1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
6 <link rel="stylesheet" href="llvm.css" type="text/css">
7 <title>LLVM 2.7 Release Notes</title>
11 <div class="doc_title">LLVM 2.7 Release Notes</div>
13 <img align=right src="http://llvm.org/img/DragonSmall.png"
14 width="136" height="136" alt="LLVM Dragon Logo">
17 <li><a href="#intro">Introduction</a></li>
18 <li><a href="#subproj">Sub-project Status Update</a></li>
19 <li><a href="#externalproj">External Projects Using LLVM 2.7</a></li>
20 <li><a href="#whatsnew">What's New in LLVM 2.7?</a></li>
21 <li><a href="GettingStarted.html">Installation Instructions</a></li>
22 <li><a href="#portability">Portability and Supported Platforms</a></li>
23 <li><a href="#knownproblems">Known Problems</a></li>
24 <li><a href="#additionalinfo">Additional Information</a></li>
27 <div class="doc_author">
28 <p>Written by the <a href="http://llvm.org">LLVM Team</a></p>
32 <h1 style="color:red">These are in-progress notes for the upcoming LLVM 2.8
35 <a href="http://llvm.org/releases/2.6/docs/ReleaseNotes.html">LLVM 2.7
36 Release Notes</a>.</h1>-->
38 <!-- *********************************************************************** -->
39 <div class="doc_section">
40 <a name="intro">Introduction</a>
42 <!-- *********************************************************************** -->
44 <div class="doc_text">
46 <p>This document contains the release notes for the LLVM Compiler
47 Infrastructure, release 2.7. 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 the <a
50 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
54 web site</a>. If you have questions or comments, the <a
55 href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM Developer's
56 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
59 main 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>
68 include/llvm/Analysis/LiveValues.h => Dan
69 lib/Transforms/IPO/MergeFunctions.cpp => consider for 2.8.
70 llvm/Analysis/PointerTracking.h => Edwin wants this, consider for 2.8.
73 lib/Transforms/Utils/SSI.cpp -> ABCD depends on it.
77 <!-- Features that need text if they're finished for 2.7:
80 llvm.dbg.value: variable debug info for optimized code
81 loop dependence analysis
84 <!-- for announcement email:
88 KLEE web page at klee.llvm.org
89 Many new papers added to /pubs/
93 <!-- *********************************************************************** -->
94 <div class="doc_section">
95 <a name="subproj">Sub-project Status Update</a>
97 <!-- *********************************************************************** -->
99 <div class="doc_text">
101 The LLVM 2.7 distribution currently consists of code from the core LLVM
102 repository (which roughly includes the LLVM optimizers, code generators
103 and supporting tools), the Clang repository and the llvm-gcc repository. In
104 addition to this code, the LLVM Project includes other sub-projects that are in
105 development. Here we include updates on these subprojects.
111 <!--=========================================================================-->
112 <div class="doc_subsection">
113 <a name="clang">Clang: C/C++/Objective-C Frontend Toolkit</a>
116 <div class="doc_text">
118 <p><a href="http://clang.llvm.org/">Clang</a> is an LLVM front end for the C, C++, and Objective-C languages. Clang aims to provide a better user experience through expressive diagnostics, a high level of conformance to language standards, fast compilation, and low memory use. Like LLVM, Clang provides a modular, library-based architecture that makes it suitable for creating or integrating with other development tools. Clang is considered a production-quality compiler for C and Objective-C on x86 (32- and 64-bit).</p>
120 <p>In the LLVM 2.7 time-frame, the Clang team has made many improvements:</p>
124 <li>C++ Support: Clang is now capable of self-hosting! While still alpha-quality, Clang's C++ support has matured enough to build LLVM and Clang, and C++ is now enabled by default. See the <a href="http://clang.llvm.org/cxx_compatibility.html">Clang C++ compatibility page</a> for common C++ migration issues.</li>
126 <li>New warnings: Clang contains a number of new warnings, including control-flow warnings (unreachable code, missing return statements in a non-<code>void</code> function, etc.), sign-comparison warnings, and improved format-string warnings.</li>
128 <li>CIndex API and Python bindings: Clang now includes a C API as part of the
129 CIndex library. Although we make make some changes to the API in the future, it
130 is intended to be stable and has been designed for use by external projects. See
132 doxygen <a href="http://clang.llvm.org/doxygen/group__CINDEX.html">CIndex</a>
133 documentation for more details. The CIndex API also includes a preliminary
134 set of Python bindings.</li>
136 <li>ARM Support: Clang now has ABI support for both the Darwin and Linux ARM
137 ABIs. Coupled with many improvements to the LLVM ARM backend, Clang is now
138 suitable for use as a a beta quality ARM compiler.</li>
142 <!--=========================================================================-->
143 <div class="doc_subsection">
144 <a name="clangsa">Clang Static Analyzer</a>
147 <div class="doc_text">
149 <p>The <a href="http://clang-analyzer.llvm.org/">Clang Static Analyzer</a>
150 project is an effort to use static source code analysis techniques to
151 automatically find bugs in C and Objective-C programs (and hopefully <a
152 href="http://clang-analyzer.llvm.org/dev_cxx.html">C++ in the
153 future</a>!). The tool is very good at finding bugs that occur on specific
154 paths through code, such as on error conditions.</p>
156 <p>In the LLVM 2.7 time-frame, the analyzer core has made several major and
157 minor improvements, including better support for tracking the fields of
158 structures, initial support (not enabled by default yet) for doing
159 interprocedural (cross-function) analysis, and new checks have been added.
164 <!--=========================================================================-->
165 <div class="doc_subsection">
166 <a name="vmkit">VMKit: JVM/CLI Virtual Machine Implementation</a>
169 <div class="doc_text">
171 The <a href="http://vmkit.llvm.org/">VMKit project</a> is an implementation of
172 a JVM and a CLI Virtual Machine (Microsoft .NET is an
173 implementation of the CLI) using LLVM for static and just-in-time
177 With the release of LLVM 2.7, VMKit has shifted to a great framework for writing
178 virtual machines. VMKit now offers precise and efficient garbage collection with
179 multi-threading support, thanks to the MMTk memory management toolkit, as well
180 as just in time and ahead of time compilation with LLVM. The major changes in
185 <li>Garbage collection: VMKit now uses the MMTk toolkit for garbage collectors.
186 The first collector to be ported is the MarkSweep collector, which is precise,
187 and drastically improves the performance of VMKit.</li>
188 <li>Line number information in the JVM: by using the debug metadata of LLVM, the
189 JVM now supports precise line number information, useful when printing a stack
191 <li>Interface calls in the JVM: we implemented a variant of the Interface Method
192 Table technique for interface calls in the JVM.
199 <!--=========================================================================-->
200 <div class="doc_subsection">
201 <a name="compiler-rt">compiler-rt: Compiler Runtime Library</a>
204 <div class="doc_text">
206 The new LLVM <a href="http://compiler-rt.llvm.org/">compiler-rt project</a>
207 is a simple library that provides an implementation of the low-level
208 target-specific hooks required by code generation and other runtime components.
209 For example, when compiling for a 32-bit target, converting a double to a 64-bit
210 unsigned integer is compiled into a runtime call to the "__fixunsdfdi"
211 function. The compiler-rt library provides highly optimized implementations of
212 this and other low-level routines (some are 3x faster than the equivalent
213 libgcc routines).</p>
216 All of the code in the compiler-rt project is available under the standard LLVM
217 License, a "BSD-style" license. New in LLVM 2.7: compiler_rt now
218 supports ARM targets.</p>
222 <!--=========================================================================-->
223 <div class="doc_subsection">
224 <a name="dragonegg">DragonEgg: llvm-gcc ported to gcc-4.5</a>
227 <div class="doc_text">
229 <a href="http://dragonegg.llvm.org/">DragonEgg</a> is a port of llvm-gcc to
230 gcc-4.5. Unlike llvm-gcc, which makes many intrusive changes to the underlying
231 gcc-4.2 code, dragonegg in theory does not require any gcc-4.5 modifications
232 whatsoever (currently one small patch is needed). This is thanks to the new
233 <a href="http://gcc.gnu.org/wiki/plugins">gcc plugin architecture</a>, which
234 makes it possible to modify the behaviour of gcc at runtime by loading a plugin,
235 which is nothing more than a dynamic library which conforms to the gcc plugin
236 interface. DragonEgg is a gcc plugin that causes the LLVM optimizers to be run
237 instead of the gcc optimizers, and the LLVM code generators instead of the gcc
238 code generators, just like llvm-gcc. To use it, you add
239 "-fplugin=path/dragonegg.so" to the gcc-4.5 command line, and gcc-4.5 magically
240 becomes llvm-gcc-4.5!
244 DragonEgg is still a work in progress. Currently C works very well, while C++,
245 Ada and Fortran work fairly well. All other languages either don't work at all,
246 or only work poorly. For the moment only the x86-32 and x86-64 targets are
247 supported, and only on linux and darwin (darwin needs an additional gcc patch).
251 DragonEgg is a new project which is seeing its first release with llvm-2.7.
257 <!--=========================================================================-->
258 <div class="doc_subsection">
259 <a name="mc">llvm-mc: Machine Code Toolkit</a>
262 <div class="doc_text">
264 The LLVM Machine Code (aka MC) sub-project of LLVM was created to solve a number
265 of problems in the realm of assembly, disassembly, object file format handling,
266 and a number of other related areas that CPU instruction-set level tools work
267 in. It is a sub-project of LLVM which provides it with a number of advantages
268 over other compilers that do not have tightly integrated assembly-level tools.
269 For a gentle introduction, please see the <a
270 href="http://blog.llvm.org/2010/04/intro-to-llvm-mc-project.html">Intro to the
271 LLVM MC Project Blog Post</a>.
274 <p>2.7 includes major parts of the work required by the new MC Project. A few
275 targets have been refactored to support it, and work is underway to support a
276 native assembler in LLVM. This work is not complete in LLVM 2.7, but you has
277 made substantially more progress on LLVM mainline.</p>
279 <p>One minor example of what MC can do is to transcode an AT&T syntax
280 X86 .s file into intel syntax. You can do this with something like:</p>
282 llvm-mc foo.s -output-asm-variant=1 -o foo-intel.s
288 <!-- *********************************************************************** -->
289 <div class="doc_section">
290 <a name="externalproj">External Open Source Projects Using LLVM 2.7</a>
292 <!-- *********************************************************************** -->
294 <div class="doc_text">
296 <p>An exciting aspect of LLVM is that it is used as an enabling technology for
297 a lot of other language and tools projects. This section lists some of the
298 projects that have already been updated to work with LLVM 2.7.</p>
301 <!--=========================================================================-->
302 <div class="doc_subsection">
303 <a name="pure">Pure</a>
306 <div class="doc_text">
308 <a href="http://pure-lang.googlecode.com/">Pure</a>
309 is an algebraic/functional programming language based on term rewriting.
310 Programs are collections of equations which are used to evaluate expressions in
311 a symbolic fashion. Pure offers dynamic typing, eager and lazy evaluation,
312 lexical closures, a hygienic macro system (also based on term rewriting),
313 built-in list and matrix support (including list and matrix comprehensions) and
314 an easy-to-use C interface. The interpreter uses LLVM as a backend to
315 JIT-compile Pure programs to fast native code.</p>
317 <p>Pure versions 0.43 and later have been tested and are known to work with
318 LLVM 2.7 (and continue to work with older LLVM releases >= 2.5).</p>
322 <!--=========================================================================-->
323 <div class="doc_subsection">
324 <a name="RoadsendPHP">Roadsend PHP</a>
327 <div class="doc_text">
329 <a href="http://code.roadsend.com/rphp">Roadsend PHP</a> (rphp) is an open
330 source implementation of the PHP programming
331 language that uses LLVM for its optimizer, JIT and static compiler. This is a
332 reimplementation of an earlier project that is now based on LLVM.
336 <!--=========================================================================-->
337 <div class="doc_subsection">
338 <a name="UnladenSwallow">Unladen Swallow</a>
341 <div class="doc_text">
343 <a href="http://code.google.com/p/unladen-swallow/">Unladen Swallow</a> is a
344 branch of <a href="http://python.org/">Python</a> intended to be fully
345 compatible and significantly faster. It uses LLVM's optimization passes and JIT
350 <!--=========================================================================-->
351 <div class="doc_subsection">
352 <a name="tce">TTA-based Codesign Environment (TCE)</a>
355 <div class="doc_text">
357 <a href="http://tce.cs.tut.fi/">TCE</a> is a toolset for designing
358 application-specific processors (ASP) based on the Transport triggered
359 architecture (TTA). The toolset provides a complete co-design flow from C/C++
360 programs down to synthesizable VHDL and parallel program binaries. Processor
361 customization points include the register files, function units, supported
362 operations, and the interconnection network.</p>
364 <p>TCE uses llvm-gcc/Clang and LLVM for C/C++ language support, target
365 independent optimizations and also for parts of code generation. It generates
366 new LLVM-based code generators "on the fly" for the designed TTA processors and
367 loads them in to the compiler backend as runtime libraries to avoid per-target
368 recompilation of larger parts of the compiler chain.</p>
372 <!--=========================================================================-->
373 <div class="doc_subsection">
374 <a name="safecode">SAFECode Compiler</a>
377 <div class="doc_text">
379 <a href="http://safecode.cs.illinois.edu">SAFECode</a> is a memory safe C
380 compiler built using LLVM. It takes standard, unannotated C code, analyzes the
381 code to ensure that memory accesses and array indexing operations are safe, and
382 instruments the code with run-time checks when safety cannot be proven
387 <!--=========================================================================-->
388 <div class="doc_subsection">
389 <a name="icedtea">IcedTea Java Virtual Machine Implementation</a>
392 <div class="doc_text">
394 <a href="http://icedtea.classpath.org/wiki/Main_Page">IcedTea</a> provides a
395 harness to build OpenJDK using only free software build tools and to provide
396 replacements for the not-yet free parts of OpenJDK. One of the extensions that
397 IcedTea provides is a new JIT compiler named <a
398 href="http://icedtea.classpath.org/wiki/ZeroSharkFaq">Shark</a> which uses LLVM
399 to provide native code generation without introducing processor-dependent
402 <p>Icedtea6 1.8 and later have been tested and are known to work with
403 LLVM 2.7 (and continue to work with older LLVM releases >= 2.6 as well).
407 <!--=========================================================================-->
408 <div class="doc_subsection">
409 <a name="llvm-lua">LLVM-Lua</a>
412 <div class="doc_text">
414 <a href="http://code.google.com/p/llvm-lua/">LLVM-Lua</a> uses LLVM
415 to add JIT and static compiling support to the Lua VM. Lua
416 bytecode is analyzed to remove type checks, then LLVM is used to compile the
417 bytecode down to machine code.
419 <p>LLVM-Lua 1.2.0 have been tested and is known to work with LLVM 2.7.
423 <!--=========================================================================-->
424 <div class="doc_subsection">
425 <a name="MacRuby">MacRuby</a>
428 <div class="doc_text">
430 <a href="http://macruby.org">MacRuby</a> is an implementation of Ruby based on
431 core Mac OS technologies, sponsored by Apple Inc. It uses LLVM at runtime for
432 optimization passes, JIT compilation and exception handling. It also allows
433 static (ahead-of-time) compilation of Ruby code straight to machine code.
435 <p>The upcoming MacRuby 0.6 release works with LLVM 2.7.
440 <!-- *********************************************************************** -->
441 <div class="doc_section">
442 <a name="whatsnew">What's New in LLVM 2.7?</a>
444 <!-- *********************************************************************** -->
446 <div class="doc_text">
448 <p>This release includes a huge number of bug fixes, performance tweaks and
449 minor improvements. Some of the major improvements and new features are listed
455 <!--=========================================================================-->
456 <div class="doc_subsection">
457 <a name="orgchanges">LLVM Community Changes</a>
460 <div class="doc_text">
462 <p>In addition to changes to the code, between LLVM 2.6 and 2.7, a number of
463 organization changes have happened:
467 <li>LLVM has a new <a href="http://llvm.org/Logo.html">official logo</a>!</li>
469 <li>Ted Kremenek and Doug Gregor have stepped forward as <a
470 href="http://llvm.org/docs/DeveloperPolicy.html#owners">Code Owners</a> of the
471 Clang static analyzer and the Clang frontend, respectively.</li>
473 <li>LLVM now has an <a href="http://blog.llvm.org">official Blog</a> at
474 <a href="http://blog.llvm.org">http://blog.llvm.org</a>. This is a great way
475 to learn about new LLVM-related features as they are implemented. Several
476 features in this release are already explained on the blog.</li>
478 <li>The LLVM web pages are now checked into the SVN server, in the "www",
479 "www-pubs" and "www-releases" SVN modules. Previously they were hidden in a
480 largely inaccessible old CVS server.</li>
482 <li><a href="http://llvm.org">llvm.org</a> is now hosted on a new (and much
483 faster) server. It is still graciously hosted at the University of Illinois
484 of Urbana Champaign.</li>
488 <!--=========================================================================-->
489 <div class="doc_subsection">
490 <a name="majorfeatures">Major New Features</a>
493 <div class="doc_text">
495 <p>LLVM 2.7 includes several major new capabilities:</p>
498 <li>2.7 includes initial support for the <a
499 href="http://en.wikipedia.org/wiki/MicroBlaze">MicroBlaze</a> target.
500 MicroBlaze is a soft processor core designed for Xilinx FPGAs.</li>
502 <li>2.7 includes a new LLVM IR "extensible metadata" feature. This feature
503 supports many different use cases, including allowing front-end authors to
504 encode source level information into LLVM IR, which is consumed by later
505 language-specific passes. This is a great way to do high-level optimizations
506 like devirtualization, type-based alias analysis, etc. See the <a
507 href="http://blog.llvm.org/2010/04/extensible-metadata-in-llvm-ir.html">
508 Extensible Metadata Blog Post</a> for more information.</li>
510 <li>2.7 encodes <a href="SourceLevelDebugging.html">debug information</a>
511 in a completely new way, built on extensible metadata. The new implementation
512 is much more memory efficient and paves the way for improvements to optimized
513 code debugging experience.</li>
515 <li>2.7 now directly supports taking the address of a label and doing an
516 indirect branch through a pointer. This is particularly useful for
517 interpreter loops, and is used to implement the GCC "address of label"
518 extension. For more information, see the <a
519 href="http://blog.llvm.org/2010/01/address-of-label-and-indirect-branches.html">
520 Address of Label and Indirect Branches in LLVM IR Blog Post</a>.
522 <li>2.7 is the first release to start supporting APIs for assembling and
523 disassembling target machine code. These APIs are useful for a variety of
524 low level clients, and are surfaced in the new "enhanced disassembly" API.
525 For more information see the <a
526 href="http://blog.llvm.org/2010/01/x86-disassembler.html">The X86
527 Disassembler Blog Post</a> for more information.</li>
529 <li>2.7 includes major parts of the work required by the new MC Project,
530 see the <a href="#mc">MC update above</a> for more information.</li>
536 <!--=========================================================================-->
537 <div class="doc_subsection">
538 <a name="coreimprovements">LLVM IR and Core Improvements</a>
541 <div class="doc_text">
542 <p>LLVM IR has several new features for better support of new targets and that
543 expose new optimization opportunities:</p>
546 <li>LLVM IR now supports a 16-bit "half float" data type through <a
547 href="LangRef.html#int_fp16">two new intrinsics</a> and APFloat support.</li>
548 <li>LLVM IR supports two new <a href="LangRef.html#fnattrs">function
549 attributes</a>: inlinehint and alignstack(n). The former is a hint to the
550 optimizer that a function was declared 'inline' and thus the inliner should
551 weight it higher when considering inlining it. The later
552 indicates to the code generator that the function diverges from the platform
553 ABI on stack alignment.</li>
554 <li>The new <a href="LangRef.html#int_objectsize">llvm.objectsize</a> intrinsic
555 allows the optimizer to infer the sizes of memory objects in some cases.
556 This intrinsic is used to implement the GCC <tt>__builtin_object_size</tt>
558 <li>LLVM IR now supports marking load and store instructions with <a
559 href="LangRef.html#i_load">"non-temporal" hints</a> (building on the new
560 metadata feature). This hint encourages the code
561 generator to generate non-temporal accesses when possible, which are useful
562 for code that is carefully managing cache behavior. Currently, only the
563 X86 backend provides target support for this feature.</li>
565 <li>LLVM 2.7 has pre-alpha support for <a
566 href="LangRef.html#t_union">unions in LLVM IR</a>.
567 Unfortunately, this support is not really usable in 2.7, so if you're
568 interested in pushing it forward, please help contribute to LLVM mainline.</li>
574 <!--=========================================================================-->
575 <div class="doc_subsection">
576 <a name="optimizer">Optimizer Improvements</a>
579 <div class="doc_text">
581 <p>In addition to a large array of minor performance tweaks and bug fixes, this
582 release includes a few major enhancements and additions to the optimizers:</p>
586 <li>The inliner reuses now merges arrays stack objects in different callees when
587 inlining multiple call sites into one function. This reduces the stack size
588 of the resultant function.</li>
589 <li>The -basicaa alias analysis pass (which is the default) has been improved to
590 be less dependent on "type safe" pointers. It can now look through bitcasts
591 and other constructs more aggressively, allowing better load/store
593 <li>The load elimination optimization in the GVN Pass [<a
594 href="http://blog.llvm.org/2009/12/introduction-to-load-elimination-in-gvn.html">intro
595 blog post</a>] has been substantially improved to be more aggressive about
596 partial redundancy elimination and do more aggressive phi translation. Please
598 href="http://blog.llvm.org/2009/12/advanced-topics-in-redundant-load.html">
599 Advanced Topics in Redundant Load Elimination with a Focus on PHI Translation
600 Blog Post</a> for more details.</li>
601 <li>The module <a href="LangRef.html#datalayout">target data string</a> now
602 includes a notion of 'native' integer data types for the target. This
603 helps mid-level optimizations avoid promoting complex sequences of
604 operations to data types that are not natively supported (e.g. converting
605 i32 operations to i64 on 32-bit chips).</li>
606 <li>The mid-level optimizer is now conservative when operating on a module with
607 no target data. Previously, it would default to SparcV9 settings, which is
608 not what most people expected.</li>
609 <li>Jump threading is now much more aggressive at simplifying correlated
610 conditionals and threading blocks with otherwise complex logic. It has
611 subsumed the old "Conditional Propagation" pass, and -condprop has been
612 removed from LLVM 2.7.</li>
613 <li>The -instcombine pass has been refactored from being one huge file to being
614 a library of its own. Internally, it uses a customized IRBuilder to clean
615 it up and simplify it.</li>
617 <li>The optimal edge profiling pass is reliable and much more complete than in
618 2.6. It can be used with the llvm-prof tool but isn't wired up to the
619 llvm-gcc and clang command line options yet.</li>
621 <li>A new experimental alias analysis implementation, -scev-aa, has been added.
622 It uses LLVM's Scalar Evolution implementation to do symbolic analysis of
623 pointer offset expressions to disambiguate pointers. It can catch a few
624 cases that basicaa cannot, particularly in complex loop nests.</li>
626 <li>The default pass ordering has been tweaked for improved optimization
634 <!--=========================================================================-->
635 <div class="doc_subsection">
636 <a name="executionengine">Interpreter and JIT Improvements</a>
639 <div class="doc_text">
642 <li>The JIT now supports generating debug information and is compatible with
643 the new GDB 7.0 (and later) interfaces for registering dynamically generated
646 <li>The JIT now <a href="http://llvm.org/PR5184">defaults
647 to compiling eagerly</a> to avoid a race condition in the lazy JIT.
648 Clients that still want the lazy JIT can switch it on by calling
649 <tt>ExecutionEngine::DisableLazyCompilation(false)</tt>.</li>
651 <li>It is now possible to create more than one JIT instance in the same process.
652 These JITs can generate machine code in parallel,
653 although <a href="http://llvm.org/docs/ProgrammersManual.html#jitthreading">you
654 still have to obey the other threading restrictions</a>.</li>
660 <!--=========================================================================-->
661 <div class="doc_subsection">
662 <a name="codegen">Target Independent Code Generator Improvements</a>
665 <div class="doc_text">
667 <p>We have put a significant amount of work into the code generator
668 infrastructure, which allows us to implement more aggressive algorithms and make
672 <li>The 'llc -asm-verbose' option (which is now the default) has been enhanced
673 to emit many useful comments to .s files indicating information about spill
674 slots and loop nest structure. This should make it much easier to read and
675 understand assembly files. This is wired up in llvm-gcc and clang to
676 the <tt>-fverbose-asm</tt> option.</li>
678 <li>New LSR with "full strength reduction" mode, which can reduce address
679 register pressure in loops where address generation is important.</li>
681 <li>A new codegen level Common Subexpression Elimination pass (MachineCSE)
682 is available and enabled by default. It catches redundancies exposed by
684 <li>A new pre-register-allocation tail duplication pass is available and enabled
685 by default, it can substantially improve branch prediction quality in some
687 <li>A new sign and zero extension optimization pass (OptimizeExtsPass)
688 is available and enabled by default. This pass can takes advantage
689 architecture features like x86-64 implicit zero extension behavior and
691 <li>The code generator now supports a mode where it attempts to preserve the
692 order of instructions in the input code. This is important for source that
693 is hand scheduled and extremely sensitive to scheduling. It is compatible
694 with the GCC <tt>-fno-schedule-insns</tt> option.</li>
695 <li>The target-independent code generator now supports generating code with
696 arbitrary numbers of result values. Returning more values than was
697 previously supported is handled by returning through a hidden pointer. In
698 2.7, only the X86 and XCore targets have adopted support for this
700 <li>The code generator now supports generating code that follows the
701 <a href="LangRef.html#callingconv">Glasgow Haskell Compiler Calling
702 Convention</a> and ABI.</li>
703 <li>The "<a href="CodeGenerator.html#selectiondag_select">DAG instruction
704 selection</a>" phase of the code generator has been largely rewritten for
705 2.7. Previously, tblgen spit out tons of C++ code which was compiled and
706 linked into the target to do the pattern matching, now it emits a much
707 smaller table which is read by the target-independent code. The primary
708 advantages of this approach is that the size and compile time of various
709 targets is much improved. The X86 code generator shrunk by 1.5MB of code,
711 <li>Almost the entire code generator has switched to emitting code through the
712 MC interfaces instead of printing textually to the .s file. This led to a
713 number of cleanups and speedups. In 2.7, debug an exception handling
714 information does not go through MC yet.</li>
718 <!--=========================================================================-->
719 <div class="doc_subsection">
720 <a name="x86">X86-32 and X86-64 Target Improvements</a>
723 <div class="doc_text">
724 <p>New features of the X86 target include:
728 <li>The X86 backend now optimizes tails calls much more aggressively for
729 functions that use the standard C calling convention.</li>
730 <li>The X86 backend now models scalar SSE registers as subregs of the SSE vector
731 registers, making the code generator more aggressive in cases where scalars
732 and vector types are mixed.</li>
738 <!--=========================================================================-->
739 <div class="doc_subsection">
740 <a name="ARM">ARM Target Improvements</a>
743 <div class="doc_text">
744 <p>New features of the ARM target include:
749 <li>The ARM backend now generates instructions in unified assembly syntax.</li>
751 <li>llvm-gcc now has complete support for the ARM v7 NEON instruction set. This
752 support differs slightly from the GCC implementation. Please see the
754 href="http://blog.llvm.org/2010/04/arm-advanced-simd-neon-intrinsics-and.html">
755 ARM Advanced SIMD (NEON) Intrinsics and Types in LLVM Blog Post</a> for
756 helpful information if migrating code from GCC to LLVM-GCC.</li>
758 <li>The ARM and Thumb code generators now use register scavenging for stack
759 object address materialization. This allows the use of R3 as a general
760 purpose register in Thumb1 code, as it was previous reserved for use in
761 stack address materialization. Secondly, sequential uses of the same
762 value will now re-use the materialized constant.</li>
764 <li>The ARM backend now has good support for ARMv4 targets and has been tested
765 on StrongARM hardware. Previously, LLVM only supported ARMv4T and
768 <li>Atomic builtins are now supported for ARMv6 and ARMv7 (__sync_synchronize,
769 __sync_fetch_and_add, etc.).</li>
776 <!--=========================================================================-->
777 <div class="doc_subsection">
778 <a name="newapis">New Useful APIs</a>
781 <div class="doc_text">
783 <p>This release includes a number of new APIs that are used internally, which
784 may also be useful for external clients.
788 <li>The optimizer uses the new CodeMetrics class to measure the size of code.
789 Various passes (like the inliner, loop unswitcher, etc) all use this to make
790 more accurate estimates of the code size impact of various
792 <li>A new <a href="http://llvm.org/doxygen/InstructionSimplify_8h-source.html">
793 llvm/Analysis/InstructionSimplify.h</a> interface is available for doing
794 symbolic simplification of instructions (e.g. <tt>a+0</tt> -> <tt>a</tt>)
795 without requiring the instruction to exist. This centralizes a lot of
796 ad-hoc symbolic manipulation code scattered in various passes.</li>
797 <li>The optimizer now uses a new <a
798 href="http://llvm.org/doxygen/SSAUpdater_8h-source.html">SSAUpdater</a>
799 class which efficiently supports
800 doing unstructured SSA update operations. This centralized a bunch of code
801 scattered throughout various passes (e.g. jump threading, lcssa,
802 loop rotate, etc) for doing this sort of thing. The code generator has a
803 similar <a href="http://llvm.org/doxygen/MachineSSAUpdater_8h-source.html">
804 MachineSSAUpdater</a> class.</li>
805 <li>The <a href="http://llvm.org/doxygen/Regex_8h-source.html">
806 llvm/Support/Regex.h</a> header exposes a platform independent regular
807 expression API. Building on this, the <a
808 href="TestingGuide.html#FileCheck">FileCheck</a> utility now supports
809 regular exressions.</li>
810 <li>raw_ostream now supports a circular "debug stream" accessed with "dbgs()".
811 By default, this stream works the same way as "errs()", but if you pass
812 <tt>-debug-buffer-size=1000</tt> to opt, the debug stream is capped to a
813 fixed sized circular buffer and the output is printed at the end of the
814 program's execution. This is helpful if you have a long lived compiler
815 process and you're interested in seeing snapshots in time.</li>
821 <!--=========================================================================-->
822 <div class="doc_subsection">
823 <a name="otherimprovements">Other Improvements and New Features</a>
826 <div class="doc_text">
827 <p>Other miscellaneous features include:</p>
830 <li>You can now build LLVM as a big dynamic library (e.g. "libllvm2.7.so"). To
831 get this, configure LLVM with the --enable-shared option.</li>
833 <li>LLVM command line tools now overwrite their output by default. Previously,
834 they would only do this with -f. This makes them more convenient to use, and
835 behave more like standard unix tools.</li>
837 <li>The opt and llc tools now autodetect whether their input is a .ll or .bc
838 file, and automatically do the right thing. This means you don't need to
839 explicitly use the llvm-as tool for most things.</li>
845 <!--=========================================================================-->
846 <div class="doc_subsection">
847 <a name="changes">Major Changes and Removed Features</a>
850 <div class="doc_text">
852 <p>If you're already an LLVM user or developer with out-of-tree changes based
853 on LLVM 2.6, this section lists some "gotchas" that you may run into upgrading
854 from the previous release.</p>
859 The Andersen's alias analysis ("anders-aa") pass, the Predicate Simplifier
860 ("predsimplify") pass, the LoopVR pass, the GVNPRE pass, and the random sampling
861 profiling ("rsprofiling") passes have all been removed. They were not being
862 actively maintained and had substantial problems. If you are interested in
863 these components, you are welcome to ressurect them from SVN, fix the
864 correctness problems, and resubmit them to mainline.</li>
866 <li>LLVM now defaults to building most libraries with RTTI turned off, providing
867 a code size reduction. Packagers who are interested in building LLVM to support
868 plugins that require RTTI information should build with "make REQUIRE_RTTI=1"
869 and should read the new <a href="Packaging.html">Advice on Packaging LLVM</a>
872 <li>The LLVM interpreter now defaults to <em>not</em> using <tt>libffi</tt> even
873 if you have it installed. This makes it more likely that an LLVM built on one
874 system will work when copied to a similar system. To use <tt>libffi</tt>,
875 configure with <tt>--enable-libffi</tt>.</li>
877 <li>Debug information uses a completely different representation, an LLVM 2.6
878 .bc file should work with LLVM 2.7, but debug info won't come forward.</li>
880 <li>The LLVM 2.6 (and earlier) "malloc" and "free" instructions got removed,
881 along with LowerAllocations pass. Now you should just use a call to the
882 malloc and free functions in libc. These calls are optimized as well as
883 the old instructions were.</li>
886 <p>In addition, many APIs have changed in this release. Some of the major LLVM
890 <li>Just about everything has been converted to use raw_ostream instead of
892 <li>llvm/ADT/iterator.h has been removed, just use <iterator>
894 <li>The Streams.h file and "DOUT" got removed, use "DEBUG(errs() << ...);"
896 <li>The TargetAsmInfo interface was renamed to MCAsmInfo.</li>
897 <li><tt>ModuleProvider</tt> has been <a
898 href="http://llvm.org/viewvc/llvm-project?view=rev&revision=94686">removed</a>
899 and its methods moved to <tt>Module</tt> and <tt>GlobalValue</tt>.
900 Most clients can remove uses of <tt>ExistingModuleProvider</tt>,
901 replace <tt>getBitcodeModuleProvider</tt> with
902 <tt>getLazyBitcodeModule</tt>, and pass their <tt>Module</tt> to
903 functions that used to accept <tt>ModuleProvider</tt>. Clients who
904 wrote their own <tt>ModuleProvider</tt>s will need to derive from
905 <tt>GVMaterializer</tt> instead and use
906 <tt>Module::setMaterializer</tt> to attach it to a
907 <tt>Module</tt>.</li>
909 <li><tt>GhostLinkage</tt> has given up the ghost.
910 <tt>GlobalValue</tt>s that have not yet been read from their backing
911 storage have the same linkage they will have after being read in.
912 Clients must replace calls to
913 <tt>GlobalValue::hasNotBeenReadFromBitcode</tt> with
914 <tt>GlobalValue::isMaterializable</tt>.</li>
916 <li>The <tt>isInteger</tt>, <tt>isIntOrIntVector</tt>, <tt>isFloatingPoint</tt>,
917 <tt>isFPOrFPVector</tt> and <tt>isFPOrFPVector</tt> methods have been renamed
918 <tt>isIntegerTy</tt>, <tt>isIntOrIntVectorTy</tt>, <tt>isFloatingPointTy</tt>,
919 <tt>isFPOrFPVectorTy</tt> and <tt>isFPOrFPVectorTy</tt> respectively.</li>
921 <li><tt>llvm::Instruction::clone()</tt> no longer takes argument.</li>
922 <li><tt>raw_fd_ostream</tt>'s constructor now takes a flag argument, not individual
923 booleans (see <tt>include/llvm/Support/raw_ostream.h</tt> for details).</li>
924 <li>Some header files have been renamed:
926 <li><tt>llvm/Support/AIXDataTypesFix.h</tt> to
927 <tt>llvm/System/AIXDataTypesFix.h</tt></li>
928 <li><tt>llvm/Support/DataTypes.h</tt> to <tt>llvm/System/DataTypes.h</tt></li>
929 <li><tt>llvm/Transforms/Utils/InlineCost.h</tt> to
930 <tt>llvm/Analysis/InlineCost.h</tt></li>
931 <li><tt>llvm/Support/Mangler.h</tt> to <tt>llvm/Target/Mangler.h</tt></li>
932 <li><tt>llvm/Analysis/Passes.h</tt> to <tt>llvm/CodeGen/Passes.h</tt></li>
940 <!-- *********************************************************************** -->
941 <div class="doc_section">
942 <a name="portability">Portability and Supported Platforms</a>
944 <!-- *********************************************************************** -->
946 <div class="doc_text">
948 <p>LLVM is known to work on the following platforms:</p>
951 <li>Intel and AMD machines (IA32, X86-64, AMD64, EMT-64) running Red Hat
952 Linux, Fedora Core, FreeBSD and AuroraUX (and probably other unix-like
954 <li>PowerPC and X86-based Mac OS X systems, running 10.4 and above in 32-bit
955 and 64-bit modes.</li>
956 <li>Intel and AMD machines running on Win32 using MinGW libraries (native).</li>
957 <li>Intel and AMD machines running on Win32 with the Cygwin libraries (limited
958 support is available for native builds with Visual C++).</li>
959 <li>Sun x86 and AMD64 machines running Solaris 10, OpenSolaris 0906.</li>
960 <li>Alpha-based machines running Debian GNU/Linux.</li>
963 <p>The core LLVM infrastructure uses GNU autoconf to adapt itself
964 to the machine and operating system on which it is built. However, minor
965 porting may be required to get LLVM to work on new platforms. We welcome your
966 portability patches and reports of successful builds or error messages.</p>
970 <!-- *********************************************************************** -->
971 <div class="doc_section">
972 <a name="knownproblems">Known Problems</a>
974 <!-- *********************************************************************** -->
976 <div class="doc_text">
978 <p>This section contains significant known problems with the LLVM system,
979 listed by component. If you run into a problem, please check the <a
980 href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
981 there isn't already one.</p>
984 <li>LLVM will not correctly compile on Solaris and/or OpenSolaris
985 using the stock GCC 3.x.x series 'out the box',
986 See: <a href="GettingStarted.html#brokengcc">Broken versions of GCC and other tools</a>.
987 However, A <a href="http://pkg.auroraux.org/GCC">Modern GCC Build</a>
988 for x86/x86-64 has been made available from the third party AuroraUX Project
989 that has been meticulously tested for bootstrapping LLVM & Clang.</li>
994 <!-- ======================================================================= -->
995 <div class="doc_subsection">
996 <a name="experimental">Experimental features included with this release</a>
999 <div class="doc_text">
1001 <p>The following components of this LLVM release are either untested, known to
1002 be broken or unreliable, or are in early development. These components should
1003 not be relied on, and bugs should not be filed against them, but they may be
1004 useful to some people. In particular, if you would like to work on one of these
1005 components, please contact us on the <a
1006 href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev list</a>.</p>
1009 <li>The MSIL, Alpha, SPU, MIPS, PIC16, Blackfin, MSP430, SystemZ and MicroBlaze
1010 backends are experimental.</li>
1011 <li><tt>llc</tt> "<tt>-filetype=asm</tt>" (the default) is the only
1012 supported value for this option. The MachO writer is experimental, and
1013 works much better in mainline SVN.</li>
1018 <!-- ======================================================================= -->
1019 <div class="doc_subsection">
1020 <a name="x86-be">Known problems with the X86 back-end</a>
1023 <div class="doc_text">
1026 <li>The X86 backend does not yet support
1027 all <a href="http://llvm.org/PR879">inline assembly that uses the X86
1028 floating point stack</a>. It supports the 'f' and 't' constraints, but not
1030 <li>The X86 backend generates inefficient floating point code when configured
1031 to generate code for systems that don't have SSE2.</li>
1032 <li>Win64 code generation wasn't widely tested. Everything should work, but we
1033 expect small issues to happen. Also, llvm-gcc cannot build the mingw64
1034 runtime currently due to lack of support for the 'u' inline assembly
1035 constraint and for X87 floating point inline assembly.</li>
1036 <li>The X86-64 backend does not yet support the LLVM IR instruction
1037 <tt>va_arg</tt>. Currently, front-ends support variadic
1038 argument constructs on X86-64 by lowering them manually.</li>
1043 <!-- ======================================================================= -->
1044 <div class="doc_subsection">
1045 <a name="ppc-be">Known problems with the PowerPC back-end</a>
1048 <div class="doc_text">
1051 <li>The Linux PPC32/ABI support needs testing for the interpreter and static
1052 compilation, and lacks support for debug information.</li>
1057 <!-- ======================================================================= -->
1058 <div class="doc_subsection">
1059 <a name="arm-be">Known problems with the ARM back-end</a>
1062 <div class="doc_text">
1065 <li>Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6
1066 processors, thumb programs can crash or produce wrong
1067 results (<a href="http://llvm.org/PR1388">PR1388</a>).</li>
1068 <li>Compilation for ARM Linux OABI (old ABI) is supported but not fully tested.
1074 <!-- ======================================================================= -->
1075 <div class="doc_subsection">
1076 <a name="sparc-be">Known problems with the SPARC back-end</a>
1079 <div class="doc_text">
1082 <li>The SPARC backend only supports the 32-bit SPARC ABI (-m32); it does not
1083 support the 64-bit SPARC ABI (-m64).</li>
1088 <!-- ======================================================================= -->
1089 <div class="doc_subsection">
1090 <a name="mips-be">Known problems with the MIPS back-end</a>
1093 <div class="doc_text">
1096 <li>64-bit MIPS targets are not supported yet.</li>
1101 <!-- ======================================================================= -->
1102 <div class="doc_subsection">
1103 <a name="alpha-be">Known problems with the Alpha back-end</a>
1106 <div class="doc_text">
1110 <li>On 21164s, some rare FP arithmetic sequences which may trap do not have the
1111 appropriate nops inserted to ensure restartability.</li>
1116 <!-- ======================================================================= -->
1117 <div class="doc_subsection">
1118 <a name="c-be">Known problems with the C back-end</a>
1121 <div class="doc_text">
1124 <li><a href="http://llvm.org/PR802">The C backend has only basic support for
1125 inline assembly code</a>.</li>
1126 <li><a href="http://llvm.org/PR1658">The C backend violates the ABI of common
1127 C++ programs</a>, preventing intermixing between C++ compiled by the CBE and
1128 C++ code compiled with <tt>llc</tt> or native compilers.</li>
1129 <li>The C backend does not support all exception handling constructs.</li>
1130 <li>The C backend does not support arbitrary precision integers.</li>
1136 <!-- ======================================================================= -->
1137 <div class="doc_subsection">
1138 <a name="c-fe">Known problems with the llvm-gcc C and C++ front-end</a>
1141 <div class="doc_text">
1143 <p>The only major language feature of GCC not supported by llvm-gcc is
1144 the <tt>__builtin_apply</tt> family of builtins. However, some extensions
1145 are only supported on some targets. For example, trampolines are only
1146 supported on some targets (these are used when you take the address of a
1147 nested function).</p>
1151 <!-- ======================================================================= -->
1152 <div class="doc_subsection">
1153 <a name="fortran-fe">Known problems with the llvm-gcc Fortran front-end</a>
1156 <div class="doc_text">
1158 <li>Fortran support generally works, but there are still several unresolved bugs
1159 in <a href="http://llvm.org/bugs/">Bugzilla</a>. Please see the
1160 tools/gfortran component for details.</li>
1164 <!-- ======================================================================= -->
1165 <div class="doc_subsection">
1166 <a name="ada-fe">Known problems with the llvm-gcc Ada front-end</a>
1169 <div class="doc_text">
1170 The llvm-gcc 4.2 Ada compiler works fairly well; however, this is not a mature
1171 technology, and problems should be expected.
1173 <li>The Ada front-end currently only builds on X86-32. This is mainly due
1174 to lack of trampoline support (pointers to nested functions) on other platforms.
1175 However, it <a href="http://llvm.org/PR2006">also fails to build on X86-64</a>
1176 which does support trampolines.</li>
1177 <li>The Ada front-end <a href="http://llvm.org/PR2007">fails to bootstrap</a>.
1178 This is due to lack of LLVM support for <tt>setjmp</tt>/<tt>longjmp</tt> style
1179 exception handling, which is used internally by the compiler.
1180 Workaround: configure with <tt>--disable-bootstrap</tt>.</li>
1181 <li>The c380004, <a href="http://llvm.org/PR2010">c393010</a>
1182 and <a href="http://llvm.org/PR2421">cxg2021</a> ACATS tests fail
1183 (c380004 also fails with gcc-4.2 mainline).
1184 If the compiler is built with checks disabled then <a href="http://llvm.org/PR2010">c393010</a>
1185 causes the compiler to go into an infinite loop, using up all system memory.</li>
1186 <li>Some GCC specific Ada tests continue to crash the compiler.</li>
1187 <li>The <tt>-E</tt> binder option (exception backtraces)
1188 <a href="http://llvm.org/PR1982">does not work</a> and will result in programs
1189 crashing if an exception is raised. Workaround: do not use <tt>-E</tt>.</li>
1190 <li>Only discrete types <a href="http://llvm.org/PR1981">are allowed to start
1191 or finish at a non-byte offset</a> in a record. Workaround: do not pack records
1192 or use representation clauses that result in a field of a non-discrete type
1193 starting or finishing in the middle of a byte.</li>
1194 <li>The <tt>lli</tt> interpreter <a href="http://llvm.org/PR2009">considers
1195 'main' as generated by the Ada binder to be invalid</a>.
1196 Workaround: hand edit the file to use pointers for <tt>argv</tt> and
1197 <tt>envp</tt> rather than integers.</li>
1198 <li>The <tt>-fstack-check</tt> option <a href="http://llvm.org/PR2008">is
1203 <!-- *********************************************************************** -->
1204 <div class="doc_section">
1205 <a name="additionalinfo">Additional Information</a>
1207 <!-- *********************************************************************** -->
1209 <div class="doc_text">
1211 <p>A wide variety of additional information is available on the <a
1212 href="http://llvm.org">LLVM web page</a>, in particular in the <a
1213 href="http://llvm.org/docs/">documentation</a> section. The web page also
1214 contains versions of the API documentation which is up-to-date with the
1215 Subversion version of the source code.
1216 You can access versions of these documents specific to this release by going
1217 into the "<tt>llvm/doc/</tt>" directory in the LLVM tree.</p>
1219 <p>If you have any questions or comments about LLVM, please feel free to contact
1220 us via the <a href="http://llvm.org/docs/#maillist"> mailing
1225 <!-- *********************************************************************** -->
1229 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1230 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1231 <a href="http://validator.w3.org/check/referer"><img
1232 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1234 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
1235 Last modified: $Date$