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 and significant known problems.
49 All LLVM releases may be downloaded from
50 the <a href="http://llvm.org/releases/">LLVM releases web site</a>.</p>
52 <p>For more information about LLVM, including information about the latest
53 release, please check out the <a href="http://llvm.org/">main LLVM web
54 site</a>. If you have questions or comments,
55 the <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM
56 Developer's Mailing List</a> is a good place to send them.</p>
58 <p>Note that if you are reading this file from a Subversion checkout or the main
59 LLVM web page, this document applies to the <i>next</i> release, not the
60 current one. To see the release notes for a specific release, please see the
61 <a href="http://llvm.org/releases/">releases page</a>.</p>
65 <!-- Features that need text if they're finished for 3.1:
69 loop dependence analysis
70 CorrelatedValuePropagation
71 lib/Transforms/IPO/MergeFunctions.cpp => consider for 3.1.
74 <!-- *********************************************************************** -->
76 <a name="subproj">Sub-project Status Update</a>
78 <!-- *********************************************************************** -->
82 <p>The LLVM 3.0 distribution currently consists of code from the core LLVM
83 repository (which roughly includes the LLVM optimizers, code generators and
84 supporting tools), and the Clang repository. In
85 addition to this code, the LLVM Project includes other sub-projects that are
86 in development. Here we include updates on these subprojects.</p>
88 <!--=========================================================================-->
90 <a name="clang">Clang: C/C++/Objective-C Frontend Toolkit</a>
95 <p><a href="http://clang.llvm.org/">Clang</a> is an LLVM front end for the C,
96 C++, and Objective-C languages. Clang aims to provide a better user
97 experience through expressive diagnostics, a high level of conformance to
98 language standards, fast compilation, and low memory use. Like LLVM, Clang
99 provides a modular, library-based architecture that makes it suitable for
100 creating or integrating with other development tools. Clang is considered a
101 production-quality compiler for C, Objective-C, C++ and Objective-C++ on x86
102 (32- and 64-bit), and for darwin/arm targets.</p>
104 <p>In the LLVM 3.0 time-frame, the Clang team has made many improvements:</p>
107 <li>Greatly improved support for building C++ applications, with greater
108 stability and better diagnostics.</li>
110 <li><a href="http://clang.llvm.org/cxx_status.html">Improved support</a> for
111 the <a href="http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_detail.htm?csnumber=50372">C++
112 2011</a> standard, including implementations of non-static data member
113 initializers, alias templates, delegating constructors, the range-based
114 for loop, and implicitly-generated move constructors and move assignment
115 operators, among others.</li>
117 <li>Implemented support for some features of the upcoming C1x standard,
118 including static assertions and generic selections.</li>
120 <li>Better detection of include and linking paths for system headers and
121 libraries, especially for Linux distributions.</li>
123 <li>Implemented support
124 for <a href="http://clang.llvm.org/docs/AutomaticReferenceCounting.html">Automatic
125 Reference Counting</a> for Objective-C.</li>
127 <li>Implemented a number of optimizations in <tt>libclang</tt>, the Clang C
128 interface, to improve the performance of code completion and the mapping
129 from source locations to abstract syntax tree nodes.</li>
133 <p>If Clang rejects your code but another compiler accepts it, please take a
134 look at the <a href="http://clang.llvm.org/compatibility.html">language
135 compatibility</a> guide to make sure this is not intentional or a known
140 <!--=========================================================================-->
142 <a name="dragonegg">DragonEgg: GCC front-ends, LLVM back-end</a>
146 <p><a href="http://dragonegg.llvm.org/">DragonEgg</a> is a
147 <a href="http://gcc.gnu.org/wiki/plugins">gcc plugin</a> that replaces GCC's
148 optimizers and code generators with LLVM's. It works with gcc-4.5 or gcc-4.6,
149 targets the x86-32 and x86-64 processor families, and has been successfully
150 used on the Darwin, FreeBSD, KFreeBSD, Linux and OpenBSD platforms. It fully
151 supports Ada, C, C++ and Fortran. It has partial support for Go, Java, Obj-C
154 <p>The 3.0 release has the following notable changes:</p>
156 <li>GCC version 4.6 is now fully supported.</li>
158 <li>Patching and building GCC is no longer required: the plugin should work
159 with your system GCC (version 4.5 or 4.6; on Debian/Ubuntu systems the
160 gcc-4.5-plugin-dev or gcc-4.6-plugin-dev package is also needed).</li>
162 <li>The <tt>-fplugin-arg-dragonegg-enable-gcc-optzns</tt> option, which runs
163 GCC's optimizers as well as LLVM's, now works much better. This is the
164 option to use if you want ultimate performance! It not yet completely
165 stable: it may cause the plugin to crash.</li>
167 <li>The type and constant conversion logic has been almost entirely rewritten,
168 fixing a multitude of obscure bugs.</li>
178 <!--=========================================================================-->
180 <a name="compiler-rt">compiler-rt: Compiler Runtime Library</a>
185 <p>The new LLVM <a href="http://compiler-rt.llvm.org/">compiler-rt project</a>
186 is a simple library that provides an implementation of the low-level
187 target-specific hooks required by code generation and other runtime
188 components. For example, when compiling for a 32-bit target, converting a
189 double to a 64-bit unsigned integer is compiled into a runtime call to the
190 "__fixunsdfdi" function. The compiler-rt library provides highly optimized
191 implementations of this and other low-level routines (some are 3x faster than
192 the equivalent libgcc routines).</p>
194 <p>In the LLVM 3.0 timeframe,</p>
198 <!--=========================================================================-->
200 <a name="lldb">LLDB: Low Level Debugger</a>
205 <p>LLDB has advanced by leaps and bounds in the 3.0 timeframe. It is
206 dramatically more stable and useful, and includes both a
207 new <a href="http://lldb.llvm.org/tutorial.html">tutorial</a> and
208 a <a href="http://lldb.llvm.org/lldb-gdb.html">side-by-side comparison with
213 <!--=========================================================================-->
215 <a name="libc++">libc++: C++ Standard Library</a>
220 <p>Like compiler_rt, libc++ is now <a href="DeveloperPolicy.html#license">dual
221 licensed</a> under the MIT and UIUC license, allowing it to be used more
227 <!--=========================================================================-->
229 <a name="LLBrowse">LLBrowse: IR Browser</a>
234 <p><a href="http://llvm.org/svn/llvm-project/llbrowse/trunk/doc/LLBrowse.html">
235 LLBrowse</a> is an interactive viewer for LLVM modules. It can load any LLVM
236 module and displays its contents as an expandable tree view, facilitating an
237 easy way to inspect types, functions, global variables, or metadata nodes. It
238 is fully cross-platform, being based on the popular wxWidgets GUI
243 <!--=========================================================================-->
245 <a name="vmkit">VMKit</a>
250 <p>The <a href="http://vmkit.llvm.org/">VMKit project</a> is an
251 implementation of a Java Virtual Machine (Java VM or JVM) that uses LLVM for
252 static and just-in-time compilation.
254 <p>In the LLVM 3.0 time-frame, VMKit has had significant improvements on both
255 runtime and startup performance:</p>
258 <li>Precompilation: by compiling ahead of time a small subset of Java's core
259 library, the startup performance have been highly optimized to the point that
260 running a 'Hello World' program takes less than 30 milliseconds.</li>
262 <li>Customization: by customizing virtual methods for individual classes,
263 the VM can statically determine the target of a virtual call, and decide to
266 <li>Inlining: the VM does more inlining than it did before, by allowing more
267 bytecode instructions to be inlined, and thanks to customization. It also
268 inlines GC barriers, and object allocations.</li>
270 <li>New exception model: the generated code for a method that does not do
271 any try/catch is not penalized anymore by the eventuality of calling a
272 method that throws an exception. Instead, the method that throws the
273 exception jumps directly to the method that could catch it.</li>
279 <!--=========================================================================-->
282 <a name="klee">KLEE: A Symbolic Execution Virtual Machine</a>
287 <a href="http://klee.llvm.org/">KLEE</a> is a symbolic execution framework for
288 programs in LLVM bitcode form. KLEE tries to symbolically evaluate "all" paths
289 through the application and records state transitions that lead to fault
290 states. This allows it to construct testcases that lead to faults and can even
291 be used to verify some algorithms.
299 <!-- *********************************************************************** -->
301 <a name="externalproj">External Open Source Projects Using LLVM 3.0</a>
303 <!-- *********************************************************************** -->
307 <p>An exciting aspect of LLVM is that it is used as an enabling technology for
308 a lot of other language and tools projects. This section lists some of the
309 projects that have already been updated to work with LLVM 3.0.</p>
311 <!--=========================================================================-->
312 <h3>AddressSanitizer</h3>
316 <p><a href="http://code.google.com/p/address-sanitizer/">AddressSanitizer</a>
317 uses compiler instrumentation and a specialized malloc library to find C/C++
318 bugs such as use-after-free and out-of-bound accesses to heap, stack, and
319 globals. The key feature of the tool is speed: the average slowdown
320 introduced by AddressSanitizer is less than 2x.</p>
324 <!--=========================================================================-->
329 <p><a href="http://www.clamav.net">Clam AntiVirus</a> is an open source (GPL)
330 anti-virus toolkit for UNIX, designed especially for e-mail scanning on mail
333 <p>Since version 0.96 it
334 has <a href="http://vrt-sourcefire.blogspot.com/2010/09/introduction-to-clamavs-low-level.html">bytecode
335 signatures</a> that allow writing detections for complex malware.</p>
337 <p>It uses LLVM's JIT to speed up the execution of bytecode on X86, X86-64,
338 PPC32/64, falling back to its own interpreter otherwise. The git version was
339 updated to work with LLVM 3.0.</p>
343 <!--=========================================================================-->
344 <h3>clang_complete for VIM</h3>
348 <p><a href="https://github.com/Rip-Rip/clang_complete">clang_complete</a> is a
349 VIM plugin, that provides accurate C/C++ autocompletion using the clang front
350 end. The development version of clang complete, can directly use libclang
351 which can maintain a cache to speed up auto completion.</p>
355 <!--=========================================================================-->
360 <p><a href="https://bitbucket.org/dwilliamson/clreflect">clReflect</a> is a C++
361 parser that uses clang/LLVM to derive a light-weight reflection database
362 suitable for use in game development. It comes with a very simple runtime
363 library for loading and querying the database, requiring no external
364 dependencies (including CRT), and an additional utility library for object
365 management and serialisation.</p>
369 <!--=========================================================================-->
370 <h3>Cling C++ Interpreter</h3>
374 <p><a href="http://cern.ch/cling">Cling</a> is an interactive compiler interface
375 (aka C++ interpreter). It uses LLVM's JIT and clang; it currently supports
376 C++ and C. It has a prompt interface, runs source files, calls into shared
377 libraries, prints the value of expressions, even does runtime lookup of
378 identifiers (dynamic scopes). And it just behaves like one would expect from
383 <!--=========================================================================-->
384 <h3>Crack Programming Language</h3>
388 <p><a href="http://code.google.com/p/crack-language/">Crack</a> aims to provide
389 the ease of development of a scripting language with the performance of a
390 compiled language. The language derives concepts from C++, Java and Python,
391 incorporating object-oriented programming, operator overloading and strong
396 <!--=========================================================================-->
401 <p><a href="http://eerolanguage.org/">Eero</a> is a fully
402 header-and-binary-compatible dialect of Objective-C 2.0, implemented with a
403 patched version of the Clang/LLVM compiler. It features a streamlined syntax,
404 Python-like indentation, and new operators, for improved readability and
405 reduced code clutter. It also has new features such as limited forms of
406 operator overloading and namespaces, and strict (type-and-operator-safe)
407 enumerations. It is inspired by languages such as Smalltalk, Python, and
412 <!--=========================================================================-->
413 <h3>Glasgow Haskell Compiler (GHC)</h3>
417 <p>GHC is an open source, state-of-the-art programming suite for Haskell, a
418 standard lazy functional programming language. It includes an optimizing
419 static compiler generating good code for a variety of platforms, together
420 with an interactive system for convenient, quick development.</p>
422 <p>GHC 7.0 and onwards include an LLVM code generator, supporting LLVM 2.8 and
423 later. Since LLVM 2.9, GHC now includes experimental support for the ARM
424 platform with LLVM 3.0.</p>
428 <!--=========================================================================-->
433 <p><a href="http://botwars.tk/gwscript/">gwXscript</a> is an object oriented,
434 aspect oriented programming language which can create both executables (ELF,
435 EXE) and shared libraries (DLL, SO, DYNLIB). The compiler is implemented in
436 its own language and translates scripts into LLVM-IR which can be optimized
437 and translated into native code by the LLVM framework. Source code in
438 gwScript contains definitions that expand the namespaces. So you can build
439 your project and simply 'plug out' features by removing a file. The remaining
440 project does not leave scars since you directly separate concerns by the
441 'template' feature of gwX. It is also possible to add new features to a
442 project by just adding files and without editing the original project. This
443 language is used for example to create games or content management systems
444 that should be extendable.</p>
446 <p>gwXscript is strongly typed and offers comfort with its native types string,
447 hash and array. You can easily write new libraries in gwXscript or native
448 code. gwXscript is type safe and users should not be able to crash your
449 program or execute malicious code except code that is eating CPU time.</p>
453 <!--=========================================================================-->
454 <h3>include-what-you-use</h3>
458 <p><a href="http://code.google.com/p/include-what-you-use">include-what-you-use</a>
459 is a tool to ensure that a file directly <code>#include</code>s
460 all <code>.h</code> files that provide a symbol that the file uses. It also
461 removes superfluous <code>#include</code>s from source files.</p>
465 <!--=========================================================================-->
466 <h3>ispc: The Intel SPMD Program Compiler</h3>
470 <p><a href="http://ispc.github.com">ispc</a> is a compiler for "single program,
471 multiple data" (SPMD) programs. It compiles a C-based SPMD programming
472 language to run on the SIMD units of CPUs; it often delivers 5-6x speedups on
473 a single core of a CPU with an 8-wide SIMD unit compared to serial code,
474 while still providing a clean and easy-to-understand programming model. For
475 an introduction to the language and its performance,
476 see <a href="http://ispc.github.com/example.html">the walkthrough</a> of a short
477 example program. ispc is licensed under the BSD license.</p>
481 <!--=========================================================================-->
482 <h3>LanguageKit and Pragmatic Smalltalk</h3>
486 <p><a href="http://etoileos.com/etoile/features/languagekit/">LanguageKit</a> is
487 a framework for implementing dynamic languages sharing an object model with
488 Objective-C. It provides static and JIT compilation using LLVM along with
489 its own interpreter. Pragmatic Smalltalk is a dialect of Smalltalk, built on
490 top of LanguageKit, that interfaces directly with Objective-C, sharing the
491 same object representation and message sending behaviour. These projects are
492 developed as part of the Étoié desktop environment.</p>
496 <!--=========================================================================-->
501 <p><a href="http://lua-av.mat.ucsb.edu/blog/">LuaAV</a> is a real-time
502 audiovisual scripting environment based around the Lua language and a
503 collection of libraries for sound, graphics, and other media protocols. LuaAV
504 uses LLVM and Clang to JIT compile efficient user-defined audio synthesis
505 routines specified in a declarative syntax.</p>
509 <!--=========================================================================-->
514 <p>An open source, cross-platform implementation of C# and the CLR that is
515 binary compatible with Microsoft.NET. Has an optional, dynamically-loaded
516 LLVM code generation backend in Mini, the JIT compiler.</p>
518 <p>Note that we use a Git mirror of LLVM with some patches. See:
519 https://github.com/mono/llvm</p>
523 <!--=========================================================================-->
528 <p><a href="http://polly.grosser.es">Polly</a> is an advanced data-locality
529 optimizer and automatic parallelizer. It uses an advanced, mathematical
530 model to calculate detailed data dependency information which it uses to
531 optimize the loop structure of a program. Polly can speed up sequential code
532 by improving memory locality and consequently the cache use. Furthermore,
533 Polly is able to expose different kind of parallelism which it exploits by
534 introducing (basic) OpenMP and SIMD code. A mid-term goal of Polly is to
535 automatically create optimized GPU code.</p>
539 <!--=========================================================================-->
540 <h3>Portable OpenCL (pocl)</h3>
544 <p>Portable OpenCL is an open source implementation of the OpenCL standard which
545 can be easily adapted for new targets. One of the goals of the project is
546 improving performance portability of OpenCL programs, avoiding the need for
547 target-dependent manual optimizations. A "native" target is included, which
548 allows running OpenCL kernels on the host (CPU).</p>
552 <!--=========================================================================-->
556 <p><a href="http://pure-lang.googlecode.com/">Pure</a> is an
557 algebraic/functional programming language based on term rewriting. Programs
558 are collections of equations which are used to evaluate expressions in a
559 symbolic fashion. The interpreter uses LLVM as a backend to JIT-compile Pure
560 programs to fast native code. Pure offers dynamic typing, eager and lazy
561 evaluation, lexical closures, a hygienic macro system (also based on term
562 rewriting), built-in list and matrix support (including list and matrix
563 comprehensions) and an easy-to-use interface to C and other programming
564 languages (including the ability to load LLVM bitcode modules, and inline C,
565 C++, Fortran and Faust code in Pure programs if the corresponding LLVM-enabled
566 compilers are installed).</p>
568 <p>Pure version 0.48 has been tested and is known to work with LLVM 3.0
569 (and continues to work with older LLVM releases >= 2.5).</p>
573 <!--=========================================================================-->
574 <h3>Renderscript</h3>
578 <p><a href="http://developer.android.com/guide/topics/renderscript/index.html">Renderscript</a>
579 is Android's advanced 3D graphics rendering and compute API. It provides a
580 portable C99-based language with extensions to facilitate common use cases
581 for enhancing graphics and thread level parallelism. The Renderscript
582 compiler frontend is based on Clang/LLVM. It emits a portable bitcode format
583 for the actual compiled script code, as well as reflects a Java interface for
584 developers to control the execution of the compiled bitcode. Executable
585 machine code is then generated from this bitcode by an LLVM backend on the
586 device. Renderscript is thus able to provide a mechanism by which Android
587 developers can improve performance of their applications while retaining
592 <!--=========================================================================-->
597 <p><a href="http://safecode.cs.illinois.edu">SAFECode</a> is a memory safe C/C++
598 compiler built using LLVM. It takes standard, unannotated C/C++ code,
599 analyzes the code to ensure that memory accesses and array indexing
600 operations are safe, and instruments the code with run-time checks when
601 safety cannot be proven statically. SAFECode can be used as a debugging aid
602 (like Valgrind) to find and repair memory safety bugs. It can also be used
603 to protect code from security attacks at run-time.</p>
607 <!--=========================================================================-->
608 <h3>The Stupid D Compiler (SDC)</h3>
612 <p><a href="https://github.com/bhelyer/SDC">The Stupid D Compiler</a> is a
613 project seeking to write a self-hosting compiler for the D programming
614 language without using the frontend of the reference compiler (DMD).</p>
618 <!--=========================================================================-->
619 <h3>TTA-based Co-design Environment (TCE)</h3>
623 <p>TCE is a toolset for designing application-specific processors (ASP) based on
624 the Transport triggered architecture (TTA). The toolset provides a complete
625 co-design flow from C/C++ programs down to synthesizable VHDL and parallel
626 program binaries. Processor customization points include the register files,
627 function units, supported operations, and the interconnection network.</p>
629 <p>TCE uses Clang and LLVM for C/C++ language support, target independent
630 optimizations and also for parts of code generation. It generates new
631 LLVM-based code generators "on the fly" for the designed TTA processors and
632 loads them in to the compiler backend as runtime libraries to avoid
633 per-target recompilation of larger parts of the compiler chain.</p>
637 <!--=========================================================================-->
638 <h3>Tart Programming Language</h3>
642 <p><a href="http://code.google.com/p/tart/">Tart</a> is a general-purpose,
643 strongly typed programming language designed for application
644 developers. Strongly inspired by Python and C#, Tart focuses on practical
645 solutions for the professional software developer, while avoiding the clutter
646 and boilerplate of legacy languages like Java and C++. Although Tart is still
647 in development, the current implementation supports many features expected of
648 a modern programming language, such as garbage collection, powerful
649 bidirectional type inference, a greatly simplified syntax for template
650 metaprogramming, closures and function literals, reflection, operator
651 overloading, explicit mutability and immutability, and much more. Tart is
652 flexible enough to accommodate a broad range of programming styles and
653 philosophies, while maintaining a strong commitment to simplicity, minimalism
654 and elegance in design.</p>
658 <!--=========================================================================-->
659 <h3>ThreadSanitizer</h3>
663 <p><a href="http://code.google.com/p/data-race-test/">ThreadSanitizer</a> is a
664 data race detector for (mostly) C and C++ code, available for Linux, Mac OS
665 and Windows. On different systems, we use binary instrumentation frameworks
666 (Valgrind and Pin) as frontends that generate the program events for the race
667 detection algorithm. On Linux, there's an option of using LLVM-based
668 compile-time instrumentation.</p>
672 <!--=========================================================================-->
673 <h3>The ZooLib C++ Cross-Platform Application Framework</h3>
677 <p><a href="http://www.zoolib.org/">ZooLib</a> is Open Source under the MIT
678 License. It provides GUI, filesystem access, TCP networking, thread-safe
679 memory management, threading and locking for Mac OS X, Classic Mac OS,
680 Microsoft Windows, POSIX operating systems with X11, BeOS, Haiku, Apple's iOS
681 and Research in Motion's BlackBerry.</p>
683 <p>My current work is to use CLang's static analyzer to improve ZooLib's code
684 quality. I also plan to set up LLVM compiles of the demo programs and test
685 programs using CLang and LLVM on all the platforms that CLang, LLVM and
686 ZooLib all support.</p>
692 <!-- *********************************************************************** -->
694 <a name="whatsnew">What's New in LLVM 3.0?</a>
696 <!-- *********************************************************************** -->
700 <p>This release includes a huge number of bug fixes, performance tweaks and
701 minor improvements. Some of the major improvements and new features are
702 listed in this section.</p>
704 <!--=========================================================================-->
706 <a name="majorfeatures">Major New Features</a>
711 <p><b>llvm-gcc is gone</b></p>
713 <p>LLVM 3.0 includes several major new capabilities:</p>
716 Analysis/RegionInfo.h + Dom Frontiers
717 SparseBitVector: used in LiveVar.
723 Better performance for Neon code in clang due to SRoA improvements.
724 New regalloc on by default. Lin scan going away in 3.1
725 PGO / builtin_expect improvements (summary needed)
727 AVX support, assembler, compiler and disassembler.
728 IndVar improvements: andy
729 PTX backend improvements: Justin
730 llvm-rtdyld & MC JIT: JimG
731 InstAliases now automatically used in the asmprinter where they are shorter.
732 Integrated assembler on by default for arm/thumb?
733 PostOrder Dominator frontiers were removed.
734 Line Profiling / gcov support
735 EH and debug information produced with CFI directives, yielding smaller executables: http://blog.mozilla.com/respindola/2011/05/12/cfi-directives/
736 X86-64 generates smaller and faster code at -O0 (fast isel improvements)
737 Better code generation for Cortex-A9
738 Many APIs take ArrayRef's now.
739 Pass manager extension API.
753 <!--=========================================================================-->
755 <a name="coreimprovements">LLVM IR and Core Improvements</a>
760 <p>LLVM IR has several new features for better support of new targets and that
761 expose new optimization opportunities:</p>
763 <p>One of the biggest changes is that 3.0 has a new exception handling
764 system. The old system used LLVM intrinsics to convey the exception handling
765 information to the code generator. It worked in most cases, but not
766 all. Inlining was especially difficult to get right. Also, the intrinsics
767 could be moved away from the <code>invoke</code> instruction, making it hard
768 to recover that information.</p>
770 <p>The new EH system makes exception handling a first-class member of the IR. It
771 adds two new instructions:</p>
774 <li><a href="LangRef.html#i_landingpad"><code>landingpad</code></a> —
775 this instruction defines a landing pad basic block. It contains all of the
776 information that's needed by the code generator. It's also required to be
777 the first non-PHI instruction in the landing pad. In addition, a landing
778 pad may be jumped to only by the unwind edge of an <code>invoke</code>
781 <li><a href="LangRef.html#i_resume"><code>resume</code></a> — this
782 instruction causes the current exception to resume traveling up the
783 stack. It replaces the <code>@llvm.eh.resume</code> intrinsic.</li>
786 <p>Converting from the old EH API to the new EH API is rather simple, because a
787 lot of complexity has been removed. The two intrinsics,
788 <code>@llvm.eh.exception</code> and <code>@llvm.eh.selector</code> have been
789 superceded by the <code>landingpad</code> instruction. Instead of generating
790 a call to <code>@llvm.eh.exception</code> and <code>@llvm.eh.selector</code>:
792 <div class="doc_code">
794 Function *ExcIntr = Intrinsic::getDeclaration(TheModule,
795 Intrinsic::eh_exception);
796 Function *SlctrIntr = Intrinsic::getDeclaration(TheModule,
797 Intrinsic::eh_selector);
799 // The exception pointer.
800 Value *ExnPtr = Builder.CreateCall(ExcIntr, "exc_ptr");
802 std::vector<Value*> Args;
803 Args.push_back(ExnPtr);
804 Args.push_back(Builder.CreateBitCast(Personality,
805 Type::getInt8PtrTy(Context)));
807 <i>// Add selector clauses to Args.</i>
809 // The selector call.
810 Builder.CreateCall(SlctrIntr, Args, "exc_sel");
814 <p>You should instead generate a <code>landingpad</code> instruction, that
815 returns an exception object and selector value:</p>
817 <div class="doc_code">
819 LandingPadInst *LPadInst =
820 Builder.CreateLandingPad(StructType::get(Int8PtrTy, Int32Ty, NULL),
823 Value *LPadExn = Builder.CreateExtractValue(LPadInst, 0);
824 Builder.CreateStore(LPadExn, getExceptionSlot());
826 Value *LPadSel = Builder.CreateExtractValue(LPadInst, 1);
827 Builder.CreateStore(LPadSel, getEHSelectorSlot());
831 <p>It's now trivial to add the individual clauses to the <code>landingpad</code>
834 <div class="doc_code">
836 <i><b>// Adding a catch clause</b></i>
837 Constant *TypeInfo = getTypeInfo();
838 LPadInst->addClause(TypeInfo);
840 <i><b>// Adding a C++ catch-all</b></i>
841 LPadInst->addClause(Constant::getNullValue(Builder.getInt8PtrTy()));
843 <i><b>// Adding a cleanup</b></i>
844 LPadInst->setCleanup(true);
846 <i><b>// Adding a filter clause</b></i>
847 std::vector<Constant*> TypeInfos;
848 Constant *TypeInfo = getFilterTypeInfo();
849 TypeInfos.push_back(Builder.CreateBitCast(TypeInfo, Builder.getInt8PtrTy()));
851 ArrayType *FilterTy = ArrayType::get(Int8PtrTy, TypeInfos.size());
852 LPadInst->addClause(ConstantArray::get(FilterTy, TypeInfos));
856 <p>Converting from using the <code>@llvm.eh.resume</code> intrinsic to
857 the <code>resume</code> instruction is trivial. It takes the exception
858 pointer and exception selector values returned by
859 the <code>landingpad</code> instruction:</p>
861 <div class="doc_code">
863 Type *UnwindDataTy = StructType::get(Builder.getInt8PtrTy(),
864 Builder.getInt32Ty(), NULL);
865 Value *UnwindData = UndefValue::get(UnwindDataTy);
866 Value *ExcPtr = Builder.CreateLoad(getExceptionObjSlot());
867 Value *ExcSel = Builder.CreateLoad(getExceptionSelSlot());
868 UnwindData = Builder.CreateInsertValue(UnwindData, ExcPtr, 0, "exc_ptr");
869 UnwindData = Builder.CreateInsertValue(UnwindData, ExcSel, 1, "exc_sel");
870 Builder.CreateResume(UnwindData);
876 <!--=========================================================================-->
878 <a name="loopoptimization">Loop Optimization Improvements</a>
882 <p>The induction variable simplification pass in 3.0 only modifies
883 induction variables when profitable. Sign and zero extension
884 elimination, linear function test replacement, loop unrolling, and
885 other simplifications that require induction variable analysis have
886 been generalized so they no longer require loops to be rewritten in a
887 typically suboptimal form prior to optimization. This new design
888 preserves more IR level information, avoids undoing earlier loop
889 optimizations (particularly hand-optimized loops), and no longer
890 strongly depends on the code generator rewriting loops a second time
891 in a now optimal form--an intractable problem.</p>
893 <p>The original behavior can be restored with -mllvm -enable-iv-rewrite;
894 however, support for this mode will be short lived. As such, bug
895 reports should be filed for any significant performance regressions
896 when moving from -mllvm -enable-iv-rewrite to the 3.0 default mode.</p>
899 <!--=========================================================================-->
901 <a name="optimizer">Optimizer Improvements</a>
906 <p>In addition to a large array of minor performance tweaks and bug fixes, this
907 release includes a few major enhancements and additions to the
920 <!--=========================================================================-->
922 <a name="mc">MC Level Improvements</a>
927 <p>The LLVM Machine Code (aka MC) subsystem was created to solve a number of
928 problems in the realm of assembly, disassembly, object file format handling,
929 and a number of other related areas that CPU instruction-set level tools work
938 <p>For more information, please see
939 the <a href="http://blog.llvm.org/2010/04/intro-to-llvm-mc-project.html">Intro
940 to the LLVM MC Project Blog Post</a>.</p>
944 <!--=========================================================================-->
946 <a name="codegen">Target Independent Code Generator Improvements</a>
951 <p>We have put a significant amount of work into the code generator
952 infrastructure, which allows us to implement more aggressive algorithms and
953 make it run faster:</p>
962 <!--=========================================================================-->
964 <a name="x86">X86-32 and X86-64 Target Improvements</a>
969 <p>New features and major changes in the X86 target include:</p>
973 <li>The X86 backend now supports
974 all <a href="http://llvm.org/PR879">inline assembly that uses the X86
975 floating point stack</a>.</li>
977 <li>The CRC32 intrinsics have been renamed. The intrinsics were previously
978 <code>@llvm.x86.sse42.crc32.[8|16|32]</code>
979 and <code>@llvm.x86.sse42.crc64.[8|64]</code>. They have been renamed to
980 <code>@llvm.x86.sse42.crc32.32.[8|16|32]</code> and
981 <code>@llvm.x86.sse42.crc32.64.[8|64]</code>.</li>
987 <!--=========================================================================-->
989 <a name="ARM">ARM Target Improvements</a>
994 <p>New features of the ARM target include:</p>
1004 <!--=========================================================================-->
1006 <a name="MIPS">MIPS Target Improvements</a>
1011 <p>New features and major changes in the MIPS target include:</p>
1014 <li>Most MIPS32r1 and r2 instructions are now supported.</li>
1015 <li>LE/BE MIPS32r1/r2 has been tested extensively.</li>
1016 <li>O32 ABI has been fully tested.</li>
1017 <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>
1018 <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>
1019 <li>Support for old-style JIT is complete.</li>
1020 <li>Support for old architectures (MIPS1 and MIPS2) has been removed.</li>
1021 <li>Initial support for MIPS64 has been added.</li>
1025 <!--=========================================================================-->
1027 <a name="PTX">PTX Target Improvements</a>
1033 The PTX back-end is still experimental, but is fairly usable for compute kernels
1034 in LLVM 3.0. Most scalar arithmetic is implemented, as well as intrinsics to
1035 access the special PTX registers and sync instructions. The major missing
1036 pieces are texture/sampler support and some vector operations.</p>
1038 <p>That said, the backend is already being used for domain-specific languages
1039 and works well with the <a href="http://www.pcc.me.uk/~peter/libclc/">libclc
1040 library</a> to supply OpenCL built-ins. With it, you can use Clang to compile
1041 OpenCL code into PTX and execute it by loading the resulting PTX as a binary
1042 blob using the nVidia OpenCL library. It has been tested with several OpenCL
1043 programs, including some from the nVidia GPU Computing SDK, and the performance
1044 is on par with the nVidia compiler.</p>
1048 <!--=========================================================================-->
1050 <a name="OtherTS">Other Target Specific Improvements</a>
1055 <p>PPC32/ELF va_arg was implemented.</p>
1056 <p>PPC32 initial support for .o file writing was implemented.</p>
1057 <p>MicroBlaze scheduling itineraries were added that model the
1058 3-stage and the 5-stage pipeline architectures. The 3-stage
1059 pipeline model can be selected with <code>-mcpu=mblaze3</code>
1060 and the 5-stage pipeline model can be selected with
1061 <code>-mcpu=mblaze5</code>.</p>
1071 <!--=========================================================================-->
1073 <a name="changes">Major Changes and Removed Features</a>
1078 <p>If you're already an LLVM user or developer with out-of-tree changes based on
1079 LLVM 2.9, this section lists some "gotchas" that you may run into upgrading
1080 from the previous release.</p>
1083 <li>The <code>LLVMC</code> front end code was removed while separating
1084 out language independence.</li>
1085 <li>The <code>LowerSetJmp</code> pass wasn't used effectively by any
1086 target and has been removed.</li>
1087 <li>The old <code>TailDup</code> pass was not used in the standard pipeline
1088 and was unable to update ssa form, so it has been removed.
1089 <li>The syntax of volatile loads and stores in IR has been changed to
1090 "<code>load volatile</code>"/"<code>store volatile</code>". The old
1091 syntax ("<code>volatile load</code>"/"<code>volatile store</code>")
1092 is still accepted, but is now considered deprecated.</li>
1093 <li>The old atomic intrinscs (<code>llvm.memory.barrier</code> and
1094 <code>llvm.atomic.*</code>) are now gone. Please use the new atomic
1095 instructions, described in the <a href="Atomics.html">atomics guide</a>.
1098 <h4>Windows (32-bit)</h4>
1102 <li>On Win32(MinGW32 and MSVC), Windows 2000 will not be supported.
1103 Windows XP or higher is required.</li>
1110 <!--=========================================================================-->
1112 <a name="api_changes">Internal API Changes</a>
1117 <p>In addition, many APIs have changed in this release. Some of the major
1118 LLVM API changes are:</p>
1121 <li>The biggest and most pervasive change is that llvm::Type's are no longer
1122 returned or accepted as 'const' values. Instead, just pass around
1123 non-const Type's.</li>
1125 <li><code>PHINode::reserveOperandSpace</code> has been removed. Instead, you
1126 must specify how many operands to reserve space for when you create the
1127 PHINode, by passing an extra argument
1128 into <code>PHINode::Create</code>.</li>
1130 <li>PHINodes no longer store their incoming BasicBlocks as operands. Instead,
1131 the list of incoming BasicBlocks is stored separately, and can be accessed
1132 with new functions <code>PHINode::block_begin</code>
1133 and <code>PHINode::block_end</code>.</li>
1135 <li>Various functions now take an <code>ArrayRef</code> instead of either a
1136 pair of pointers (or iterators) to the beginning and end of a range, or a
1137 pointer and a length. Others now return an <code>ArrayRef</code> instead
1138 of a reference to a <code>SmallVector</code>
1139 or <code>std::vector</code>. These include:
1141 <!-- Please keep this list sorted. -->
1142 <li><code>CallInst::Create</code></li>
1143 <li><code>ComputeLinearIndex</code> (in <code>llvm/CodeGen/Analysis.h</code>)</li>
1144 <li><code>ConstantArray::get</code></li>
1145 <li><code>ConstantExpr::getExtractElement</code></li>
1146 <li><code>ConstantExpr::getGetElementPtr</code></li>
1147 <li><code>ConstantExpr::getInBoundsGetElementPtr</code></li>
1148 <li><code>ConstantExpr::getIndices</code></li>
1149 <li><code>ConstantExpr::getInsertElement</code></li>
1150 <li><code>ConstantExpr::getWithOperands</code></li>
1151 <li><code>ConstantFoldCall</code> (in <code>llvm/Analysis/ConstantFolding.h</code>)</li>
1152 <li><code>ConstantFoldInstOperands</code> (in <code>llvm/Analysis/ConstantFolding.h</code>)</li>
1153 <li><code>ConstantVector::get</code></li>
1154 <li><code>DIBuilder::createComplexVariable</code></li>
1155 <li><code>DIBuilder::getOrCreateArray</code></li>
1156 <li><code>ExtractValueInst::Create</code></li>
1157 <li><code>ExtractValueInst::getIndexedType</code></li>
1158 <li><code>ExtractValueInst::getIndices</code></li>
1159 <li><code>FindInsertedValue</code> (in <code>llvm/Analysis/ValueTracking.h</code>)</li>
1160 <li><code>gep_type_begin</code> (in <code>llvm/Support/GetElementPtrTypeIterator.h</code>)</li>
1161 <li><code>gep_type_end</code> (in <code>llvm/Support/GetElementPtrTypeIterator.h</code>)</li>
1162 <li><code>GetElementPtrInst::Create</code></li>
1163 <li><code>GetElementPtrInst::CreateInBounds</code></li>
1164 <li><code>GetElementPtrInst::getIndexedType</code></li>
1165 <li><code>InsertValueInst::Create</code></li>
1166 <li><code>InsertValueInst::getIndices</code></li>
1167 <li><code>InvokeInst::Create</code></li>
1168 <li><code>IRBuilder::CreateCall</code></li>
1169 <li><code>IRBuilder::CreateExtractValue</code></li>
1170 <li><code>IRBuilder::CreateGEP</code></li>
1171 <li><code>IRBuilder::CreateInBoundsGEP</code></li>
1172 <li><code>IRBuilder::CreateInsertValue</code></li>
1173 <li><code>IRBuilder::CreateInvoke</code></li>
1174 <li><code>MDNode::get</code></li>
1175 <li><code>MDNode::getIfExists</code></li>
1176 <li><code>MDNode::getTemporary</code></li>
1177 <li><code>MDNode::getWhenValsUnresolved</code></li>
1178 <li><code>SimplifyGEPInst</code> (in <code>llvm/Analysis/InstructionSimplify.h</code>)</li>
1179 <li><code>TargetData::getIndexedOffset</code></li>
1182 <li>All forms of <code>StringMap::getOrCreateValue</code> have been remove
1183 except for the one which takes a <code>StringRef</code>.</li>
1185 <li>The <code>LLVMBuildUnwind</code> function from the C API was removed. The
1186 LLVM <code>unwind</code> instruction has been deprecated for a long time
1187 and isn't used by the current front-ends. So this was removed during the
1188 exception handling rewrite.</li>
1190 <li>The <code>LLVMAddLowerSetJmpPass</code> function from the C API was
1191 removed because the <code>LowerSetJmp</code> pass was removed.</li>
1193 <li>The <code>DIBuilder</code> interface used by front ends to encode
1194 debugging information in the LLVM IR now expects clients to
1195 use <code>DIBuilder::finalize()</code> at the end of translation unit to
1196 complete debugging information encoding.</li>
1198 <li>The way the type system works has been
1199 rewritten: <code>PATypeHolder</code> and <code>OpaqueType</code> are gone,
1200 and all APIs deal with <code>Type*</code> instead of <code>const
1201 Type*</code>. If you need to create recursive structures, then create a
1202 named structure, and use <code>setBody()</code> when all its elements are
1203 built. Type merging and refining is gone too: named structures are not
1204 merged with other structures, even if their layout is identical. (of
1205 course anonymous structures are still uniqued by layout).</li>
1207 <li>TargetSelect.h moved to Support/ from Target/</li>
1209 <li>UpgradeIntrinsicCall no longer upgrades pre-2.9 intrinsic calls (for
1210 example <code>llvm.memset.i32</code>).</li>
1212 <li>It is mandatory to initialize all out-of-tree passes too and their dependencies now with
1213 <code>INITIALIZE_PASS{BEGIN,END,}</code>
1214 and <code>INITIALIZE_{PASS,AG}_DEPENDENCY</code>.</li>
1216 <li>The interface for MemDepResult in MemoryDependenceAnalysis has been
1217 enhanced with new return types Unknown and NonFuncLocal, in addition to
1218 the existing types Clobber, Def, and NonLocal.</li>
1225 <!-- *********************************************************************** -->
1227 <a name="knownproblems">Known Problems</a>
1229 <!-- *********************************************************************** -->
1233 <p>This section contains significant known problems with the LLVM system, listed
1234 by component. If you run into a problem, please check
1235 the <a href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
1236 there isn't already one.</p>
1238 <!-- ======================================================================= -->
1240 <a name="experimental">Experimental features included with this release</a>
1245 <p>The following components of this LLVM release are either untested, known to
1246 be broken or unreliable, or are in early development. These components
1247 should not be relied on, and bugs should not be filed against them, but they
1248 may be useful to some people. In particular, if you would like to work on
1249 one of these components, please contact us on
1250 the <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev
1254 <li>The Alpha, Blackfin, CellSPU, MicroBlaze, MSP430, MIPS, PTX, SystemZ and
1255 XCore backends are experimental.</li>
1257 <li><tt>llc</tt> "<tt>-filetype=obj</tt>" is experimental on all targets other
1258 than darwin and ELF X86 systems.</li>
1263 <!-- ======================================================================= -->
1265 <a name="x86-be">Known problems with the X86 back-end</a>
1271 <li>The X86-64 backend <a href="http://llvm.org/PR1740">does not yet support
1272 the <tt>va_arg</tt> LLVM IR instruction</a>. Currently, front-ends support
1273 variadic argument constructs on X86-64 by lowering them manually.</li>
1278 <!-- ======================================================================= -->
1280 <a name="ppc-be">Known problems with the PowerPC back-end</a>
1286 <li>The PPC32/ELF support lacks PIC support.</li>
1291 <!-- ======================================================================= -->
1293 <a name="arm-be">Known problems with the ARM back-end</a>
1299 <li>Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6
1300 processors, thumb programs can crash or produce wrong results
1301 (<a href="http://llvm.org/PR1388">PR1388</a>).</li>
1303 <li>Compilation for ARM Linux OABI (old ABI) is supported but not fully
1309 <!-- ======================================================================= -->
1311 <a name="sparc-be">Known problems with the SPARC back-end</a>
1317 <li>The SPARC backend only supports the 32-bit SPARC ABI (-m32); it does not
1318 support the 64-bit SPARC ABI (-m64).</li>
1323 <!-- ======================================================================= -->
1325 <a name="mips-be">Known problems with the MIPS back-end</a>
1331 <li>64-bit MIPS targets are not supported yet.</li>
1336 <!-- ======================================================================= -->
1338 <a name="alpha-be">Known problems with the Alpha back-end</a>
1344 <li>On 21164s, some rare FP arithmetic sequences which may trap do not have
1345 the appropriate nops inserted to ensure restartability.</li>
1350 <!-- ======================================================================= -->
1352 <a name="c-be">Known problems with the C back-end</a>
1357 <p>The C backend has numerous problems and is not being actively maintained.
1358 Depending on it for anything serious is not advised.</p>
1361 <li><a href="http://llvm.org/PR802">The C backend has only basic support for
1362 inline assembly code</a>.</li>
1364 <li><a href="http://llvm.org/PR1658">The C backend violates the ABI of common
1365 C++ programs</a>, preventing intermixing between C++ compiled by the CBE
1366 and C++ code compiled with <tt>llc</tt> or native compilers.</li>
1368 <li>The C backend does not support all exception handling constructs.</li>
1370 <li>The C backend does not support arbitrary precision integers.</li>
1377 <!-- *********************************************************************** -->
1379 <a name="additionalinfo">Additional Information</a>
1381 <!-- *********************************************************************** -->
1385 <p>A wide variety of additional information is available on
1386 the <a href="http://llvm.org/">LLVM web page</a>, in particular in
1387 the <a href="http://llvm.org/docs/">documentation</a> section. The web page
1388 also contains versions of the API documentation which is up-to-date with the
1389 Subversion version of the source code. You can access versions of these
1390 documents specific to this release by going into the "<tt>llvm/doc/</tt>"
1391 directory in the LLVM tree.</p>
1393 <p>If you have any questions or comments about LLVM, please feel free to contact
1394 us via the <a href="http://llvm.org/docs/#maillist"> mailing lists</a>.</p>
1398 <!-- *********************************************************************** -->
1402 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1403 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1404 <a href="http://validator.w3.org/check/referer"><img
1405 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1407 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
1408 Last modified: $Date$