support and some vector operations.</p>
1068 <p>That said, the backend is already being used for domain-specific languages
1069 and can be used by Clang to
1070 <a href="http://clang.llvm.org/docs/ReleaseNotes.html#opencl">compile OpenCL
1071 C code</a> into PTX.</p>
1075 <!--=========================================================================-->
1077 <a name="OtherTS">Other Target Specific Improvements</a>
1083 <li>Many PowerPC improvements have been implemented for ELF targets, including
1084 support for varargs and initial support for direct .o file emission.</li>
1086 <li>MicroBlaze scheduling itineraries were added that model the
1087 3-stage and the 5-stage pipeline architectures. The 3-stage
1088 pipeline model can be selected with <code>-mcpu=mblaze3</code>
1089 and the 5-stage pipeline model can be selected with
1090 <code>-mcpu=mblaze5</code>.</li>
1096 <!--=========================================================================-->
1098 <a name="changes">Major Changes and Removed Features</a>
1103 <p>If you're already an LLVM user or developer with out-of-tree changes based on
1104 LLVM 2.9, this section lists some "gotchas" that you may run into upgrading
1105 from the previous release.</p>
1108 <li>LLVM 3.0 removes support for reading LLVM 2.8 and earlier files, and LLVM
1109 3.1 will eliminate support for reading LLVM 2.9 files. Going forward, we
1110 aim for all future versions of LLVM to read bitcode files and .ll files
1111 produced by LLVM 3.0.</li>
1112 <li>Tablegen has been split into a library, allowing the clang tblgen pieces
1113 to now live in the clang tree. The llvm version has been renamed to
1114 llvm-tblgen instead of tblgen.</li>
1115 <li>The <code>LLVMC</code> meta compiler driver was removed.</li>
1116 <li>The unused PostOrder Dominator Frontiers and LowerSetJmp passes were removed.</li>
1119 <li>The old <code>TailDup</code> pass was not used in the standard pipeline
1120 and was unable to update ssa form, so it has been removed.
1121 <li>The syntax of volatile loads and stores in IR has been changed to
1122 "<code>load volatile</code>"/"<code>store volatile</code>". The old
1123 syntax ("<code>volatile load</code>"/"<code>volatile store</code>")
1124 is still accepted, but is now considered deprecated and will be removed in
1126 <li>llvm-gcc's frontend tests have been removed from llvm/test/Frontend*, sunk
1127 into the clang and dragonegg testsuites.</li>
1128 <li>The old atomic intrinsics (<code>llvm.memory.barrier</code> and
1129 <code>llvm.atomic.*</code>) are now gone. Please use the new atomic
1130 instructions, described in the <a href="Atomics.html">atomics guide</a>.
1131 <li>LLVM's configure script doesn't depend on llvm-gcc anymore, eliminating a
1132 strange circular dependence between projects.</li>
1135 <h4>Windows (32-bit)</h4>
1139 <li>On Win32(MinGW32 and MSVC), Windows 2000 will not be supported.
1140 Windows XP or higher is required.</li>
1147 <!--=========================================================================-->
1149 <a name="api_changes">Internal API Changes</a>
1154 <p>In addition, many APIs have changed in this release. Some of the major
1155 LLVM API changes are:</p>
1158 <li>The biggest and most pervasive change is that the type system has been
1159 rewritten: <code>PATypeHolder</code> and <code>OpaqueType</code> are gone,
1160 and all APIs deal with <code>Type*</code> instead of <code>const
1161 Type*</code>. If you need to create recursive structures, then create a
1162 named structure, and use <code>setBody()</code> when all its elements are
1163 built. Type merging and refining is gone too: named structures are not
1164 merged with other structures, even if their layout is identical. (of
1165 course anonymous structures are still uniqued by layout).</li>
1167 <li><code>PHINode::reserveOperandSpace</code> has been removed. Instead, you
1168 must specify how many operands to reserve space for when you create the
1169 PHINode, by passing an extra argument
1170 into <code>PHINode::Create</code>.</li>
1172 <li>PHINodes no longer store their incoming BasicBlocks as operands. Instead,
1173 the list of incoming BasicBlocks is stored separately, and can be accessed
1174 with new functions <code>PHINode::block_begin</code>
1175 and <code>PHINode::block_end</code>.</li>
1177 <li>Various functions now take an <code>ArrayRef</code> instead of either a
1178 pair of pointers (or iterators) to the beginning and end of a range, or a
1179 pointer and a length. Others now return an <code>ArrayRef</code> instead
1180 of a reference to a <code>SmallVector</code>
1181 or <code>std::vector</code>. These include:
1183 <!-- Please keep this list sorted. -->
1184 <li><code>CallInst::Create</code></li>
1185 <li><code>ComputeLinearIndex</code> (in <code>llvm/CodeGen/Analysis.h</code>)</li>
1186 <li><code>ConstantArray::get</code></li>
1187 <li><code>ConstantExpr::getExtractElement</code></li>
1188 <li><code>ConstantExpr::getGetElementPtr</code></li>
1189 <li><code>ConstantExpr::getInBoundsGetElementPtr</code></li>
1190 <li><code>ConstantExpr::getIndices</code></li>
1191 <li><code>ConstantExpr::getInsertElement</code></li>
1192 <li><code>ConstantExpr::getWithOperands</code></li>
1193 <li><code>ConstantFoldCall</code> (in <code>llvm/Analysis/ConstantFolding.h</code>)</li>
1194 <li><code>ConstantFoldInstOperands</code> (in <code>llvm/Analysis/ConstantFolding.h</code>)</li>
1195 <li><code>ConstantVector::get</code></li>
1196 <li><code>DIBuilder::createComplexVariable</code></li>
1197 <li><code>DIBuilder::getOrCreateArray</code></li>
1198 <li><code>ExtractValueInst::Create</code></li>
1199 <li><code>ExtractValueInst::getIndexedType</code></li>
1200 <li><code>ExtractValueInst::getIndices</code></li>
1201 <li><code>FindInsertedValue</code> (in <code>llvm/Analysis/ValueTracking.h</code>)</li>
1202 <li><code>gep_type_begin</code> (in <code>llvm/Support/GetElementPtrTypeIterator.h</code>)</li>
1203 <li><code>gep_type_end</code> (in <code>llvm/Support/GetElementPtrTypeIterator.h</code>)</li>
1204 <li><code>GetElementPtrInst::Create</code></li>
1205 <li><code>GetElementPtrInst::CreateInBounds</code></li>
1206 <li><code>GetElementPtrInst::getIndexedType</code></li>
1207 <li><code>InsertValueInst::Create</code></li>
1208 <li><code>InsertValueInst::getIndices</code></li>
1209 <li><code>InvokeInst::Create</code></li>
1210 <li><code>IRBuilder::CreateCall</code></li>
1211 <li><code>IRBuilder::CreateExtractValue</code></li>
1212 <li><code>IRBuilder::CreateGEP</code></li>
1213 <li><code>IRBuilder::CreateInBoundsGEP</code></li>
1214 <li><code>IRBuilder::CreateInsertValue</code></li>
1215 <li><code>IRBuilder::CreateInvoke</code></li>
1216 <li><code>MDNode::get</code></li>
1217 <li><code>MDNode::getIfExists</code></li>
1218 <li><code>MDNode::getTemporary</code></li>
1219 <li><code>MDNode::getWhenValsUnresolved</code></li>
1220 <li><code>SimplifyGEPInst</code> (in <code>llvm/Analysis/InstructionSimplify.h</code>)</li>
1221 <li><code>TargetData::getIndexedOffset</code></li>
1224 <li>All forms of <code>StringMap::getOrCreateValue</code> have been remove
1225 except for the one which takes a <code>StringRef</code>.</li>
1227 <li>The <code>LLVMBuildUnwind</code> function from the C API was removed. The
1228 LLVM <code>unwind</code> instruction has been deprecated for a long time
1229 and isn't used by the current front-ends. So this was removed during the
1230 exception handling rewrite.</li>
1232 <li>The <code>LLVMAddLowerSetJmpPass</code> function from the C API was
1233 removed because the <code>LowerSetJmp</code> pass was removed.</li>
1235 <li>The <code>DIBuilder</code> interface used by front ends to encode
1236 debugging information in the LLVM IR now expects clients to
1237 use <code>DIBuilder::finalize()</code> at the end of translation unit to
1238 complete debugging information encoding.</li>
1240 <li>TargetSelect.h moved to Support/ from Target/</li>
1242 <li>UpgradeIntrinsicCall no longer upgrades pre-2.9 intrinsic calls (for
1243 example <code>llvm.memset.i32</code>).</li>
1245 <li>It is mandatory to initialize all out-of-tree passes too and their dependencies now with
1246 <code>INITIALIZE_PASS{BEGIN,END,}</code>
1247 and <code>INITIALIZE_{PASS,AG}_DEPENDENCY</code>.</li>
1249 <li>The interface for MemDepResult in MemoryDependenceAnalysis has been
1250 enhanced with new return types Unknown and NonFuncLocal, in addition to
1251 the existing types Clobber, Def, and NonLocal.</li>
1258 <!-- *********************************************************************** -->
1260 <a name="knownproblems">Known Problems</a>
1262 <!-- *********************************************************************** -->
1266 <p>LLVM is generally a production quality compiler, and is used by a broad range
1267 of applications and shipping in many products. That said, not every
1268 subsystem is as mature as the aggregate, particularly the more obscure
1269 targets. If you run into a problem, please check the <a
1270 href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
1271 there isn't already one or ask on the <a
1272 href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev
1275 <p>Known problem areas include:</p>
1278 <li>The Alpha, Blackfin, CellSPU, MSP430, PTX, SystemZ and
1279 XCore backends are experimental, and the Alpha, Blackfin and SystemZ
1280 targets have already been removed from mainline.</li>
1282 <li>The integrated assembler, disassembler, and JIT is not supported by
1283 several targets. If an integrated assembler is not supported, then a
1284 system assembler is required. For more details, see the <a
1285 href="CodeGenerator.html#targetfeatures">Target Features Matrix</a>.
1288 <li>The C backend has numerous problems and is not being actively maintained.
1289 Depending on it for anything serious is not advised.</li>
1294 <!-- *********************************************************************** -->
1296 <a name="additionalinfo">Additional Information</a>
1298 <!-- *********************************************************************** -->
1302 <p>A wide variety of additional information is available on
1303 the <a href="http://llvm.org/">LLVM web page</a>, in particular in
1304 the <a href="http://llvm.org/docs/">documentation</a> section. The web page
1305 also contains versions of the API documentation which is up-to-date with the
1306 Subversion version of the source code. You can access versions of these
1307 documents specific to this release by going into the "<tt>llvm/doc/</tt>"
1308 directory in the LLVM tree.</p>
1310 <p>If you have any questions or comments about LLVM, please feel free to contact
1311 us via the <a href="http://llvm.org/docs/#maillist"> mailing lists</a>.</p>
1315 <!--=========================================================================-->
1317 <!-- EH details: to be moved to a blog post:
1322 <p>One of the biggest changes is that 3.0 has a new exception handling
1323 system. The old system used LLVM intrinsics to convey the exception handling
1324 information to the code generator. It worked in most cases, but not
1325 all. Inlining was especially difficult to get right. Also, the intrinsics
1326 could be moved away from the <code>invoke</code> instruction, making it hard
1327 to recover that information.</p>
1329 <p>The new EH system makes exception handling a first-class member of the IR. It
1330 adds two new instructions:</p>
1333 <li><a href="LangRef.html#i_landingpad"><code>landingpad</code></a> —
1334 this instruction defines a landing pad basic block. It contains all of the
1335 information that's needed by the code generator. It's also required to be
1336 the first non-PHI instruction in the landing pad. In addition, a landing
1337 pad may be jumped to only by the unwind edge of an <code>invoke</code>
1340 <li><a href="LangRef.html#i_resume"><code>resume</code></a> — this
1341 instruction causes the current exception to resume traveling up the
1342 stack. It replaces the <code>@llvm.eh.resume</code> intrinsic.</li>
1345 <p>Converting from the old EH API to the new EH API is rather simple, because a
1346 lot of complexity has been removed. The two intrinsics,
1347 <code>@llvm.eh.exception</code> and <code>@llvm.eh.selector</code> have been
1348 superseded by the <code>landingpad</code> instruction. Instead of generating
1349 a call to <code>@llvm.eh.exception</code> and <code>@llvm.eh.selector</code>:
1351 <div class="doc_code">
1353 Function *ExcIntr = Intrinsic::getDeclaration(TheModule,
1354 Intrinsic::eh_exception);
1355 Function *SlctrIntr = Intrinsic::getDeclaration(TheModule,
1356 Intrinsic::eh_selector);
1358 // The exception pointer.
1359 Value *ExnPtr = Builder.CreateCall(ExcIntr, "exc_ptr");
1361 std::vector<Value*> Args;
1362 Args.push_back(ExnPtr);
1363 Args.push_back(Builder.CreateBitCast(Personality,
1364 Type::getInt8PtrTy(Context)));
1366 <i>// Add selector clauses to Args.</i>
1368 // The selector call.
1369 Builder.CreateCall(SlctrIntr, Args, "exc_sel");
1373 <p>You should instead generate a <code>landingpad</code> instruction, that
1374 returns an exception object and selector value:</p>
1376 <div class="doc_code">
1378 LandingPadInst *LPadInst =
1379 Builder.CreateLandingPad(StructType::get(Int8PtrTy, Int32Ty, NULL),
1382 Value *LPadExn = Builder.CreateExtractValue(LPadInst, 0);
1383 Builder.CreateStore(LPadExn, getExceptionSlot());
1385 Value *LPadSel = Builder.CreateExtractValue(LPadInst, 1);
1386 Builder.CreateStore(LPadSel, getEHSelectorSlot());
1390 <p>It's now trivial to add the individual clauses to the <code>landingpad</code>
1393 <div class="doc_code">
1395 <i><b>// Adding a catch clause</b></i>
1396 Constant *TypeInfo = getTypeInfo();
1397 LPadInst->addClause(TypeInfo);
1399 <i><b>// Adding a C++ catch-all</b></i>
1400 LPadInst->addClause(Constant::getNullValue(Builder.getInt8PtrTy()));
1402 <i><b>// Adding a cleanup</b></i>
1403 LPadInst->setCleanup(true);
1405 <i><b>// Adding a filter clause</b></i>
1406 std::vector<Constant*> TypeInfos;
1407 Constant *TypeInfo = getFilterTypeInfo();
1408 TypeInfos.push_back(Builder.CreateBitCast(TypeInfo, Builder.getInt8PtrTy()));
1410 ArrayType *FilterTy = ArrayType::get(Int8PtrTy, TypeInfos.size());
1411 LPadInst->addClause(ConstantArray::get(FilterTy, TypeInfos));
1415 <p>Converting from using the <code>@llvm.eh.resume</code> intrinsic to
1416 the <code>resume</code> instruction is trivial. It takes the exception
1417 pointer and exception selector values returned by
1418 the <code>landingpad</code> instruction:</p>
1420 <div class="doc_code">
1422 Type *UnwindDataTy = StructType::get(Builder.getInt8PtrTy(),
1423 Builder.getInt32Ty(), NULL);
1424 Value *UnwindData = UndefValue::get(UnwindDataTy);
1425 Value *ExcPtr = Builder.CreateLoad(getExceptionObjSlot());
1426 Value *ExcSel = Builder.CreateLoad(getExceptionSelSlot());
1427 UnwindData = Builder.CreateInsertValue(UnwindData, ExcPtr, 0, "exc_ptr");
1428 UnwindData = Builder.CreateInsertValue(UnwindData, ExcSel, 1, "exc_sel");
1429 Builder.CreateResume(UnwindData);
1439 <!-- *********************************************************************** -->
1443 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1444 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1445 <a href="http://validator.w3.org/check/referer"><img
1446 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1448 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
1449 Last modified: $Date$