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:
99 <li>Greatly improved support for building C++ applications, with greater
100 stability and better diagnostics.</li>
102 <li><a href="http://clang.llvm.org/cxx_status.html">Improved support</a> for
103 the <a href="http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_detail.htm?csnumber=50372">C++
104 2011</a> standard (aka "C++'0x"), including implementations of non-static data member
105 initializers, alias templates, delegating constructors, range-based
106 for loops, and implicitly-generated move constructors and move assignment
107 operators, among others.</li>
109 <li>Implemented support for some features of the upcoming C1x standard,
110 including static assertions and generic selections.</li>
112 <li>Better detection of include and linking paths for system headers and
113 libraries, especially for Linux distributions.</li>
115 <li>Several improvements to Objective-C support, including:
118 <li><a href="http://clang.llvm.org/docs/AutomaticReferenceCounting.html">
119 Automatic Reference Counting</a> (ARC) and an improved memory model
120 cleanly separating object and C memory.</li>
122 <li>A migration tool for moving manual retain/release code to ARC</li>
124 <li>Better support for data hiding, allowing instance variables to be
125 declared in implementation contexts or class extensions</li>
126 <li>Weak linking support for Objective-C classes</li>
127 <li>Improved static type checking by inferring the return type of methods
128 such as +alloc and -init.</li>
131 Some new Objective-C features require either the Mac OS X 10.7 / iOS 5
132 Objective-C runtime, or version 1.6 or later of the GNUstep Objective-C
133 runtime version.</li>
135 <li>Implemented a number of optimizations in <tt>libclang</tt>, the Clang C
136 interface, to improve the performance of code completion and the mapping
137 from source locations to abstract syntax tree nodes.</li>
139 For more details about the changes to Clang since the 2.9 release, see the
140 <a href="http://clang.llvm.org/docs/ReleaseNotes.html">Clang release notes</a>
144 <p>If Clang rejects your code but another compiler accepts it, please take a
145 look at the <a href="http://clang.llvm.org/compatibility.html">language
146 compatibility</a> guide to make sure this is not intentional or a known
151 <!--=========================================================================-->
153 <a name="dragonegg">DragonEgg: GCC front-ends, LLVM back-end</a>
157 <p><a href="http://dragonegg.llvm.org/">DragonEgg</a> is a
158 <a href="http://gcc.gnu.org/wiki/plugins">gcc plugin</a> that replaces GCC's
159 optimizers and code generators with LLVM's. It works with gcc-4.5 or gcc-4.6,
160 targets the x86-32 and x86-64 processor families, and has been successfully
161 used on the Darwin, FreeBSD, KFreeBSD, Linux and OpenBSD platforms. It fully
162 supports Ada, C, C++ and Fortran. It has partial support for Go, Java, Obj-C
165 <p>The 3.0 release has the following notable changes:</p>
168 <li>GCC version 4.6 is now fully supported.</li>
170 <li>Patching and building GCC is no longer required: the plugin should work
171 with your system GCC (version 4.5 or 4.6; on Debian/Ubuntu systems the
172 gcc-4.5-plugin-dev or gcc-4.6-plugin-dev package is also needed).</li>
174 <li>The <tt>-fplugin-arg-dragonegg-enable-gcc-optzns</tt> option, which runs
175 GCC's optimizers as well as LLVM's, now works much better. This is the
176 option to use if you want ultimate performance! It is still experimental
177 though: it may cause the plugin to crash. Setting the optimization level
178 to <tt>-O4</tt> when using this option will optimize even harder, though
179 this usually doesn't result in any improvement over <tt>-O3</tt>.</li>
181 <li>The type and constant conversion logic has been almost entirely rewritten,
182 fixing a multitude of obscure bugs.</li>
188 <!--=========================================================================-->
190 <a name="compiler-rt">compiler-rt: Compiler Runtime Library</a>
195 <p>The new LLVM <a href="http://compiler-rt.llvm.org/">compiler-rt project</a>
196 is a simple library that provides an implementation of the low-level
197 target-specific hooks required by code generation and other runtime
198 components. For example, when compiling for a 32-bit target, converting a
199 double to a 64-bit unsigned integer is compiled into a runtime call to the
200 "__fixunsdfdi" function. The compiler-rt library provides highly optimized
201 implementations of this and other low-level routines (some are 3x faster than
202 the equivalent libgcc routines).</p>
204 <p>In the LLVM 3.0 timeframe, the target specific ARM code has converted to
205 "unified" assembly syntax, and several new functions have been added to the
210 <!--=========================================================================-->
212 <a name="lldb">LLDB: Low Level Debugger</a>
217 <p>LLDB is a ground-up implementation of a command line debugger, as well as a
218 debugger API that can be used from other applications. LLDB makes use of the
219 Clang parser to provide high-fidelity expression parsing (particularly for
220 C++) and uses the LLVM JIT for target support.</p>
222 <p>LLDB has advanced by leaps and bounds in the 3.0 timeframe. It is
223 dramatically more stable and useful, and includes both a
224 new <a href="http://lldb.llvm.org/tutorial.html">tutorial</a> and
225 a <a href="http://lldb.llvm.org/lldb-gdb.html">side-by-side comparison with
230 <!--=========================================================================-->
232 <a name="libc++">libc++: C++ Standard Library</a>
237 <p>Like compiler_rt, libc++ is now <a href="DeveloperPolicy.html#license">dual
238 licensed</a> under the MIT and UIUC license, allowing it to be used more
241 <p>Libc++ has been ported to FreeBSD and imported into the base system. It is
242 planned to be the default STL implementation for FreeBSD 10.</p>
246 <!--=========================================================================-->
248 <a name="vmkit">VMKit</a>
253 <p>The <a href="http://vmkit.llvm.org/">VMKit project</a> is an
254 implementation of a Java Virtual Machine (Java VM or JVM) that uses LLVM for
255 static and just-in-time compilation.
257 <p>In the LLVM 3.0 time-frame, VMKit has had significant improvements on both
258 runtime and startup performance:</p>
261 <li>Precompilation: by compiling ahead of time a small subset of Java's core
262 library, the startup performance have been highly optimized to the point that
263 running a 'Hello World' program takes less than 30 milliseconds.</li>
265 <li>Customization: by customizing virtual methods for individual classes,
266 the VM can statically determine the target of a virtual call, and decide to
269 <li>Inlining: the VM does more inlining than it did before, by allowing more
270 bytecode instructions to be inlined, and thanks to customization. It also
271 inlines GC barriers, and object allocations.</li>
273 <li>New exception model: the generated code for a method that does not do
274 any try/catch is not penalized anymore by the eventuality of calling a
275 method that throws an exception. Instead, the method that throws the
276 exception jumps directly to the method that could catch it.</li>
282 <!--=========================================================================-->
284 <a name="LLBrowse">LLBrowse: IR Browser</a>
289 <p><a href="http://llvm.org/svn/llvm-project/llbrowse/trunk/doc/LLBrowse.html">
290 LLBrowse</a> is an interactive viewer for LLVM modules. It can load any LLVM
291 module and displays its contents as an expandable tree view, facilitating an
292 easy way to inspect types, functions, global variables, or metadata nodes. It
293 is fully cross-platform, being based on the popular wxWidgets GUI
299 <!--=========================================================================-->
302 <a name="klee">KLEE: A Symbolic Execution Virtual Machine</a>
307 <a href="http://klee.llvm.org/">KLEE</a> is a symbolic execution framework for
308 programs in LLVM bitcode form. KLEE tries to symbolically evaluate "all" paths
309 through the application and records state transitions that lead to fault
310 states. This allows it to construct testcases that lead to faults and can even
311 be used to verify some algorithms.
319 <!-- *********************************************************************** -->
321 <a name="externalproj">External Open Source Projects Using LLVM 3.0</a>
323 <!-- *********************************************************************** -->
327 <p>An exciting aspect of LLVM is that it is used as an enabling technology for
328 a lot of other language and tools projects. This section lists some of the
329 projects that have already been updated to work with LLVM 3.0.</p>
331 <!--=========================================================================-->
332 <h3>AddressSanitizer</h3>
336 <p><a href="http://code.google.com/p/address-sanitizer/">AddressSanitizer</a>
337 uses compiler instrumentation and a specialized malloc library to find C/C++
338 bugs such as use-after-free and out-of-bound accesses to heap, stack, and
339 globals. The key feature of the tool is speed: the average slowdown
340 introduced by AddressSanitizer is less than 2x.</p>
344 <!--=========================================================================-->
349 <p><a href="http://www.clamav.net">Clam AntiVirus</a> is an open source (GPL)
350 anti-virus toolkit for UNIX, designed especially for e-mail scanning on mail
353 <p>Since version 0.96 it
354 has <a href="http://vrt-sourcefire.blogspot.com/2010/09/introduction-to-clamavs-low-level.html">bytecode
355 signatures</a> that allow writing detections for complex malware.
356 It uses LLVM's JIT to speed up the execution of bytecode on X86, X86-64,
357 PPC32/64, falling back to its own interpreter otherwise. The git version was
358 updated to work with LLVM 3.0.</p>
362 <!--=========================================================================-->
363 <h3>clang_complete for VIM</h3>
367 <p><a href="https://github.com/Rip-Rip/clang_complete">clang_complete</a> is a
368 VIM plugin, that provides accurate C/C++ autocompletion using the clang front
369 end. The development version of clang complete, can directly use libclang
370 which can maintain a cache to speed up auto completion.</p>
374 <!--=========================================================================-->
379 <p><a href="https://bitbucket.org/dwilliamson/clreflect">clReflect</a> is a C++
380 parser that uses clang/LLVM to derive a light-weight reflection database
381 suitable for use in game development. It comes with a very simple runtime
382 library for loading and querying the database, requiring no external
383 dependencies (including CRT), and an additional utility library for object
384 management and serialisation.</p>
388 <!--=========================================================================-->
389 <h3>Cling C++ Interpreter</h3>
393 <p><a href="http://cern.ch/cling">Cling</a> is an interactive compiler interface
394 (aka C++ interpreter). It supports C++ and C, and uses LLVM's JIT and the
395 Clang parser. It has a prompt interface, runs source files, calls into shared
396 libraries, prints the value of expressions, even does runtime lookup of
397 identifiers (dynamic scopes). And it just behaves like one would expect from
402 <!--=========================================================================-->
403 <h3>Crack Programming Language</h3>
407 <p><a href="http://code.google.com/p/crack-language/">Crack</a> aims to provide
408 the ease of development of a scripting language with the performance of a
409 compiled language. The language derives concepts from C++, Java and Python,
410 incorporating object-oriented programming, operator overloading and strong
415 <!--=========================================================================-->
420 <p><a href="http://eerolanguage.org/">Eero</a> is a fully
421 header-and-binary-compatible dialect of Objective-C 2.0, implemented with a
422 patched version of the Clang/LLVM compiler. It features a streamlined syntax,
423 Python-like indentation, and new operators, for improved readability and
424 reduced code clutter. It also has new features such as limited forms of
425 operator overloading and namespaces, and strict (type-and-operator-safe)
426 enumerations. It is inspired by languages such as Smalltalk, Python, and
431 <!--=========================================================================-->
432 <h3>FAUST Real-Time Audio Signal Processing Language</h3>
436 <p><a href="http://faust.grame.fr/">FAUST</a> is a compiled language for
437 real-time audio signal processing. The name FAUST stands for Functional
438 AUdio STream. Its programming model combines two approaches: functional
439 programming and block diagram composition. In addition with the C, C++, Java
440 output formats, the Faust compiler can now generate LLVM bitcode, and works
446 <!--=========================================================================-->
447 <h3>Glasgow Haskell Compiler (GHC)</h3>
451 <p>GHC is an open source, state-of-the-art programming suite for Haskell, a
452 standard lazy functional programming language. It includes an optimizing
453 static compiler generating good code for a variety of platforms, together
454 with an interactive system for convenient, quick development.</p>
456 <p>GHC 7.0 and onwards include an LLVM code generator, supporting LLVM 2.8 and
457 later. Since LLVM 2.9, GHC now includes experimental support for the ARM
458 platform with LLVM 3.0.</p>
462 <!--=========================================================================-->
467 <p><a href="http://botwars.tk/gwscript/">gwXscript</a> is an object oriented,
468 aspect oriented programming language which can create both executables (ELF,
469 EXE) and shared libraries (DLL, SO, DYNLIB). The compiler is implemented in
470 its own language and translates scripts into LLVM-IR which can be optimized
471 and translated into native code by the LLVM framework. Source code in
472 gwScript contains definitions that expand the namespaces. So you can build
473 your project and simply 'plug out' features by removing a file. The remaining
474 project does not leave scars since you directly separate concerns by the
475 'template' feature of gwX. It is also possible to add new features to a
476 project by just adding files and without editing the original project. This
477 language is used for example to create games or content management systems
478 that should be extendable.</p>
480 <p>gwXscript is strongly typed and offers comfort with its native types string,
481 hash and array. You can easily write new libraries in gwXscript or native
482 code. gwXscript is type safe and users should not be able to crash your
483 program or execute malicious code except code that is eating CPU time.</p>
487 <!--=========================================================================-->
488 <h3>include-what-you-use</h3>
492 <p><a href="http://code.google.com/p/include-what-you-use">include-what-you-use</a>
493 is a tool to ensure that a file directly <code>#include</code>s
494 all <code>.h</code> files that provide a symbol that the file uses. It also
495 removes superfluous <code>#include</code>s from source files.</p>
499 <!--=========================================================================-->
500 <h3>ispc: The Intel SPMD Program Compiler</h3>
504 <p><a href="http://ispc.github.com">ispc</a> is a compiler for "single program,
505 multiple data" (SPMD) programs. It compiles a C-based SPMD programming
506 language to run on the SIMD units of CPUs; it often delivers 5-6x speedups on
507 a single core of a CPU with an 8-wide SIMD unit compared to serial code,
508 while still providing a clean and easy-to-understand programming model. For
509 an introduction to the language and its performance,
510 see <a href="http://ispc.github.com/example.html">the walkthrough</a> of a short
511 example program. ispc is licensed under the BSD license.</p>
515 <!--=========================================================================-->
516 <h3>The Julia Programming Language</h3>
520 <p><a href="http://github.com/JuliaLang/julia">Julia</a> is a high-level,
521 high-performance dynamic language for technical
522 computing. It provides a sophisticated compiler, distributed parallel
523 execution, numerical accuracy, and an extensive mathematical function
524 library. The compiler uses type inference to generate fast code
525 without any type declarations, and uses LLVM's optimization passes and
526 JIT compiler. The language is designed around multiple dispatch,
527 giving programs a large degree of flexibility. It is ready for use on many
528 kinds of problems.</p>
531 <!--=========================================================================-->
532 <h3>LanguageKit and Pragmatic Smalltalk</h3>
536 <p><a href="http://etoileos.com/etoile/features/languagekit/">LanguageKit</a> is
537 a framework for implementing dynamic languages sharing an object model with
538 Objective-C. It provides static and JIT compilation using LLVM along with
539 its own interpreter. Pragmatic Smalltalk is a dialect of Smalltalk, built on
540 top of LanguageKit, that interfaces directly with Objective-C, sharing the
541 same object representation and message sending behaviour. These projects are
542 developed as part of the Étoilé desktop environment.</p>
546 <!--=========================================================================-->
551 <p><a href="http://lua-av.mat.ucsb.edu/blog/">LuaAV</a> is a real-time
552 audiovisual scripting environment based around the Lua language and a
553 collection of libraries for sound, graphics, and other media protocols. LuaAV
554 uses LLVM and Clang to JIT compile efficient user-defined audio synthesis
555 routines specified in a declarative syntax.</p>
559 <!--=========================================================================-->
564 <p>An open source, cross-platform implementation of C# and the CLR that is
565 binary compatible with Microsoft.NET. Has an optional, dynamically-loaded
566 LLVM code generation backend in Mini, the JIT compiler.</p>
568 <p>Note that we use a Git mirror of LLVM <a
569 href="https://github.com/mono/llvm">with some patches</a>.</p>
573 <!--=========================================================================-->
578 <p><a href="http://polly.grosser.es">Polly</a> is an advanced data-locality
579 optimizer and automatic parallelizer. It uses an advanced, mathematical
580 model to calculate detailed data dependency information which it uses to
581 optimize the loop structure of a program. Polly can speed up sequential code
582 by improving memory locality and consequently the cache use. Furthermore,
583 Polly is able to expose different kind of parallelism which it exploits by
584 introducing (basic) OpenMP and SIMD code. A mid-term goal of Polly is to
585 automatically create optimized GPU code.</p>
589 <!--=========================================================================-->
590 <h3>Portable OpenCL (pocl)</h3>
594 <p>Portable OpenCL is an open source implementation of the OpenCL standard which
595 can be easily adapted for new targets. One of the goals of the project is
596 improving performance portability of OpenCL programs, avoiding the need for
597 target-dependent manual optimizations. A "native" target is included, which
598 allows running OpenCL kernels on the host (CPU).</p>
602 <!--=========================================================================-->
606 <p><a href="http://pure-lang.googlecode.com/">Pure</a> is an
607 algebraic/functional programming language based on term rewriting. Programs
608 are collections of equations which are used to evaluate expressions in a
609 symbolic fashion. The interpreter uses LLVM as a backend to JIT-compile Pure
610 programs to fast native code. Pure offers dynamic typing, eager and lazy
611 evaluation, lexical closures, a hygienic macro system (also based on term
612 rewriting), built-in list and matrix support (including list and matrix
613 comprehensions) and an easy-to-use interface to C and other programming
614 languages (including the ability to load LLVM bitcode modules, and inline C,
615 C++, Fortran and Faust code in Pure programs if the corresponding LLVM-enabled
616 compilers are installed).</p>
618 <p>Pure version 0.48 has been tested and is known to work with LLVM 3.0
619 (and continues to work with older LLVM releases >= 2.5).</p>
623 <!--=========================================================================-->
624 <h3>Renderscript</h3>
628 <p><a href="http://developer.android.com/guide/topics/renderscript/index.html">Renderscript</a>
629 is Android's advanced 3D graphics rendering and compute API. It provides a
630 portable C99-based language with extensions to facilitate common use cases
631 for enhancing graphics and thread level parallelism. The Renderscript
632 compiler frontend is based on Clang/LLVM. It emits a portable bitcode format
633 for the actual compiled script code, as well as reflects a Java interface for
634 developers to control the execution of the compiled bitcode. Executable
635 machine code is then generated from this bitcode by an LLVM backend on the
636 device. Renderscript is thus able to provide a mechanism by which Android
637 developers can improve performance of their applications while retaining
642 <!--=========================================================================-->
647 <p><a href="http://safecode.cs.illinois.edu">SAFECode</a> is a memory safe C/C++
648 compiler built using LLVM. It takes standard, unannotated C/C++ code,
649 analyzes the code to ensure that memory accesses and array indexing
650 operations are safe, and instruments the code with run-time checks when
651 safety cannot be proven statically. SAFECode can be used as a debugging aid
652 (like Valgrind) to find and repair memory safety bugs. It can also be used
653 to protect code from security attacks at run-time.</p>
657 <!--=========================================================================-->
658 <h3>The Stupid D Compiler (SDC)</h3>
662 <p><a href="https://github.com/bhelyer/SDC">The Stupid D Compiler</a> is a
663 project seeking to write a self-hosting compiler for the D programming
664 language without using the frontend of the reference compiler (DMD).</p>
668 <!--=========================================================================-->
669 <h3>TTA-based Co-design Environment (TCE)</h3>
673 <p>TCE is a toolset for designing application-specific processors (ASP) based on
674 the Transport triggered architecture (TTA). The toolset provides a complete
675 co-design flow from C/C++ programs down to synthesizable VHDL and parallel
676 program binaries. Processor customization points include the register files,
677 function units, supported operations, and the interconnection network.</p>
679 <p>TCE uses Clang and LLVM for C/C++ language support, target independent
680 optimizations and also for parts of code generation. It generates new
681 LLVM-based code generators "on the fly" for the designed TTA processors and
682 loads them in to the compiler backend as runtime libraries to avoid
683 per-target recompilation of larger parts of the compiler chain.</p>
687 <!--=========================================================================-->
688 <h3>Tart Programming Language</h3>
692 <p><a href="http://code.google.com/p/tart/">Tart</a> is a general-purpose,
693 strongly typed programming language designed for application
694 developers. Strongly inspired by Python and C#, Tart focuses on practical
695 solutions for the professional software developer, while avoiding the clutter
696 and boilerplate of legacy languages like Java and C++. Although Tart is still
697 in development, the current implementation supports many features expected of
698 a modern programming language, such as garbage collection, powerful
699 bidirectional type inference, a greatly simplified syntax for template
700 metaprogramming, closures and function literals, reflection, operator
701 overloading, explicit mutability and immutability, and much more. Tart is
702 flexible enough to accommodate a broad range of programming styles and
703 philosophies, while maintaining a strong commitment to simplicity, minimalism
704 and elegance in design.</p>
708 <!--=========================================================================-->
709 <h3>ThreadSanitizer</h3>
713 <p><a href="http://code.google.com/p/data-race-test/">ThreadSanitizer</a> is a
714 data race detector for (mostly) C and C++ code, available for Linux, Mac OS
715 and Windows. On different systems, we use binary instrumentation frameworks
716 (Valgrind and Pin) as frontends that generate the program events for the race
717 detection algorithm. On Linux, there's an option of using LLVM-based
718 compile-time instrumentation.</p>
724 <!-- *********************************************************************** -->
726 <a name="whatsnew">What's New in LLVM 3.0?</a>
728 <!-- *********************************************************************** -->
732 <p>This release includes a huge number of bug fixes, performance tweaks and
733 minor improvements. Some of the major improvements and new features are
734 listed in this section.</p>
736 <!--=========================================================================-->
738 <a name="majorfeatures">Major New Features</a>
743 <!-- Features that need text if they're finished for 3.1:
747 loop dependence analysis
748 CorrelatedValuePropagation
749 lib/Transforms/IPO/MergeFunctions.cpp => consider for 3.1.
750 Integrated assembler on by default for arm/thumb?
755 Analysis/RegionInfo.h + Dom Frontiers
756 SparseBitVector: used in LiveVar.
757 llvm/lib/Archive - replace with lib object?
760 <p>LLVM 3.0 includes several major changes and big features:</p>
763 <li>llvm-gcc is no longer supported, and not included in the release. We
764 recommend switching to <a
765 href="http://clang.llvm.org/">Clang</a> or <a
766 href="http://dragonegg.llvm.org/">DragonEgg</a>.</li>
768 <li>The linear scan register allocator has been replaced with a new "greedy"
769 register allocator, enabling live range splitting and many other
770 optimizations that lead to better code quality. Please see its <a
771 href="http://blog.llvm.org/2011/09/greedy-register-allocation-in-llvm-30.html">blog post</a> or its talk at the <a
772 href="http://llvm.org/devmtg/2011-11/">Developer Meeting</a>
773 for more information.</li>
774 <li>LLVM IR now includes full support for <a href="Atomics.html">atomics
775 memory operations</a> intended to support the C++'11 and C'1x memory models.
776 This includes <a href="LangRef.html#memoryops">atomic load and store,
777 compare and exchange, and read/modify/write instructions</a> as well as a
778 full set of <a href="LangRef.html#ordering">memory ordering constraints</a>.
779 Please see the <a href="Atomics.html">Atomics Guide</a> for more
782 <li>The LLVM IR exception handling representation has been redesigned and
783 reimplemented, making it more elegant, fixing a huge number of bugs, and
784 enabling inlining and other optimizations. Please see its <a href=
785 "http://blog.llvm.org/2011/11/llvm-30-exception-handling-redesign.html">blog
786 post</a> and the <a href="ExceptionHandling.html">Exception Handling
787 documentation</a> for more information.</li>
788 <li>The LLVM IR Type system has been redesigned and reimplemented, making it
789 faster and solving some long-standing problems.
791 href="http://blog.llvm.org/2011/11/llvm-30-type-system-rewrite.html">blog
792 post</a> for more information.</li>
794 <li>The MIPS backend has made major leaps in this release, going from an
795 experimental target to being virtually production quality and supporting a
796 wide variety of MIPS subtargets. See the <a href="#MIPS">MIPS section</a>
797 below for more information.</li>
799 <li>The optimizer and code generator now supports gprof and gcov-style coverage
800 and profiling information, and includes a new llvm-cov tool (but also works
801 with gcov). Clang exposes coverage and profiling through GCC-compatible
802 command line options.</li>
808 <!--=========================================================================-->
810 <a name="coreimprovements">LLVM IR and Core Improvements</a>
815 <p>LLVM IR has several new features for better support of new targets and that
816 expose new optimization opportunities:</p>
819 <li><a href="Atomics.html">Atomic memory accesses and memory ordering</a> are
820 now directly expressible in the IR.</li>
821 <li>A new <a href="LangRef.html#int_fma">llvm.fma intrinsic</a> directly
822 represents floating point multiply accumulate operations without an
823 intermediate rounding stage.</li>
824 <li>A new <a href="LangRef.html#int_expect">llvm.expect intrinsic</a> allows a
825 frontend to express expected control flow (and the
826 <a href="BranchWeightMetadata.html#builtin_expect">
827 <code>__builtin_expect</code></a> from GNU C).</li>
828 <li>The <a href="LangRef.html#int_prefetch">llvm.prefetch intrinsic</a> now
829 takes a 4th argument that specifies whether the prefetch happens from the
830 icache or dcache.</li>
831 <li>The new <a href="LangRef.html#uwtable">uwtable function attribute</a>
832 allows a frontend to control emission of unwind tables.</li>
833 <li>The new <a href="LangRef.html#fnattrs">nonlazybind function
834 attribute</a> allow optimization of Global Offset Table (GOT) accesses.</li>
835 <li>The new <a href="LangRef.html#returns_twice">returns_twice attribute</a>
836 allows better modeling of functions like setjmp.</li>
837 <li>The <a href="LangRef.html#datalayout">target datalayout</a> string can now
838 encode the natural alignment of the target's stack for better optimization.
843 <!--=========================================================================-->
845 <a name="optimizer">Optimizer Improvements</a>
850 <p>In addition to many minor performance tweaks and bug fixes, this
851 release includes a few major enhancements and additions to the
855 <li>The pass manager now has an extension API that allows front-ends and plugins
856 to insert their own optimizations in the well-known places in the standard
857 pass optimization pipeline.</li>
859 <li>Information about <a href="BranchWeightMetadata.html">branch probability</a>
860 and basic block frequency is now available within LLVM, based on a
861 combination of static branch prediction heuristics and
862 <code>__builtin_expect</code> calls. That information is currently used for
863 register spill placement and if-conversion, with additional optimizations
864 planned for future releases. The same framework is intended for eventual
865 use with profile-guided optimization.</li>
867 <li>The "-indvars" induction variable simplification pass only modifies
868 induction variables when profitable. Sign and zero extension
869 elimination, linear function test replacement, loop unrolling, and
870 other simplifications that require induction variable analysis have
871 been generalized so they no longer require loops to be rewritten into
872 canonical form prior to optimization. This new design
873 preserves more IR level information, avoids undoing earlier loop
874 optimizations (particularly hand-optimized loops), and no longer
875 requires the code generator to reconstruct loops into an optimal form -
876 an intractable problem.</li>
878 <li>LLVM now includes a pass to optimize retain/release calls for the
879 <a href="http://clang.llvm.org/docs/AutomaticReferenceCounting.html">Automatic
880 Reference Counting</a> (ARC) Objective-C language feature (in
881 lib/Transforms/Scalar/ObjCARC.cpp). It is a decent example of implementing
882 a source-language-specific optimization in LLVM.</li>
888 <!--=========================================================================-->
890 <a name="mc">MC Level Improvements</a>
895 <p>The LLVM Machine Code (aka MC) subsystem was created to solve a number of
896 problems in the realm of assembly, disassembly, object file format handling,
897 and a number of other related areas that CPU instruction-set level tools work
898 in. For more information, please see
899 the <a href="http://blog.llvm.org/2010/04/intro-to-llvm-mc-project.html">Intro
900 to the LLVM MC Project Blog Post</a>.</p>
903 <li>The MC layer has undergone significant refactoring to eliminate layering
904 violations that caused it to pull in the LLVM compiler backend code.</li>
905 <li>The ELF object file writers are much more full featured.</li>
906 <li>The integrated assembler now supports #line directives.</li>
907 <li>An early implementation of a JIT built on top of the MC framework (known
908 as MC-JIT) has been implemented and will eventually replace the old JIT.
909 It emits object files direct to memory and uses a runtime dynamic linker to
910 resolve references and drive lazy compilation. The MC-JIT enables much
911 greater code reuse between the JIT and the static compiler and provides
912 better integration with the platform ABI as a result.
914 <li>The assembly printer now makes uses of assemblers instruction aliases
915 (InstAliases) to print simplified mneumonics when possible.</li>
916 <li>TableGen can now autogenerate MC expansion logic for pseudo
917 instructions that expand to multiple MC instructions (through the
918 PseudoInstExpansion class).</li>
919 <li>A new llvm-dwarfdump tool provides a start of a drop-in
920 replacement for the corresponding tool that use LLVM libraries. As part of
921 this, LLVM has the beginnings of a dwarf parsing library.</li>
922 <li>llvm-objdump has more output including, symbol by symbol disassembly,
923 inline relocations, section headers, symbol tables, and section contents.
924 Support for archive files has also been added.</li>
925 <li>llvm-nm has gained support for archives of binary files.</li>
926 <li>llvm-size has been added. This tool prints out section sizes.</li>
931 <!--=========================================================================-->
933 <a name="codegen">Target Independent Code Generator Improvements</a>
938 <p>We have put a significant amount of work into the code generator
939 infrastructure, which allows us to implement more aggressive algorithms and
940 make it run faster:</p>
943 <li>LLVM can now produce code that works with libgcc
944 to <a href="SegmentedStacks.html">dynamically allocate stack
945 segments</a>, as opposed to allocating a worst-case chunk of
946 virtual memory for each thread.</li>
947 <li>LLVM generates substantially better code for indirect gotos due to a new
948 tail duplication pass, which can be a substantial performance win for
949 interpreter loops that use them.</li>
950 <li>Exception handling and debug frame information is now emitted with CFI
951 directives. This lets the assembler produce more compact info as it knows
952 the final offsets, yielding <a href="http://blog.mozilla.com/respindola/2011/05/12/cfi-directives/">much smaller executables</a> for some C++ applications.
953 If the system assembler doesn't support it, MC exands the directives when
954 the integrated assembler is not used.
957 <li>The code generator now supports vector "select" operations on vector
958 comparisons, turning them into various optimized code sequences (e.g.
959 using the SSE4/AVX "blend" instructions).</li>
960 <li>The SSE execution domain fix pass and the ARM NEON move fix pass have been
961 merged to a target independent execution dependency fix pass. This pass is
962 used to select alternative equivalent opcodes in a way that minimizes
963 execution domain crossings. Closely connected instructions are moved to
964 the same execution domain when possible. Targets can override the
965 <code>getExecutionDomain</code> and <code>setExecutionDomain</code> hooks
966 to use the pass.</li>
970 <!--=========================================================================-->
972 <a name="x86">X86-32 and X86-64 Target Improvements</a>
977 <p>New features and major changes in the X86 target include:</p>
980 <li>The X86 backend, assembler and disassembler now have full support for AVX 1.
981 To enable it pass <code>-mavx</code> to the compiler. AVX2 implementation is
982 underway on mainline.</li>
983 <li>The integrated assembler and disassembler now support a broad range of new
984 instructions including Atom, Ivy Bridge, <a
985 href="http://en.wikipedia.org/wiki/SSE4a">SSE4a/BMI</a> instructions, <a
986 href="http://en.wikipedia.org/wiki/RdRand">rdrand</a> and many others.</li>
987 <li>The X86 backend now fully supports the <a href="http://llvm.org/PR879">X87
988 floating point stack inline assembly constraints</a>.</li>
989 <li>The integrated assembler now supports the <tt>.code32</tt> and
990 <tt>.code64</tt> directives to switch between 32-bit and 64-bit
992 <li>The X86 backend now synthesizes horizontal add/sub instructions from generic
993 vector code when the appropriate instructions are enabled.</li>
994 <li>The X86-64 backend generates smaller and faster code at -O0 due to
995 improvements in fast instruction selection.</li>
996 <li><a href="http://code.google.com/p/nativeclient/">Native Client</a>
997 subtarget support has been added.</li>
999 <li>The CRC32 intrinsics have been renamed. The intrinsics were previously
1000 <code>@llvm.x86.sse42.crc32.[8|16|32]</code>
1001 and <code>@llvm.x86.sse42.crc64.[8|64]</code>. They have been renamed to
1002 <code>@llvm.x86.sse42.crc32.32.[8|16|32]</code> and
1003 <code>@llvm.x86.sse42.crc32.64.[8|64]</code>.</li>
1008 <!--=========================================================================-->
1010 <a name="ARM">ARM Target Improvements</a>
1015 <p>New features of the ARM target include:</p>
1018 <li>The ARM backend generates much faster code for Cortex-A9 chips.</li>
1019 <li>The ARM backend has improved support for Cortex-M series processors.</li>
1020 <li>The ARM inline assembly constraints have been implemented and are now fully
1022 <li>NEON code produced by Clang often runs much faster due to improvements in
1023 the Scalar Replacement of Aggregates pass.</li>
1024 <li>The old ARM disassembler is replaced with a new one based on autogenerated
1025 encoding information from ARM .td files.</li>
1026 <li>The integrated assembler has made major leaps forward, but is still beta quality in LLVM 3.0.</li>
1031 <!--=========================================================================-->
1033 <a name="MIPS">MIPS Target Improvements</a>
1038 <p>This release has seen major new work on just about every aspect of the MIPS
1039 backend. Some of the major new features include:</p>
1042 <li>Most MIPS32r1 and r2 instructions are now supported.</li>
1043 <li>LE/BE MIPS32r1/r2 has been tested extensively.</li>
1044 <li>O32 ABI has been fully tested.</li>
1045 <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>
1046 <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>
1047 <li>Support for old-style JIT is complete.</li>
1048 <li>Support for old architectures (MIPS1 and MIPS2) has been removed.</li>
1049 <li>Initial support for MIPS64 has been added.</li>
1053 <!--=========================================================================-->
1055 <a name="PTX">PTX Target Improvements</a>
1061 The PTX back-end is still experimental, but is fairly usable for compute kernels
1062 in LLVM 3.0. Most scalar arithmetic is implemented, as well as intrinsics to
1063 access the special PTX registers and sync instructions. The major missing
1064 pieces are texture/sampler support and some vector operations.</p>
1066 <p>That said, the backend is already being used for domain-specific languages
1067 and can be used by Clang to
1068 <a href="http://clang.llvm.org/docs/ReleaseNotes.html#opencl">compile OpenCL
1069 C code</a> into PTX.</p>
1073 <!--=========================================================================-->
1075 <a name="OtherTS">Other Target Specific Improvements</a>
1081 <li>Many PowerPC improvements have been implemented for ELF targets, including
1082 support for varargs and initial support for direct .o file emission.</li>
1084 <li>MicroBlaze scheduling itineraries were added that model the
1085 3-stage and the 5-stage pipeline architectures. The 3-stage
1086 pipeline model can be selected with <code>-mcpu=mblaze3</code>
1087 and the 5-stage pipeline model can be selected with
1088 <code>-mcpu=mblaze5</code>.</li>
1094 <!--=========================================================================-->
1096 <a name="changes">Major Changes and Removed Features</a>
1101 <p>If you're already an LLVM user or developer with out-of-tree changes based on
1102 LLVM 2.9, this section lists some "gotchas" that you may run into upgrading
1103 from the previous release.</p>
1106 <li>LLVM 3.0 removes support for reading LLVM 2.8 and earlier files, and LLVM
1107 3.1 will eliminate support for reading LLVM 2.9 files. Going forward, we
1108 aim for all future versions of LLVM to read bitcode files and .ll files
1109 produced by LLVM 3.0.</li>
1110 <li>Tablegen has been split into a library, allowing the clang tblgen pieces
1111 to now live in the clang tree. The llvm version has been renamed to
1112 llvm-tblgen instead of tblgen.</li>
1113 <li>The <code>LLVMC</code> meta compiler driver was removed.</li>
1114 <li>The unused PostOrder Dominator Frontiers and LowerSetJmp passes were removed.</li>
1117 <li>The old <code>TailDup</code> pass was not used in the standard pipeline
1118 and was unable to update ssa form, so it has been removed.
1119 <li>The syntax of volatile loads and stores in IR has been changed to
1120 "<code>load volatile</code>"/"<code>store volatile</code>". The old
1121 syntax ("<code>volatile load</code>"/"<code>volatile store</code>")
1122 is still accepted, but is now considered deprecated and will be removed in
1124 <li>llvm-gcc's frontend tests have been removed from llvm/test/Frontend*, sunk
1125 into the clang and dragonegg testsuites.</li>
1126 <li>The old atomic intrinsics (<code>llvm.memory.barrier</code> and
1127 <code>llvm.atomic.*</code>) are now gone. Please use the new atomic
1128 instructions, described in the <a href="Atomics.html">atomics guide</a>.
1129 <li>LLVM's configure script doesn't depend on llvm-gcc anymore, eliminating a
1130 strange circular dependence between projects.</li>
1133 <h4>Windows (32-bit)</h4>
1137 <li>On Win32(MinGW32 and MSVC), Windows 2000 will not be supported.
1138 Windows XP or higher is required.</li>
1145 <!--=========================================================================-->
1147 <a name="api_changes">Internal API Changes</a>
1152 <p>In addition, many APIs have changed in this release. Some of the major
1153 LLVM API changes are:</p>
1156 <li>The biggest and most pervasive change is that the type system has been
1157 rewritten: <code>PATypeHolder</code> and <code>OpaqueType</code> are gone,
1158 and all APIs deal with <code>Type*</code> instead of <code>const
1159 Type*</code>. If you need to create recursive structures, then create a
1160 named structure, and use <code>setBody()</code> when all its elements are
1161 built. Type merging and refining is gone too: named structures are not
1162 merged with other structures, even if their layout is identical. (of
1163 course anonymous structures are still uniqued by layout).</li>
1165 <li><code>PHINode::reserveOperandSpace</code> has been removed. Instead, you
1166 must specify how many operands to reserve space for when you create the
1167 PHINode, by passing an extra argument
1168 into <code>PHINode::Create</code>.</li>
1170 <li>PHINodes no longer store their incoming BasicBlocks as operands. Instead,
1171 the list of incoming BasicBlocks is stored separately, and can be accessed
1172 with new functions <code>PHINode::block_begin</code>
1173 and <code>PHINode::block_end</code>.</li>
1175 <li>Various functions now take an <code>ArrayRef</code> instead of either a
1176 pair of pointers (or iterators) to the beginning and end of a range, or a
1177 pointer and a length. Others now return an <code>ArrayRef</code> instead
1178 of a reference to a <code>SmallVector</code>
1179 or <code>std::vector</code>. These include:
1181 <!-- Please keep this list sorted. -->
1182 <li><code>CallInst::Create</code></li>
1183 <li><code>ComputeLinearIndex</code> (in <code>llvm/CodeGen/Analysis.h</code>)</li>
1184 <li><code>ConstantArray::get</code></li>
1185 <li><code>ConstantExpr::getExtractElement</code></li>
1186 <li><code>ConstantExpr::getGetElementPtr</code></li>
1187 <li><code>ConstantExpr::getInBoundsGetElementPtr</code></li>
1188 <li><code>ConstantExpr::getIndices</code></li>
1189 <li><code>ConstantExpr::getInsertElement</code></li>
1190 <li><code>ConstantExpr::getWithOperands</code></li>
1191 <li><code>ConstantFoldCall</code> (in <code>llvm/Analysis/ConstantFolding.h</code>)</li>
1192 <li><code>ConstantFoldInstOperands</code> (in <code>llvm/Analysis/ConstantFolding.h</code>)</li>
1193 <li><code>ConstantVector::get</code></li>
1194 <li><code>DIBuilder::createComplexVariable</code></li>
1195 <li><code>DIBuilder::getOrCreateArray</code></li>
1196 <li><code>ExtractValueInst::Create</code></li>
1197 <li><code>ExtractValueInst::getIndexedType</code></li>
1198 <li><code>ExtractValueInst::getIndices</code></li>
1199 <li><code>FindInsertedValue</code> (in <code>llvm/Analysis/ValueTracking.h</code>)</li>
1200 <li><code>gep_type_begin</code> (in <code>llvm/Support/GetElementPtrTypeIterator.h</code>)</li>
1201 <li><code>gep_type_end</code> (in <code>llvm/Support/GetElementPtrTypeIterator.h</code>)</li>
1202 <li><code>GetElementPtrInst::Create</code></li>
1203 <li><code>GetElementPtrInst::CreateInBounds</code></li>
1204 <li><code>GetElementPtrInst::getIndexedType</code></li>
1205 <li><code>InsertValueInst::Create</code></li>
1206 <li><code>InsertValueInst::getIndices</code></li>
1207 <li><code>InvokeInst::Create</code></li>
1208 <li><code>IRBuilder::CreateCall</code></li>
1209 <li><code>IRBuilder::CreateExtractValue</code></li>
1210 <li><code>IRBuilder::CreateGEP</code></li>
1211 <li><code>IRBuilder::CreateInBoundsGEP</code></li>
1212 <li><code>IRBuilder::CreateInsertValue</code></li>
1213 <li><code>IRBuilder::CreateInvoke</code></li>
1214 <li><code>MDNode::get</code></li>
1215 <li><code>MDNode::getIfExists</code></li>
1216 <li><code>MDNode::getTemporary</code></li>
1217 <li><code>MDNode::getWhenValsUnresolved</code></li>
1218 <li><code>SimplifyGEPInst</code> (in <code>llvm/Analysis/InstructionSimplify.h</code>)</li>
1219 <li><code>TargetData::getIndexedOffset</code></li>
1222 <li>All forms of <code>StringMap::getOrCreateValue</code> have been remove
1223 except for the one which takes a <code>StringRef</code>.</li>
1225 <li>The <code>LLVMBuildUnwind</code> function from the C API was removed. The
1226 LLVM <code>unwind</code> instruction has been deprecated for a long time
1227 and isn't used by the current front-ends. So this was removed during the
1228 exception handling rewrite.</li>
1230 <li>The <code>LLVMAddLowerSetJmpPass</code> function from the C API was
1231 removed because the <code>LowerSetJmp</code> pass was removed.</li>
1233 <li>The <code>DIBuilder</code> interface used by front ends to encode
1234 debugging information in the LLVM IR now expects clients to
1235 use <code>DIBuilder::finalize()</code> at the end of translation unit to
1236 complete debugging information encoding.</li>
1238 <li>TargetSelect.h moved to Support/ from Target/</li>
1240 <li>UpgradeIntrinsicCall no longer upgrades pre-2.9 intrinsic calls (for
1241 example <code>llvm.memset.i32</code>).</li>
1243 <li>It is mandatory to initialize all out-of-tree passes too and their dependencies now with
1244 <code>INITIALIZE_PASS{BEGIN,END,}</code>
1245 and <code>INITIALIZE_{PASS,AG}_DEPENDENCY</code>.</li>
1247 <li>The interface for MemDepResult in MemoryDependenceAnalysis has been
1248 enhanced with new return types Unknown and NonFuncLocal, in addition to
1249 the existing types Clobber, Def, and NonLocal.</li>
1256 <!-- *********************************************************************** -->
1258 <a name="knownproblems">Known Problems</a>
1260 <!-- *********************************************************************** -->
1264 <p>LLVM is generally a production quality compiler, and is used by a broad range
1265 of applications and shipping in many products. That said, not every
1266 subsystem is as mature as the aggregate, particularly the more obscure
1267 targets. If you run into a problem, please check the <a
1268 href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
1269 there isn't already one or ask on the <a
1270 href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev
1273 <p>Known problem areas include:</p>
1276 <li>The Alpha, Blackfin, CellSPU, MSP430, PTX, SystemZ and
1277 XCore backends are experimental, and the Alpha, Blackfin and SystemZ
1278 targets have already been removed from mainline.</li>
1280 <li>The integrated assembler, disassembler, and JIT is not supported by
1281 several targets. If an integrated assembler is not supported, then a
1282 system assembler is required. For more details, see the <a
1283 href="CodeGenerator.html#targetfeatures">Target Features Matrix</a>.
1286 <li>The C backend has numerous problems and is not being actively maintained.
1287 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: $Date$