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">
7 <link rel="stylesheet" href="llvm.css" type="text/css">
8 <title>LLVM 2.8 Release Notes</title>
12 <div class="doc_title">LLVM 2.8 Release Notes</div>
14 <img align=right src="http://llvm.org/img/DragonSmall.png"
15 width="136" height="136" alt="LLVM Dragon Logo">
18 <li><a href="#intro">Introduction</a></li>
19 <li><a href="#subproj">Sub-project Status Update</a></li>
20 <li><a href="#externalproj">External Projects Using LLVM 2.8</a></li>
21 <li><a href="#whatsnew">What's New in LLVM 2.8?</a></li>
22 <li><a href="GettingStarted.html">Installation Instructions</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/2.7/docs/ReleaseNotes.html">LLVM 2.7
36 Release Notes</a>.</h1>
39 <!-- *********************************************************************** -->
40 <div class="doc_section">
41 <a name="intro">Introduction</a>
43 <!-- *********************************************************************** -->
45 <div class="doc_text">
47 <p>This document contains the release notes for the LLVM Compiler
48 Infrastructure, release 2.8. Here we describe the status of LLVM, including
49 major improvements from the previous release and significant known problems.
50 All LLVM releases may be downloaded from the <a
51 href="http://llvm.org/releases/">LLVM releases web site</a>.</p>
53 <p>For more information about LLVM, including information about the latest
54 release, please check out the <a href="http://llvm.org/">main LLVM
55 web site</a>. If you have questions or comments, the <a
56 href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM Developer's
57 Mailing List</a> is a good place to send them.</p>
59 <p>Note that if you are reading this file from a Subversion checkout or the
60 main LLVM web page, this document applies to the <i>next</i> release, not the
61 current one. To see the release notes for a specific release, please see the
62 <a href="http://llvm.org/releases/">releases page</a>.</p>
69 include/llvm/Analysis/LiveValues.h => Dan
70 lib/Transforms/IPO/MergeFunctions.cpp => consider for 2.8.
75 <!-- Features that need text if they're finished for 2.9:
78 loop dependence analysis
80 CorrelatedValuePropagation
83 <!-- Announcement, lldb, libc++ -->
86 <!-- *********************************************************************** -->
87 <div class="doc_section">
88 <a name="subproj">Sub-project Status Update</a>
90 <!-- *********************************************************************** -->
92 <div class="doc_text">
94 The LLVM 2.8 distribution currently consists of code from the core LLVM
95 repository (which roughly includes the LLVM optimizers, code generators
96 and supporting tools), the Clang repository and the llvm-gcc repository. In
97 addition to this code, the LLVM Project includes other sub-projects that are in
98 development. Here we include updates on these subprojects.
104 <!--=========================================================================-->
105 <div class="doc_subsection">
106 <a name="clang">Clang: C/C++/Objective-C Frontend Toolkit</a>
109 <div class="doc_text">
111 <p><a href="http://clang.llvm.org/">Clang</a> is an LLVM front end for the C,
112 C++, and Objective-C languages. Clang aims to provide a better user experience
113 through expressive diagnostics, a high level of conformance to language
114 standards, fast compilation, and low memory use. Like LLVM, Clang provides a
115 modular, library-based architecture that makes it suitable for creating or
116 integrating with other development tools. Clang is considered a
117 production-quality compiler for C, Objective-C, C++ and Objective-C++ on x86
118 (32- and 64-bit), and for darwin-arm targets.</p>
120 <p>In the LLVM 2.8 time-frame, the Clang team has made many improvements:</p>
123 <li>Clang C++ is now feature-complete with respect to the ISO C++ 1998 and 2003 standards.</li>
124 <li>Added support for Objective-C++.</li>
125 <li>Clang now uses LLVM-MC to directly generate object code and to parse inline assembly (on Darwin).</li>
126 <li>Introduced many new warnings, including <code>-Wmissing-field-initializers</code>, <code>-Wshadow</code>, <code>-Wno-protocol</code>, <code>-Wtautological-compare</code>, <code>-Wstrict-selector-match</code>, <code>-Wcast-align</code>, <code>-Wunused</code> improvements, and greatly improved format-string checking.</li>
127 <li>Introduced the "libclang" library, a C interface to Clang intended to support IDE clients.</li>
128 <li>Added support for <code>#pragma GCC visibility</code>, <code>#pragma align</code>, and others.</li>
129 <li>Added support for SSE, ARM NEON, and Altivec.</li>
130 <li>Implemented support for blocks in C++.</li>
131 <li>Implemented precompiled headers for C++.</li>
132 <li>Improved abstract syntax trees to retain more accurate source information.</li>
136 <!--=========================================================================-->
137 <div class="doc_subsection">
138 <a name="clangsa">Clang Static Analyzer</a>
141 <div class="doc_text">
143 <p>The <a href="http://clang-analyzer.llvm.org/">Clang Static Analyzer</a>
144 project is an effort to use static source code analysis techniques to
145 automatically find bugs in C and Objective-C programs (and hopefully <a
146 href="http://clang-analyzer.llvm.org/dev_cxx.html">C++ in the
147 future</a>!). The tool is very good at finding bugs that occur on specific
148 paths through code, such as on error conditions.</p>
150 <p>The LLVM 2.8 release fixes a number of bugs and slightly improves precision
151 over 2.7, but there are no major new features in the release.
156 <!--=========================================================================-->
157 <div class="doc_subsection">
158 <a name="dragonegg">DragonEgg: llvm-gcc ported to gcc-4.5</a>
161 <div class="doc_text">
163 <a href="http://dragonegg.llvm.org/">DragonEgg</a> is a port of llvm-gcc to
164 gcc-4.5. Unlike llvm-gcc, dragonegg in theory does not require any gcc-4.5
165 modifications whatsoever (currently one small patch is needed) thanks to the
166 new <a href="http://gcc.gnu.org/wiki/plugins">gcc plugin architecture</a>.
167 DragonEgg is a gcc plugin that makes gcc-4.5 use the LLVM optimizers and code
168 generators instead of gcc's, just like with llvm-gcc.
172 DragonEgg is still a work in progress, but it is able to compile a lot of code,
173 for example all of gcc, LLVM and clang. Currently Ada, C, C++ and Fortran work
174 well, while all other languages either don't work at all or only work poorly.
175 For the moment only the x86-32 and x86-64 targets are supported, and only on
176 linux and darwin (darwin may need additional gcc patches).
180 The 2.8 release has the following notable changes:
182 <li>The plugin loads faster due to exporting fewer symbols.</li>
183 <li>Additional vector operations such as addps256 are now supported.</li>
184 <li>Ada global variables with no initial value are no longer zero initialized,
185 resulting in better optimization.</li>
186 <li>The '-fplugin-arg-dragonegg-enable-gcc-optzns' flag now runs all gcc
187 optimizers, rather than just a handful.</li>
188 <li>Fortran programs using common variables now link correctly.</li>
189 <li>GNU OMP constructs no longer crash the compiler.</li>
195 <!--=========================================================================-->
196 <div class="doc_subsection">
197 <a name="vmkit">VMKit: JVM/CLI Virtual Machine Implementation</a>
200 <div class="doc_text">
202 The <a href="http://vmkit.llvm.org/">VMKit project</a> is an implementation of
203 a Java Virtual Machine (Java VM or JVM) that uses LLVM for static and
204 just-in-time compilation. As of LLVM 2.8, VMKit now supports copying garbage
205 collectors, and can be configured to use MMTk's copy mark-sweep garbage
206 collector. In LLVM 2.8, the VMKit .NET VM is no longer being maintained.
210 <!--=========================================================================-->
211 <div class="doc_subsection">
212 <a name="compiler-rt">compiler-rt: Compiler Runtime Library</a>
215 <div class="doc_text">
217 The new LLVM <a href="http://compiler-rt.llvm.org/">compiler-rt project</a>
218 is a simple library that provides an implementation of the low-level
219 target-specific hooks required by code generation and other runtime components.
220 For example, when compiling for a 32-bit target, converting a double to a 64-bit
221 unsigned integer is compiled into a runtime call to the "__fixunsdfdi"
222 function. The compiler-rt library provides highly optimized implementations of
223 this and other low-level routines (some are 3x faster than the equivalent
224 libgcc routines).</p>
227 All of the code in the compiler-rt project is available under the standard LLVM
228 License, a "BSD-style" license. New in LLVM 2.8, compiler_rt now supports
229 soft floating point (for targets that don't have a real floating point unit),
230 and includes an extensive testsuite for the "blocks" language feature and the
231 blocks runtime included in compiler_rt.</p>
235 <!--=========================================================================-->
236 <div class="doc_subsection">
237 <a name="lldb">LLDB: Low Level Debugger</a>
240 <div class="doc_text">
242 <a href="http://lldb.llvm.org/">LLDB</a> is a brand new member of the LLVM
243 umbrella of projects. LLDB is a next generation, high-performance debugger. It
244 is built as a set of reusable components which highly leverage existing
245 libraries in the larger LLVM Project, such as the Clang expression parser, the
246 LLVM disassembler and the LLVM JIT.</p>
249 LLDB is in early development and not included as part of the LLVM 2.8 release,
250 but is mature enough to support basic debugging scenarios on Mac OS X in C,
251 Objective-C and C++. We'd really like help extending and expanding LLDB to
252 support new platforms, new languages, new architectures, and new features.
257 <!--=========================================================================-->
258 <div class="doc_subsection">
259 <a name="libc++">libc++: C++ Standard Library</a>
262 <div class="doc_text">
264 <a href="http://libc++.llvm.org/">libc++</a> is another new member of the LLVM
265 family. It is an implementation of the C++ standard library, written from the
266 ground up to specifically target the forthcoming C++'0X standard and focus on
267 delivering great performance.</p>
270 As of the LLVM 2.8 release, libc++ is virtually feature complete, but would
271 benefit from more testing and better integration with Clang++. It is also
272 looking forward to the C++ committee finalizing the C++'0x standard.
278 <!-- *********************************************************************** -->
279 <div class="doc_section">
280 <a name="externalproj">External Open Source Projects Using LLVM 2.8</a>
282 <!-- *********************************************************************** -->
284 <div class="doc_text">
286 <p>An exciting aspect of LLVM is that it is used as an enabling technology for
287 a lot of other language and tools projects. This section lists some of the
288 projects that have already been updated to work with LLVM 2.8.</p>
291 <!--=========================================================================-->
292 <div class="doc_subsection">
293 <a name="tce">TTA-based Codesign Environment (TCE)</a>
296 <div class="doc_text">
298 <a href="http://tce.cs.tut.fi/">TCE</a> is a toolset for designing
299 application-specific processors (ASP) based on the Transport triggered
300 architecture (TTA). The toolset provides a complete co-design flow from C/C++
301 programs down to synthesizable VHDL and parallel program binaries. Processor
302 customization points include the register files, function units, supported
303 operations, and the interconnection network.</p>
305 <p>TCE uses llvm-gcc/Clang and LLVM for C/C++ language support, target
306 independent optimizations and also for parts of code generation. It generates
307 new LLVM-based code generators "on the fly" for the designed TTA processors and
308 loads them in to the compiler backend as runtime libraries to avoid per-target
309 recompilation of larger parts of the compiler chain.</p>
313 <!--=========================================================================-->
314 <div class="doc_subsection">
315 <a name="Horizon">Horizon Bytecode Compiler</a>
318 <div class="doc_text">
320 <a href="http://www.quokforge.org/projects/horizon">Horizon</a> is a bytecode
321 language and compiler written on top of LLVM, intended for producing
322 single-address-space managed code operating systems that
323 run faster than the equivalent multiple-address-space C systems.
324 More in-depth blurb is available on <a
325 href="http://www.quokforge.org/projects/horizon/wiki/Wiki">the wiki</a>.</p>
329 <!--=========================================================================-->
330 <div class="doc_subsection">
331 <a name="clamav">Clam AntiVirus</a>
334 <div class="doc_text">
336 <a href=http://www.clamav.net>Clam AntiVirus</a> is an open source (GPL)
337 anti-virus toolkit for UNIX, designed especially for e-mail scanning on mail
338 gateways. Since version 0.96 it has <a
339 href="http://vrt-sourcefire.blogspot.com/2010/09/introduction-to-clamavs-low-level.html">bytecode
340 signatures</a> that allow writing detections for complex malware. It
341 uses LLVM's JIT to speed up the execution of bytecode on
342 X86,X86-64,PPC32/64, falling back to its own interpreter otherwise.
343 The git version was updated to work with LLVM 2.8
347 href="http://git.clamav.net/gitweb?p=clamav-bytecode-compiler.git;a=blob_plain;f=docs/user/clambc-user.pdf">
348 ClamAV bytecode compiler</a> uses Clang and LLVM to compile a C-like
349 language, insert runtime checks, and generate ClamAV bytecode.</p>
353 <!--=========================================================================-->
354 <div class="doc_subsection">
355 <a name="pure">Pure</a>
358 <div class="doc_text">
360 <a href="http://pure-lang.googlecode.com/">Pure</a>
361 is an algebraic/functional
362 programming language based on term rewriting. Programs are collections
363 of equations which are used to evaluate expressions in a symbolic
364 fashion. Pure offers dynamic typing, eager and lazy evaluation, lexical
365 closures, a hygienic macro system (also based on term rewriting),
366 built-in list and matrix support (including list and matrix
367 comprehensions) and an easy-to-use C interface. The interpreter uses
368 LLVM as a backend to JIT-compile Pure programs to fast native code.</p>
370 <p>Pure versions 0.44 and later have been tested and are known to work with
371 LLVM 2.8 (and continue to work with older LLVM releases >= 2.5).</p>
375 <!--=========================================================================-->
376 <div class="doc_subsection">
377 <a name="GHC">Glasgow Haskell Compiler (GHC)</a>
380 <div class="doc_text">
382 <a href="http://www.haskell.org/ghc/">GHC</a> is an open source,
383 state-of-the-art programming suite for
384 Haskell, a standard lazy functional programming language. It includes
385 an optimizing static compiler generating good code for a variety of
386 platforms, together with an interactive system for convenient, quick
389 <p>In addition to the existing C and native code generators, GHC 7.0 now
391 href="http://hackage.haskell.org/trac/ghc/wiki/Commentary/Compiler/Backends/LLVM">LLVM
392 code generator</a>. GHC supports LLVM 2.7 and later.</p>
396 <!--=========================================================================-->
397 <div class="doc_subsection">
398 <a name="Clay">Clay Programming Language</a>
401 <div class="doc_text">
403 <a href="http://tachyon.in/clay/">Clay</a> is a new systems programming
404 language that is specifically designed for generic programming. It makes
405 generic programming very concise thanks to whole program type propagation. It
406 uses LLVM as its backend.</p>
410 <!--=========================================================================-->
411 <div class="doc_subsection">
412 <a name="llvm-py">llvm-py Python Bindings for LLVM</a>
415 <div class="doc_text">
417 <a href="http://www.mdevan.org/llvm-py/">llvm-py</a> has been updated to work
418 with LLVM 2.8. llvm-py provides Python bindings for LLVM, allowing you to write a
419 compiler backend or a VM in Python.</p>
424 <!--=========================================================================-->
425 <div class="doc_subsection">
426 <a name="FAUST">FAUST Real-Time Audio Signal Processing Language</a>
429 <div class="doc_text">
431 <a href="http://faust.grame.fr">FAUST</a> is a compiled language for real-time
432 audio signal processing. The name FAUST stands for Functional AUdio STream. Its
433 programming model combines two approaches: functional programming and block
434 diagram composition. In addition with the C, C++, JAVA output formats, the
435 Faust compiler can now generate LLVM bitcode, and works with LLVM 2.7 and
440 <!--=========================================================================-->
441 <div class="doc_subsection">
442 <a name="jade">Jade Just-in-time Adaptive Decoder Engine</a>
445 <div class="doc_text">
447 href="http://sourceforge.net/apps/trac/orcc/wiki/JadeDocumentation">Jade</a>
448 (Just-in-time Adaptive Decoder Engine) is a generic video decoder engine using
449 LLVM for just-in-time compilation of video decoder configurations. Those
450 configurations are designed by MPEG Reconfigurable Video Coding (RVC) committee.
451 MPEG RVC standard is built on a stream-based dataflow representation of
452 decoders. It is composed of a standard library of coding tools written in
453 RVC-CAL language and a dataflow configuration — block diagram —
456 <p>Jade project is hosted as part of the <a href="http://orcc.sf.net">Open
457 RVC-CAL Compiler</a> and requires it to translate the RVC-CAL standard library
458 of video coding tools into an LLVM assembly code.</p>
462 <!--=========================================================================-->
463 <div class="doc_subsection">
464 <a name="neko_llvm_jit">LLVM JIT for Neko VM</a>
467 <div class="doc_text">
468 <p><a href="http://github.com/vava/neko_llvm_jit">Neko LLVM JIT</a>
469 replaces the standard Neko JIT with an LLVM-based implementation. While not
470 fully complete, it is already providing a 1.5x speedup on 64-bit systems.
471 Neko LLVM JIT requires LLVM 2.8 or later.</p>
475 <!--=========================================================================-->
476 <div class="doc_subsection">
477 <a name="crack">Crack Scripting Language</a>
480 <div class="doc_text">
482 <a href="http://code.google.com/p/crack-language/">Crack</a> aims to provide
483 the ease of development of a scripting language with the performance of a
484 compiled language. The language derives concepts from C++, Java and Python,
485 incorporating object-oriented programming, operator overloading and strong
486 typing. Crack 0.2 works with LLVM 2.7, and the forthcoming Crack 0.2.1 release
487 builds on LLVM 2.8.</p>
491 <!--=========================================================================-->
492 <div class="doc_subsection">
493 <a name="DresdenTM">Dresden TM Compiler (DTMC)</a>
496 <div class="doc_text">
498 <a href="http://tm.inf.tu-dresden.de">DTMC</a> provides support for
499 Transactional Memory, which is an easy-to-use and efficient way to synchronize
500 accesses to shared memory. Transactions can contain normal C/C++ code (e.g.,
501 __transaction { list.remove(x); x.refCount--; }) and will be executed
502 virtually atomically and isolated from other transactions.</p>
506 <!--=========================================================================-->
507 <div class="doc_subsection">
508 <a name="Kai">Kai Programming Language</a>
511 <div class="doc_text">
513 <a href="http://www.oriontransfer.co.nz/research/kai">Kai</a> (Japanese 会 for
514 meeting/gathering) is an experimental interpreter that provides a highly
515 extensible runtime environment and explicit control over the compilation
516 process. Programs are defined using nested symbolic expressions, which are all
517 parsed into first-class values with minimal intrinsic semantics. Kai can
518 generate optimised code at run-time (using LLVM) in order to exploit the nature
519 of the underlying hardware and to integrate with external software libraries.
520 It is a unique exploration into world of dynamic code compilation, and the
521 interaction between high level and low level semantics.</p>
525 <!--=========================================================================-->
526 <div class="doc_subsection">
527 <a name="OSL">OSL: Open Shading Language</a>
530 <div class="doc_text">
532 <a href="http://code.google.com/p/openshadinglanguage/">OSL</a> is a shading
533 language designed for use in physically based renderers and in particular
534 production rendering. By using LLVM instead of the interpreter, it was able to
535 meet its performance goals (>= C-code) while retaining the benefits of
536 runtime specialization and a portable high-level language.
543 <!-- *********************************************************************** -->
544 <div class="doc_section">
545 <a name="whatsnew">What's New in LLVM 2.8?</a>
547 <!-- *********************************************************************** -->
549 <div class="doc_text">
551 <p>This release includes a huge number of bug fixes, performance tweaks and
552 minor improvements. Some of the major improvements and new features are listed
558 <!--=========================================================================-->
559 <div class="doc_subsection">
560 <a name="majorfeatures">Major New Features</a>
563 <div class="doc_text">
565 <p>LLVM 2.8 includes several major new capabilities:</p>
568 <li>As mentioned above, <a href="#libc++">libc++</a> and <a
569 href="#lldb">LLDB</a> are major new additions to the LLVM collective.</li>
570 <li>LLVM 2.8 now has pretty decent support for debugging optimized code. You
571 should be able to reliably get debug info for function arguments, assuming
572 that the value is actually available where you have stopped.</li>
574 <li>A new 'llvm-diff' tool is available that does a semantic diff of .ll
576 <li>The <a href="#mc">MC subproject</a> has made major progress in this release.
577 Direct .o file writing support for darwin/x86[-64] is now reliable and
578 support for other targets and object file formats are in progress.</li>
583 <!--=========================================================================-->
584 <div class="doc_subsection">
585 <a name="coreimprovements">LLVM IR and Core Improvements</a>
588 <div class="doc_text">
589 <p>LLVM IR has several new features for better support of new targets and that
590 expose new optimization opportunities:</p>
593 <li>The <a href="LangRef.html#int_libc">memcpy, memmove, and memset</a>
594 intrinsics now take address space qualified pointers and a bit to indicate
595 whether the transfer is "<a href="LangRef.html#volatile">volatile</a>" or not.
597 <li>Per-instruction debug info metadata is much faster and uses less memory by
598 using the new DebugLoc class.</li>
599 <li>LLVM IR now has a more formalized concept of "<a
600 href="LangRef.html#trapvalues">trap values</a>", which allow the optimizer
601 to optimize more aggressively in the presence of undefined behavior, while
602 still producing predictable results.</li>
603 <li>LLVM IR now supports two new <a href="LangRef.html#linkage">linkage
604 types</a> (linker_private_weak and linker_private_weak_def_auto) which map
605 onto some obscure MachO concepts.</li>
610 <!--=========================================================================-->
611 <div class="doc_subsection">
612 <a name="optimizer">Optimizer Improvements</a>
615 <div class="doc_text">
617 <p>In addition to a large array of minor performance tweaks and bug fixes, this
618 release includes a few major enhancements and additions to the optimizers:</p>
621 <li>As mentioned above, the optimizer now has support for updating debug
622 information as it goes. A key aspect of this is the new <a
623 href="SourceLevelDebugging.html#format_common_value">llvm.dbg.value</a>
624 intrinsic. This intrinsic represents debug info for variables that are
625 promoted to SSA values (typically by mem2reg or the -scalarrepl passes).</li>
627 <li>The JumpThreading pass is now much more aggressive about implied value
628 relations, allowing it to thread conditions like "a == 4" when a is known to
629 be 13 in one of the predecessors of a block. It does this in conjunction
630 with the new LazyValueInfo analysis pass.</li>
631 <li>The new RegionInfo analysis pass identifies single-entry single-exit regions
632 in the CFG. You can play with it with the "opt -regions analyze" or
633 "opt -view-regions" commands.</li>
634 <li>The loop optimizer has significantly improved strength reduction and analysis
635 capabilities. Notably it is able to build on the trap value and signed
636 integer overflow information to optimize <= and >= loops.</li>
637 <li>The CallGraphSCCPassManager now has some basic support for iterating within
638 an SCC when a optimizer devirtualizes a function call. This allows inlining
639 through indirect call sites that are devirtualized by store-load forwarding
640 and other optimizations.</li>
641 <li>The new <A href="Passes.html#loweratomic">-loweratomic</a> pass is available
642 to lower atomic instructions into their non-atomic form. This can be useful
643 to optimize generic code that expects to run in a single-threaded
648 <p>In addition to these features that are done in 2.8, there is preliminary
649 support in the release for Type Based Alias Analysis
650 Preliminary work on TBAA but not usable in 2.8.
651 New CorrelatedValuePropagation pass, not on by default in 2.8 yet.
656 <!--=========================================================================-->
657 <div class="doc_subsection">
658 <a name="mc">MC Level Improvements</a>
661 <div class="doc_text">
663 The LLVM Machine Code (aka MC) subsystem was created to solve a number
664 of problems in the realm of assembly, disassembly, object file format handling,
665 and a number of other related areas that CPU instruction-set level tools work
668 <p>The MC subproject has made great leaps in LLVM 2.8. For example, support for
669 directly writing .o files from LLC (and clang) now works reliably for
670 darwin/x86[-64] (including inline assembly support) and the integrated
671 assembler is turned on by default in Clang for these targets. This provides
672 improved compile times among other things.</p>
675 <li>The entire compiler has converted over to using the MCStreamer assembler API
676 instead of writing out a .s file textually.</li>
677 <li>The "assembler parser" is far more mature than in 2.7, supporting a full
678 complement of directives, now supports assembler macros, etc.</li>
679 <li>The "assembler backend" has been completed, including support for relaxation
680 relocation processing and all the other things that an assembler does.</li>
681 <li>The MachO file format support is now fully functional and works.</li>
682 <li>The MC disassembler now fully supports ARM and Thumb. ARM assembler support
683 is still in early development though.</li>
684 <li>The X86 MC assembler now supports the X86 AES and AVX instruction set.</li>
685 <li>Work on ELF and COFF object files and ARM target support is well underway,
686 but isn't useful yet in LLVM 2.8. Please contact the llvmdev mailing list
687 if you're interested in this.</li>
690 <p>For more information, please see the <a
691 href="http://blog.llvm.org/2010/04/intro-to-llvm-mc-project.html">Intro to the
692 LLVM MC Project Blog Post</a>.
698 <!--=========================================================================-->
699 <div class="doc_subsection">
700 <a name="codegen">Target Independent Code Generator Improvements</a>
703 <div class="doc_text">
705 <p>We have put a significant amount of work into the code generator
706 infrastructure, which allows us to implement more aggressive algorithms and make
710 <li>The clang/gcc -momit-leaf-frame-pointer argument is now supported.</li>
711 <li>The clang/gcc -ffunction-sections and -fdata-sections arguments are now
712 supported on ELF targets (like GCC).</li>
713 <li>The MachineCSE pass is now tuned and on by default. It eliminates common
714 subexpressions that are exposed when lowering to machine instructions.</li>
715 <li>The "local" register allocator was replaced by a new "fast" register
716 allocator. This new allocator (which is often used at -O0) is substantially
717 faster and produces better code than the old local register allocator.</li>
718 <li>A new LLC "-regalloc=default" option is available, which automatically
719 chooses a register allocator based on the -O optimization level.</li>
720 <li>The common code generator code was modified to promote illegal argument and
721 return value vectors to wider ones when possible instead of scalarizing
722 them. For example, <3 x float> will now pass in one SSE register
723 instead of 3 on X86. This generates substantially better code since the
724 rest of the code generator was already expecting this.</li>
725 <li>The code generator uses a new "COPY" machine instruction. This speeds up
726 the code generator and eliminates the need for targets to implement the
727 isMoveInstr hook. Also, the copyRegToReg hook was renamed to copyPhysReg
729 <li>The code generator now has a "LocalStackSlotPass", which optimizes stack
730 slot access for targets (like ARM) that have limited stack displacement
732 <li>A new "PeepholeOptimizer" is available, which eliminates sign and zero
733 extends, and optimizes away compare instructions when the condition result
734 is available from a previous instruction.</li>
735 <li>Atomic operations now get legalized into simpler atomic operations if not
736 natively supported, easing the implementation burden on targets.</li>
737 <li>The bottom-up pre-allocation scheduler is now register pressure aware,
738 allowing it to avoid overscheduling in high pressure situations while still
739 aggressively scheduling when registers are available.</li>
740 <li>A new instruction-level-parallelism pre-allocation scheduler is available,
741 which is also register pressure aware. This scheduler has shown substantial
742 wins on X86-64 and is on by default.</li>
743 <li>The tblgen type inference algorithm was rewritten to be more consistent and
744 diagnose more target bugs. If you have an out-of-tree backend, you may
745 find that it finds bugs in your target description. This support also
746 allows limited support for writing patterns for instructions that return
747 multiple results (e.g. a virtual register and a flag result). The
748 'parallel' modifier in tblgen was removed, you should use the new support
749 for multiple results instead.</li>
750 <li>A new (experimental) "-rendermf" pass is available which renders a
751 MachineFunction into HTML, showing live ranges and other useful
753 <li>The new SubRegIndex tablegen class allows subregisters to be indexed
754 symbolically instead of numerically. If your target uses subregisters you
755 will need to adapt to use SubRegIndex when you upgrade to 2.8.</li>
758 <li>The -fast-isel instruction selection path (used at -O0 on X86) was rewritten
759 to work bottom-up on basic blocks instead of top down. This makes it
760 slightly faster (because the MachineDCE pass is not needed any longer) and
761 allows it to generate better code in some cases.</li>
766 <!--=========================================================================-->
767 <div class="doc_subsection">
768 <a name="x86">X86-32 and X86-64 Target Improvements</a>
771 <div class="doc_text">
772 <p>New features and major changes in the X86 target include:
776 <li>The X86 backend now supports holding X87 floating point stack values
777 in registers across basic blocks, dramatically improving performance of code
778 that uses long double, and when targetting CPUs that don't support SSE.</li>
780 <li>The X86 backend now uses a SSEDomainFix pass to optimize SSE operations. On
781 Nehalem ("Core i7") and newer CPUs there is a 2 cycle latency penalty on
782 using a register in a different domain than where it was defined. This pass
783 optimizes away these stalls.</li>
785 <li>The X86 backend now promotes 16-bit integer operations to 32-bits when
786 possible. This avoids 0x66 prefixes, which are slow on some
787 microarchitectures and bloat the code on all of them.</li>
789 <li>The X86 backend now supports the Microsoft "thiscall" calling convention,
790 and a <a href="LangRef.html#callingconv">calling convention</a> to support
791 <a href="#GHC">ghc</a>.</li>
793 <li>The X86 backend supports a new "llvm.x86.int" intrinsic, which maps onto
794 the X86 "int $42" and "int3" instructions.</li>
796 <li>At the IR level, the <2 x float> datatype is now promoted and passed
797 around as a <4 x float> instead of being passed and returned as an MMX
798 vector. If you have a frontend that uses this, please pass and return a
799 <2 x i32> instead (using bitcasts).</li>
801 <li>When printing .s files in verbose assembly mode (the default for clang -S),
802 the X86 backend now decodes X86 shuffle instructions and prints human
803 readable comments after the most inscrutible of them, e.g.:
806 insertps $113, %xmm3, %xmm0 <i># xmm0 = zero,xmm0[1,2],xmm3[1]</i>
807 unpcklps %xmm1, %xmm0 <i># xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]</i>
808 pshufd $1, %xmm1, %xmm1 <i># xmm1 = xmm1[1,0,0,0]</i>
816 <!--=========================================================================-->
817 <div class="doc_subsection">
818 <a name="ARM">ARM Target Improvements</a>
821 <div class="doc_text">
822 <p>New features of the ARM target include:
826 <li>The ARM backend now optimizes tail calls into jumps.</li>
827 <li>Scheduling is improved through the new list-hybrid scheduler as well
828 as through better modeling of structural hazards.</li>
829 <li><a href="LangRef.html#int_fp16">Half float</a> instructions are now
831 <li>NEON support has been improved to model instructions which operate onto
832 multiple consecutive registers more aggressively. This avoids lots of
833 extraneous register copies.</li>
834 <li>The ARM backend now uses a new "ARMGlobalMerge" pass, which merges several
835 global variables into one, saving extra address computation (all the global
836 variables can be accessed via same base address) and potentially reducing
837 register pressure.</li>
839 <li>The ARM has received many minor improvements and tweaks which lead to
840 substantially better performance in a wide range of different scenarios.</li>
842 <li>The ARM NEON intrinsics have been substantially reworked to reduce
843 redundancy and improve code generation. Some of the major changes are:
846 All of the NEON load and store intrinsics (llvm.arm.neon.vld* and
847 llvm.arm.neon.vst*) take an extra parameter to specify the alignment in bytes
848 of the memory being accessed.
851 The llvm.arm.neon.vaba intrinsic (vector absolute difference and
852 accumulate) has been removed. This operation is now represented using
853 the llvm.arm.neon.vabd intrinsic (vector absolute difference) followed by a
857 The llvm.arm.neon.vabdl and llvm.arm.neon.vabal intrinsics (lengthening
858 vector absolute difference with and without accumlation) have been removed.
859 They are represented using the llvm.arm.neon.vabd intrinsic (vector absolute
860 difference) followed by a vector zero-extend operation, and for vabal,
864 The llvm.arm.neon.vmovn intrinsic has been removed. Calls of this intrinsic
865 are now replaced by vector truncate operations.
868 The llvm.arm.neon.vmovls and llvm.arm.neon.vmovlu intrinsics have been
869 removed. They are now represented as vector sign-extend (vmovls) and
870 zero-extend (vmovlu) operations.
873 The llvm.arm.neon.vaddl*, llvm.arm.neon.vaddw*, llvm.arm.neon.vsubl*, and
874 llvm.arm.neon.vsubw* intrinsics (lengthening vector add and subtract) have
875 been removed. They are replaced by vector add and vector subtract operations
876 where one (vaddw, vsubw) or both (vaddl, vsubl) of the operands are either
877 sign-extended or zero-extended.
880 The llvm.arm.neon.vmulls, llvm.arm.neon.vmullu, llvm.arm.neon.vmlal*, and
881 llvm.arm.neon.vmlsl* intrinsics (lengthening vector multiply with and without
882 accumulation and subtraction) have been removed. These operations are now
883 represented as vector multiplications where the operands are either
884 sign-extended or zero-extended, followed by a vector add for vmlal or a
885 vector subtract for vmlsl. Note that the polynomial vector multiply
886 intrinsic, llvm.arm.neon.vmullp, remains unchanged.
895 <!--=========================================================================-->
896 <div class="doc_subsection">
897 <a name="changes">Major Changes and Removed Features</a>
900 <div class="doc_text">
902 <p>If you're already an LLVM user or developer with out-of-tree changes based
903 on LLVM 2.7, this section lists some "gotchas" that you may run into upgrading
904 from the previous release.</p>
907 <li>The build configuration machinery changed the output directory names. It
908 wasn't clear to many people that a "Release-Asserts" build was a release build
909 without asserts. To make this more clear, "Release" does not include
910 assertions and "Release+Asserts" does (likewise, "Debug" and
911 "Debug+Asserts").</li>
912 <li>The MSIL Backend was removed, it was unsupported and broken.</li>
913 <li>The ABCD, SSI, and SCCVN passes were removed. These were not fully
914 functional and their behavior has been or will be subsumed by the
915 LazyValueInfo pass.</li>
916 <li>The LLVM IR 'Union' feature was removed. While this is a desirable feature
917 for LLVM IR to support, the existing implementation was half baked and
918 barely useful. We'd really like anyone interested to resurrect the work and
919 finish it for a future release.</li>
920 <li>If you're used to reading .ll files, you'll probably notice that .ll file
921 dumps don't produce #uses comments anymore. To get them, run a .bc file
922 through "llvm-dis --show-annotations".</li>
923 <li>Target triples are now stored in a normalized form, and all inputs from
924 humans are expected to be normalized by Triple::normalize before being
925 stored in a module triple or passed to another library.</li>
930 <p>In addition, many APIs have changed in this release. Some of the major LLVM
933 <li>LLVM 2.8 changes the internal order of operands in <a
934 href="http://llvm.org/doxygen/classllvm_1_1InvokeInst.html"><tt>InvokeInst</tt></a>
935 and <a href="http://llvm.org/doxygen/classllvm_1_1CallInst.html"><tt>CallInst</tt></a>.
936 To be portable across releases, please use the <tt>CallSite</tt> class and the
937 high-level accessors, such as <tt>getCalledValue</tt> and
938 <tt>setUnwindDest</tt>.
941 You can no longer pass use_iterators directly to cast<> (and similar),
942 because these routines tend to perform costly dereference operations more
943 than once. You have to dereference the iterators yourself and pass them in.
946 llvm.memcpy.*, llvm.memset.*, llvm.memmove.* intrinsics take an extra
947 parameter now ("i1 isVolatile"), totaling 5 parameters, and the pointer
948 operands are now address-space qualified.
949 If you were creating these intrinsic calls and prototypes yourself (as opposed
950 to using Intrinsic::getDeclaration), you can use
951 UpgradeIntrinsicFunction/UpgradeIntrinsicCall to be portable accross releases.
954 SetCurrentDebugLocation takes a DebugLoc now instead of a MDNode.
955 Change your code to use
956 SetCurrentDebugLocation(DebugLoc::getFromDILocation(...)).
959 The <tt>RegisterPass</tt> and <tt>RegisterAnalysisGroup</tt> templates are
960 considered deprecated, but continue to function in LLVM 2.8. Clients are
961 strongly advised to use the upcoming <tt>INITIALIZE_PASS()</tt> and
962 <tt>INITIALIZE_AG_PASS()</tt> macros instead.
965 The constructor for the Triple class no longer tries to understand odd triple
966 specifications. Frontends should ensure that they only pass valid triples to
967 LLVM. The Triple::normalize utility method has been added to help front-ends
968 deal with funky triples.
972 Some APIs got renamed:
974 <li>llvm_report_error -> report_fatal_error</li>
975 <li>llvm_install_error_handler -> install_fatal_error_handler</li>
976 <li>llvm::DwarfExceptionHandling -> llvm::JITExceptionHandling</li>
977 <li>VISIBILITY_HIDDEN -> LLVM_LIBRARY_VISIBILITY</li>
986 <!-- *********************************************************************** -->
987 <div class="doc_section">
988 <a name="knownproblems">Known Problems</a>
990 <!-- *********************************************************************** -->
992 <div class="doc_text">
994 <p>This section contains significant known problems with the LLVM system,
995 listed by component. If you run into a problem, please check the <a
996 href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
997 there isn't already one.</p>
1001 <!-- ======================================================================= -->
1002 <div class="doc_subsection">
1003 <a name="experimental">Experimental features included with this release</a>
1006 <div class="doc_text">
1008 <p>The following components of this LLVM release are either untested, known to
1009 be broken or unreliable, or are in early development. These components should
1010 not be relied on, and bugs should not be filed against them, but they may be
1011 useful to some people. In particular, if you would like to work on one of these
1012 components, please contact us on the <a
1013 href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev list</a>.</p>
1016 <li>The Alpha, Blackfin, CellSPU, MicroBlaze, MSP430, MIPS, PIC16, SystemZ
1017 and XCore backends are experimental.</li>
1018 <li><tt>llc</tt> "<tt>-filetype=obj</tt>" is experimental on all targets
1019 other than darwin-i386 and darwin-x86_64.</li>
1024 <!-- ======================================================================= -->
1025 <div class="doc_subsection">
1026 <a name="x86-be">Known problems with the X86 back-end</a>
1029 <div class="doc_text">
1032 <li>The X86 backend does not yet support
1033 all <a href="http://llvm.org/PR879">inline assembly that uses the X86
1034 floating point stack</a>. It supports the 'f' and 't' constraints, but not
1036 <li>Win64 code generation wasn't widely tested. Everything should work, but we
1037 expect small issues to happen. Also, llvm-gcc cannot build the mingw64
1038 runtime currently due to lack of support for the 'u' inline assembly
1039 constraint and for X87 floating point inline assembly.</li>
1040 <li>The X86-64 backend does not yet support the LLVM IR instruction
1041 <tt>va_arg</tt>. Currently, front-ends support variadic
1042 argument constructs on X86-64 by lowering them manually.</li>
1047 <!-- ======================================================================= -->
1048 <div class="doc_subsection">
1049 <a name="ppc-be">Known problems with the PowerPC back-end</a>
1052 <div class="doc_text">
1055 <li>The Linux PPC32/ABI support needs testing for the interpreter and static
1056 compilation, and lacks support for debug information.</li>
1061 <!-- ======================================================================= -->
1062 <div class="doc_subsection">
1063 <a name="arm-be">Known problems with the ARM back-end</a>
1066 <div class="doc_text">
1069 <li>Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6
1070 processors, thumb programs can crash or produce wrong
1071 results (<a href="http://llvm.org/PR1388">PR1388</a>).</li>
1072 <li>Compilation for ARM Linux OABI (old ABI) is supported but not fully tested.
1078 <!-- ======================================================================= -->
1079 <div class="doc_subsection">
1080 <a name="sparc-be">Known problems with the SPARC back-end</a>
1083 <div class="doc_text">
1086 <li>The SPARC backend only supports the 32-bit SPARC ABI (-m32); it does not
1087 support the 64-bit SPARC ABI (-m64).</li>
1092 <!-- ======================================================================= -->
1093 <div class="doc_subsection">
1094 <a name="mips-be">Known problems with the MIPS back-end</a>
1097 <div class="doc_text">
1100 <li>64-bit MIPS targets are not supported yet.</li>
1105 <!-- ======================================================================= -->
1106 <div class="doc_subsection">
1107 <a name="alpha-be">Known problems with the Alpha back-end</a>
1110 <div class="doc_text">
1114 <li>On 21164s, some rare FP arithmetic sequences which may trap do not have the
1115 appropriate nops inserted to ensure restartability.</li>
1120 <!-- ======================================================================= -->
1121 <div class="doc_subsection">
1122 <a name="c-be">Known problems with the C back-end</a>
1125 <div class="doc_text">
1127 <p>The C backend has numerous problems and is not being actively maintained.
1128 Depending on it for anything serious is not advised.</p>
1131 <li><a href="http://llvm.org/PR802">The C backend has only basic support for
1132 inline assembly code</a>.</li>
1133 <li><a href="http://llvm.org/PR1658">The C backend violates the ABI of common
1134 C++ programs</a>, preventing intermixing between C++ compiled by the CBE and
1135 C++ code compiled with <tt>llc</tt> or native compilers.</li>
1136 <li>The C backend does not support all exception handling constructs.</li>
1137 <li>The C backend does not support arbitrary precision integers.</li>
1143 <!-- ======================================================================= -->
1144 <div class="doc_subsection">
1145 <a name="llvm-gcc">Known problems with the llvm-gcc front-end</a>
1148 <div class="doc_text">
1150 <p>llvm-gcc is generally very stable for the C family of languages. The only
1151 major language feature of GCC not supported by llvm-gcc is the
1152 <tt>__builtin_apply</tt> family of builtins. However, some extensions
1153 are only supported on some targets. For example, trampolines are only
1154 supported on some targets (these are used when you take the address of a
1155 nested function).</p>
1157 <p>Fortran support generally works, but there are still several unresolved bugs
1158 in <a href="http://llvm.org/bugs/">Bugzilla</a>. Please see the
1159 tools/gfortran component for details. Note that llvm-gcc is missing major
1160 Fortran performance work in the frontend and library that went into GCC after
1161 4.2. If you are interested in Fortran, we recommend that you consider using
1162 <a href="#dragonegg">dragonegg</a> instead.</p>
1164 <p>The llvm-gcc 4.2 Ada compiler has basic functionality, but is no longer being
1165 actively maintained. If you are interested in Ada, we recommend that you
1166 consider using <a href="#dragonegg">dragonegg</a> instead.</p>
1169 <!-- *********************************************************************** -->
1170 <div class="doc_section">
1171 <a name="additionalinfo">Additional Information</a>
1173 <!-- *********************************************************************** -->
1175 <div class="doc_text">
1177 <p>A wide variety of additional information is available on the <a
1178 href="http://llvm.org">LLVM web page</a>, in particular in the <a
1179 href="http://llvm.org/docs/">documentation</a> section. The web page also
1180 contains versions of the API documentation which is up-to-date with the
1181 Subversion version of the source code.
1182 You can access versions of these documents specific to this release by going
1183 into the "<tt>llvm/doc/</tt>" directory in the LLVM tree.</p>
1185 <p>If you have any questions or comments about LLVM, please feel free to contact
1186 us via the <a href="http://llvm.org/docs/#maillist"> mailing
1191 <!-- *********************************************************************** -->
1195 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1196 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1197 <a href="http://validator.w3.org/check/referer"><img
1198 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1200 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
1201 Last modified: $Date$