blob: 19d1902090a3b0129368d8f43d7f24bf229557e6 [file] [log] [blame]
Misha Brukman500bc302003-11-22 00:38:41 +00001<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
3<html>
4<head>
Reid Spencer6454ed32004-11-18 18:38:58 +00005 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
Chris Lattner2fb6e5c2010-10-03 23:09:03 +00006 <meta encoding="utf8">
Misha Brukman500bc302003-11-22 00:38:41 +00007 <link rel="stylesheet" href="llvm.css" type="text/css">
Dan Gohmanb44f6c62010-05-03 23:51:05 +00008 <title>LLVM 2.8 Release Notes</title>
Misha Brukman500bc302003-11-22 00:38:41 +00009</head>
10<body>
Chris Lattner79c3fe12003-10-02 04:57:28 +000011
Dan Gohmanb44f6c62010-05-03 23:51:05 +000012<div class="doc_title">LLVM 2.8 Release Notes</div>
Mikhail Glushenkovea65d7d2008-10-13 02:08:34 +000013
Chris Lattner0e464a92010-03-17 04:02:39 +000014<img align=right src="http://llvm.org/img/DragonSmall.png"
Gabor Greifee2187a2010-04-22 10:21:43 +000015 width="136" height="136" alt="LLVM Dragon Logo">
Chris Lattner0e464a92010-03-17 04:02:39 +000016
Chris Lattner79c3fe12003-10-02 04:57:28 +000017<ol>
Misha Brukman500bc302003-11-22 00:38:41 +000018 <li><a href="#intro">Introduction</a></li>
Chris Lattner96a445e2008-10-13 18:01:01 +000019 <li><a href="#subproj">Sub-project Status Update</a></li>
Dan Gohmanb44f6c62010-05-03 23:51:05 +000020 <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>
Chris Lattner4b538b92004-04-30 22:17:12 +000022 <li><a href="GettingStarted.html">Installation Instructions</a></li>
Dan Gohman44aa9212008-10-14 16:23:02 +000023 <li><a href="#knownproblems">Known Problems</a></li>
Misha Brukman500bc302003-11-22 00:38:41 +000024 <li><a href="#additionalinfo">Additional Information</a></li>
Chris Lattner79c3fe12003-10-02 04:57:28 +000025</ol>
26
Chris Lattner7911ce22004-05-23 21:07:27 +000027<div class="doc_author">
Dan Gohman44aa9212008-10-14 16:23:02 +000028 <p>Written by the <a href="http://llvm.org">LLVM Team</a></p>
Misha Brukman500bc302003-11-22 00:38:41 +000029</div>
Chris Lattner79c3fe12003-10-02 04:57:28 +000030
Chris Lattnere07043c2010-09-29 05:30:03 +000031<!--
Chris Lattner7b91eda2010-04-22 05:41:35 +000032<h1 style="color:red">These are in-progress notes for the upcoming LLVM 2.8
Jeffrey Yasskinbec48772010-01-28 01:14:43 +000033release.<br>
34You may prefer the
Dan Gohmanb44f6c62010-05-03 23:51:05 +000035<a href="http://llvm.org/releases/2.7/docs/ReleaseNotes.html">LLVM 2.7
36Release Notes</a>.</h1>
Chris Lattnere07043c2010-09-29 05:30:03 +000037-->
Jeffrey Yasskinbec48772010-01-28 01:14:43 +000038
Chris Lattner79c3fe12003-10-02 04:57:28 +000039<!-- *********************************************************************** -->
Misha Brukman500bc302003-11-22 00:38:41 +000040<div class="doc_section">
41 <a name="intro">Introduction</a>
42</div>
Chris Lattner79c3fe12003-10-02 04:57:28 +000043<!-- *********************************************************************** -->
44
Misha Brukman500bc302003-11-22 00:38:41 +000045<div class="doc_text">
46
Chris Lattnerf6662f92008-10-13 17:57:36 +000047<p>This document contains the release notes for the LLVM Compiler
Dan Gohman7ae3ac82010-05-03 23:52:21 +000048Infrastructure, release 2.8. Here we describe the status of LLVM, including
Chris Lattnerf6662f92008-10-13 17:57:36 +000049major improvements from the previous release and significant known problems.
Mikhail Glushenkovf795ef02009-03-01 18:09:47 +000050All LLVM releases may be downloaded from the <a
Chris Lattnerf6662f92008-10-13 17:57:36 +000051href="http://llvm.org/releases/">LLVM releases web site</a>.</p>
Chris Lattner19092612003-10-02 16:38:05 +000052
Chris Lattner7506b1d2004-12-07 08:04:13 +000053<p>For more information about LLVM, including information about the latest
Chris Lattnerc463b272005-10-29 07:07:09 +000054release, please check out the <a href="http://llvm.org/">main LLVM
Chris Lattner47ad72c2003-10-07 21:38:31 +000055web site</a>. If you have questions or comments, the <a
Chris Lattnerc66bfef2010-03-17 04:41:49 +000056href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM Developer's
57Mailing List</a> is a good place to send them.</p>
Chris Lattner79c3fe12003-10-02 04:57:28 +000058
Mikhail Glushenkovea65d7d2008-10-13 02:08:34 +000059<p>Note that if you are reading this file from a Subversion checkout or the
Chris Lattner84977642007-09-21 03:54:09 +000060main LLVM web page, this document applies to the <i>next</i> release, not the
Gabor Greiffa933f82008-10-14 11:00:32 +000061current one. To see the release notes for a specific release, please see the
Chris Lattner84977642007-09-21 03:54:09 +000062<a href="http://llvm.org/releases/">releases page</a>.</p>
Misha Brukman500bc302003-11-22 00:38:41 +000063
64</div>
Chris Lattner61358ab2009-10-13 17:48:04 +000065
Chris Lattner79c3fe12003-10-02 04:57:28 +000066
Chris Lattner61358ab2009-10-13 17:48:04 +000067<!--
68Almost dead code.
69 include/llvm/Analysis/LiveValues.h => Dan
70 lib/Transforms/IPO/MergeFunctions.cpp => consider for 2.8.
Chris Lattner61c70e92010-08-28 04:09:24 +000071 GEPSplitterPass
Chris Lattner61358ab2009-10-13 17:48:04 +000072-->
73
74
Chris Lattnere07043c2010-09-29 05:30:03 +000075<!-- Features that need text if they're finished for 2.9:
Chris Lattner914ce462010-04-22 06:28:20 +000076 combiner-aa?
Chris Lattner61358ab2009-10-13 17:48:04 +000077 strong phi elim
Chris Lattner61358ab2009-10-13 17:48:04 +000078 loop dependence analysis
Chris Lattner885b6612010-08-28 16:33:36 +000079 TBAA
Chris Lattnere07043c2010-09-29 05:30:03 +000080 CorrelatedValuePropagation
Chris Lattneracce85d2008-02-10 07:46:44 +000081 -->
Chris Lattnere07043c2010-09-29 05:30:03 +000082
83 <!-- Announcement, lldb, libc++ -->
Chris Lattnerafa41632010-09-29 07:25:03 +000084
Chris Lattneracce85d2008-02-10 07:46:44 +000085
Chris Lattner96a445e2008-10-13 18:01:01 +000086<!-- *********************************************************************** -->
87<div class="doc_section">
88 <a name="subproj">Sub-project Status Update</a>
Chris Lattnerea34f642008-06-08 21:34:41 +000089</div>
Chris Lattner96a445e2008-10-13 18:01:01 +000090<!-- *********************************************************************** -->
Chris Lattnerea34f642008-06-08 21:34:41 +000091
92<div class="doc_text">
Chris Lattner8348b472008-06-08 21:58:17 +000093<p>
Dan Gohmanb44f6c62010-05-03 23:51:05 +000094The LLVM 2.8 distribution currently consists of code from the core LLVM
Chris Lattner61358ab2009-10-13 17:48:04 +000095repository (which roughly includes the LLVM optimizers, code generators
96and supporting tools), the Clang repository and the llvm-gcc repository. In
97addition to this code, the LLVM Project includes other sub-projects that are in
98development. Here we include updates on these subprojects.
Bill Wendling63d8c552009-03-02 04:28:57 +000099</p>
Chris Lattner96a445e2008-10-13 18:01:01 +0000100
Chris Lattner8348b472008-06-08 21:58:17 +0000101</div>
102
Chris Lattner96a445e2008-10-13 18:01:01 +0000103
Chris Lattner8348b472008-06-08 21:58:17 +0000104<!--=========================================================================-->
Chris Lattner96a445e2008-10-13 18:01:01 +0000105<div class="doc_subsection">
Chris Lattnerfb97b2d2008-10-13 18:11:54 +0000106<a name="clang">Clang: C/C++/Objective-C Frontend Toolkit</a>
Chris Lattner8348b472008-06-08 21:58:17 +0000107</div>
108
109<div class="doc_text">
110
Chris Lattner095539f2010-04-26 17:42:18 +0000111<p><a href="http://clang.llvm.org/">Clang</a> is an LLVM front end for the C,
112C++, and Objective-C languages. Clang aims to provide a better user experience
113through expressive diagnostics, a high level of conformance to language
114standards, fast compilation, and low memory use. Like LLVM, Clang provides a
115modular, library-based architecture that makes it suitable for creating or
116integrating with other development tools. Clang is considered a
Chris Lattnere07043c2010-09-29 05:30:03 +0000117production-quality compiler for C, Objective-C, C++ and Objective-C++ on x86
118(32- and 64-bit), and for darwin-arm targets.</p>
Chris Lattner61358ab2009-10-13 17:48:04 +0000119
Chris Lattner922d00f2010-07-21 15:57:40 +0000120<p>In the LLVM 2.8 time-frame, the Clang team has made many improvements:</p>
Bill Wendling741748a2008-10-27 09:27:33 +0000121
Douglas Gregorf2409d52010-10-04 07:02:35 +0000122 <ul>
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>
Duncan Sands051f2ee2010-10-04 09:11:50 +0000129 <li>Added support for SSE, ARM NEON, and Altivec.</li>
Douglas Gregorf2409d52010-10-04 07:02:35 +0000130 <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>
133 </ul>
Chris Lattnerfb97b2d2008-10-13 18:11:54 +0000134</div>
135
136<!--=========================================================================-->
137<div class="doc_subsection">
138<a name="clangsa">Clang Static Analyzer</a>
139</div>
140
141<div class="doc_text">
142
Chris Lattner8cdd7932010-04-22 06:38:11 +0000143<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>
Chris Lattnercc042612008-10-14 00:52:49 +0000149
Chris Lattner2e38c7f2010-09-30 00:34:43 +0000150<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.
Chris Lattner8cdd7932010-04-22 06:38:11 +0000152</p>
Chris Lattner8348b472008-06-08 21:58:17 +0000153
154</div>
155
Chris Lattner96a445e2008-10-13 18:01:01 +0000156<!--=========================================================================-->
157<div class="doc_subsection">
Duncan Sands749fd832010-04-02 09:23:15 +0000158<a name="dragonegg">DragonEgg: llvm-gcc ported to gcc-4.5</a>
Chris Lattner61358ab2009-10-13 17:48:04 +0000159</div>
160
161<div class="doc_text">
162<p>
Duncan Sands749fd832010-04-02 09:23:15 +0000163<a href="http://dragonegg.llvm.org/">DragonEgg</a> is a port of llvm-gcc to
Duncan Sands4b1da2b2010-09-30 17:37:34 +0000164gcc-4.5. Unlike llvm-gcc, dragonegg in theory does not require any gcc-4.5
165modifications whatsoever (currently one small patch is needed) thanks to the
166new <a href="http://gcc.gnu.org/wiki/plugins">gcc plugin architecture</a>.
167DragonEgg is a gcc plugin that makes gcc-4.5 use the LLVM optimizers and code
168generators instead of gcc's, just like with llvm-gcc.
Duncan Sands749fd832010-04-02 09:23:15 +0000169</p>
170
171<p>
Duncan Sands4b1da2b2010-09-30 17:37:34 +0000172DragonEgg is still a work in progress, but it is able to compile a lot of code,
173for example all of gcc, LLVM and clang. Currently Ada, C, C++ and Fortran work
174well, while all other languages either don't work at all or only work poorly.
175For the moment only the x86-32 and x86-64 targets are supported, and only on
176linux and darwin (darwin may need additional gcc patches).
Duncan Sands749fd832010-04-02 09:23:15 +0000177</p>
178
179<p>
Duncan Sands4b1da2b2010-09-30 17:37:34 +0000180The 2.8 release has the following notable changes:
181<ul>
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,
185resulting in better optimization.</li>
186<li>The '-fplugin-arg-dragonegg-enable-gcc-optzns' flag now runs all gcc
187optimizers, 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>
190</ul>
Chris Lattner61358ab2009-10-13 17:48:04 +0000191</p>
192
193</div>
194
Chris Lattner75547712010-10-03 23:49:06 +0000195<!--=========================================================================-->
196<div class="doc_subsection">
197<a name="vmkit">VMKit: JVM/CLI Virtual Machine Implementation</a>
198</div>
199
200<div class="doc_text">
201<p>
202The <a href="http://vmkit.llvm.org/">VMKit project</a> is an implementation of
203a Java Virtual Machine (Java VM or JVM) that uses LLVM for static and
204just-in-time compilation. As of LLVM 2.8, VMKit now supports copying garbage
205collectors, and can be configured to use MMTk's copy mark-sweep garbage
206collector. In LLVM 2.8, the VMKit .NET VM is no longer being maintained.
207</p>
208</div>
209
210<!--=========================================================================-->
211<div class="doc_subsection">
212<a name="compiler-rt">compiler-rt: Compiler Runtime Library</a>
213</div>
214
215<div class="doc_text">
216<p>
217The new LLVM <a href="http://compiler-rt.llvm.org/">compiler-rt project</a>
218is a simple library that provides an implementation of the low-level
219target-specific hooks required by code generation and other runtime components.
220For example, when compiling for a 32-bit target, converting a double to a 64-bit
221unsigned integer is compiled into a runtime call to the "__fixunsdfdi"
222function. The compiler-rt library provides highly optimized implementations of
223this and other low-level routines (some are 3x faster than the equivalent
224libgcc routines).</p>
225
226<p>
227All of the code in the compiler-rt project is available under the standard LLVM
228License, a "BSD-style" license. New in LLVM 2.8, compiler_rt now supports
229soft floating point (for targets that don't have a real floating point unit),
230and includes an extensive testsuite for the "blocks" language feature and the
231blocks runtime included in compiler_rt.</p>
232
233</div>
Chris Lattner61358ab2009-10-13 17:48:04 +0000234
235<!--=========================================================================-->
236<div class="doc_subsection">
Chris Lattnere07043c2010-09-29 05:30:03 +0000237<a name="lldb">LLDB: Low Level Debugger</a>
238</div>
239
240<div class="doc_text">
241<p>
Chris Lattner75547712010-10-03 23:49:06 +0000242<a href="http://lldb.llvm.org/">LLDB</a> is a brand new member of the LLVM
243umbrella of projects. LLDB is a next generation, high-performance debugger. It
244is built as a set of reusable components which highly leverage existing
245libraries in the larger LLVM Project, such as the Clang expression parser, the
246LLVM disassembler and the LLVM JIT.</p>
Chris Lattnere07043c2010-09-29 05:30:03 +0000247
248<p>
Chris Lattner75547712010-10-03 23:49:06 +0000249LLDB is in early development and not included as part of the LLVM 2.8 release,
250but is mature enough to support basic debugging scenarios on Mac OS X in C,
251Objective-C and C++. We'd really like help extending and expanding LLDB to
252support new platforms, new languages, new architectures, and new features.
Chris Lattnere07043c2010-09-29 05:30:03 +0000253</p>
254
255</div>
256
257<!--=========================================================================-->
258<div class="doc_subsection">
259<a name="libc++">libc++: C++ Standard Library</a>
260</div>
261
262<div class="doc_text">
263<p>
Chris Lattner75547712010-10-03 23:49:06 +0000264<a href="http://libc++.llvm.org/">libc++</a> is another new member of the LLVM
265family. It is an implementation of the C++ standard library, written from the
266ground up to specifically target the forthcoming C++'0X standard and focus on
267delivering great performance.</p>
Chris Lattnere07043c2010-09-29 05:30:03 +0000268
269<p>
Chris Lattner75547712010-10-03 23:49:06 +0000270As of the LLVM 2.8 release, libc++ is virtually feature complete, but would
271benefit from more testing and better integration with Clang++. It is also
272looking forward to the C++ committee finalizing the C++'0x standard.
Chris Lattnere07043c2010-09-29 05:30:03 +0000273</p>
274
275</div>
276
Chris Lattner61358ab2009-10-13 17:48:04 +0000277
Chris Lattnerab68e9e2009-02-26 22:33:38 +0000278<!-- *********************************************************************** -->
279<div class="doc_section">
Chris Lattner922d00f2010-07-21 15:57:40 +0000280 <a name="externalproj">External Open Source Projects Using LLVM 2.8</a>
Chris Lattnerab68e9e2009-02-26 22:33:38 +0000281</div>
282<!-- *********************************************************************** -->
283
Chris Lattner61358ab2009-10-13 17:48:04 +0000284<div class="doc_text">
285
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
Chris Lattner922d00f2010-07-21 15:57:40 +0000288 projects that have already been updated to work with LLVM 2.8.</p>
Chris Lattner7c8e7962010-04-26 17:38:10 +0000289</div>
290
Chris Lattner2e38c7f2010-09-30 00:34:43 +0000291<!--=========================================================================-->
292<div class="doc_subsection">
293<a name="tce">TTA-based Codesign Environment (TCE)</a>
294</div>
295
296<div class="doc_text">
297<p>
298<a href="http://tce.cs.tut.fi/">TCE</a> is a toolset for designing
299application-specific processors (ASP) based on the Transport triggered
300architecture (TTA). The toolset provides a complete co-design flow from C/C++
301programs down to synthesizable VHDL and parallel program binaries. Processor
302customization points include the register files, function units, supported
303operations, and the interconnection network.</p>
304
305<p>TCE uses llvm-gcc/Clang and LLVM for C/C++ language support, target
306independent optimizations and also for parts of code generation. It generates
307new LLVM-based code generators "on the fly" for the designed TTA processors and
308loads them in to the compiler backend as runtime libraries to avoid per-target
309recompilation of larger parts of the compiler chain.</p>
310
311</div>
312
313<!--=========================================================================-->
314<div class="doc_subsection">
315<a name="Horizon">Horizon Bytecode Compiler</a>
316</div>
317
318<div class="doc_text">
319<p>
320<a href="http://www.quokforge.org/projects/horizon">Horizon</a> is a bytecode
321language and compiler written on top of LLVM, intended for producing
322single-address-space managed code operating systems that
323run faster than the equivalent multiple-address-space C systems.
Gabor Greif9f459132010-10-04 17:03:49 +0000324More in-depth blurb is available on the <a
325href="http://www.quokforge.org/projects/horizon/wiki/Wiki">wiki</a>.</p>
Chris Lattner2e38c7f2010-09-30 00:34:43 +0000326
327</div>
328
329<!--=========================================================================-->
330<div class="doc_subsection">
331<a name="clamav">Clam AntiVirus</a>
332</div>
333
334<div class="doc_text">
335<p>
336<a href=http://www.clamav.net>Clam AntiVirus</a> is an open source (GPL)
337anti-virus toolkit for UNIX, designed especially for e-mail scanning on mail
338gateways. Since version 0.96 it has <a
339href="http://vrt-sourcefire.blogspot.com/2010/09/introduction-to-clamavs-low-level.html">bytecode
340signatures</a> that allow writing detections for complex malware. It
341uses LLVM's JIT to speed up the execution of bytecode on
Gabor Greif9f459132010-10-04 17:03:49 +0000342X86, X86-64, PPC32/64, falling back to its own interpreter otherwise.
343The git version was updated to work with LLVM 2.8.
Chris Lattner2e38c7f2010-09-30 00:34:43 +0000344</p>
345
346<p>The <a
347href="http://git.clamav.net/gitweb?p=clamav-bytecode-compiler.git;a=blob_plain;f=docs/user/clambc-user.pdf">
348ClamAV bytecode compiler</a> uses Clang and LLVM to compile a C-like
349language, insert runtime checks, and generate ClamAV bytecode.</p>
350
351</div>
352
353<!--=========================================================================-->
354<div class="doc_subsection">
355<a name="pure">Pure</a>
356</div>
357
358<div class="doc_text">
359<p>
360<a href="http://pure-lang.googlecode.com/">Pure</a>
361is an algebraic/functional
362programming language based on term rewriting. Programs are collections
363of equations which are used to evaluate expressions in a symbolic
364fashion. Pure offers dynamic typing, eager and lazy evaluation, lexical
365closures, a hygienic macro system (also based on term rewriting),
366built-in list and matrix support (including list and matrix
367comprehensions) and an easy-to-use C interface. The interpreter uses
368LLVM as a backend to JIT-compile Pure programs to fast native code.</p>
369
370<p>Pure versions 0.44 and later have been tested and are known to work with
371LLVM 2.8 (and continue to work with older LLVM releases >= 2.5).</p>
372
373</div>
374
375<!--=========================================================================-->
376<div class="doc_subsection">
377<a name="GHC">Glasgow Haskell Compiler (GHC)</a>
378</div>
379
380<div class="doc_text">
381<p>
382<a href="http://www.haskell.org/ghc/">GHC</a> is an open source,
383state-of-the-art programming suite for
384Haskell, a standard lazy functional programming language. It includes
385an optimizing static compiler generating good code for a variety of
386platforms, together with an interactive system for convenient, quick
387development.</p>
388
389<p>In addition to the existing C and native code generators, GHC 7.0 now
390supports an <a
391href="http://hackage.haskell.org/trac/ghc/wiki/Commentary/Compiler/Backends/LLVM">LLVM
392code generator</a>. GHC supports LLVM 2.7 and later.</p>
393
394</div>
395
396<!--=========================================================================-->
397<div class="doc_subsection">
398<a name="Clay">Clay Programming Language</a>
399</div>
400
401<div class="doc_text">
402<p>
Chris Lattner97fe6452010-09-30 01:12:09 +0000403<a href="http://tachyon.in/clay/">Clay</a> is a new systems programming
Chris Lattner2e38c7f2010-09-30 00:34:43 +0000404language that is specifically designed for generic programming. It makes
405generic programming very concise thanks to whole program type propagation. It
406uses LLVM as its backend.</p>
407
408</div>
Chris Lattner3a1d4cf2010-04-22 21:34:16 +0000409
Chris Lattnere0518442010-10-01 06:34:49 +0000410<!--=========================================================================-->
411<div class="doc_subsection">
412<a name="llvm-py">llvm-py Python Bindings for LLVM</a>
413</div>
414
415<div class="doc_text">
416<p>
417<a href="http://www.mdevan.org/llvm-py/">llvm-py</a> has been updated to work
418with LLVM 2.8. llvm-py provides Python bindings for LLVM, allowing you to write a
419compiler backend or a VM in Python.</p>
420
421</div>
422
423
Chris Lattner2fb6e5c2010-10-03 23:09:03 +0000424<!--=========================================================================-->
425<div class="doc_subsection">
426<a name="FAUST">FAUST Real-Time Audio Signal Processing Language</a>
427</div>
428
429<div class="doc_text">
430<p>
431<a href="http://faust.grame.fr">FAUST</a> is a compiled language for real-time
432audio signal processing. The name FAUST stands for Functional AUdio STream. Its
433programming model combines two approaches: functional programming and block
434diagram composition. In addition with the C, C++, JAVA output formats, the
435Faust compiler can now generate LLVM bitcode, and works with LLVM 2.7 and
4362.8.</p>
437
438</div>
439
440<!--=========================================================================-->
441<div class="doc_subsection">
442<a name="jade">Jade Just-in-time Adaptive Decoder Engine</a>
443</div>
444
445<div class="doc_text">
446<p><a
447href="http://sourceforge.net/apps/trac/orcc/wiki/JadeDocumentation">Jade</a>
448(Just-in-time Adaptive Decoder Engine) is a generic video decoder engine using
449LLVM for just-in-time compilation of video decoder configurations. Those
450configurations are designed by MPEG Reconfigurable Video Coding (RVC) committee.
451MPEG RVC standard is built on a stream-based dataflow representation of
452decoders. It is composed of a standard library of coding tools written in
Duncan Sands30be9e42010-10-04 10:04:14 +0000453RVC-CAL language and a dataflow configuration &#8212; block diagram &#8212;
Chris Lattner2fb6e5c2010-10-03 23:09:03 +0000454of a decoder.</p>
455
456<p>Jade project is hosted as part of the <a href="http://orcc.sf.net">Open
457RVC-CAL Compiler</a> and requires it to translate the RVC-CAL standard library
458of video coding tools into an LLVM assembly code.</p>
459
460</div>
461
462<!--=========================================================================-->
463<div class="doc_subsection">
464<a name="neko_llvm_jit">LLVM JIT for Neko VM</a>
465</div>
466
467<div class="doc_text">
468<p><a href="http://github.com/vava/neko_llvm_jit">Neko LLVM JIT</a>
469replaces the standard Neko JIT with an LLVM-based implementation. While not
470fully complete, it is already providing a 1.5x speedup on 64-bit systems.
471Neko LLVM JIT requires LLVM 2.8 or later.</p>
472
473</div>
474
475<!--=========================================================================-->
476<div class="doc_subsection">
477<a name="crack">Crack Scripting Language</a>
478</div>
479
480<div class="doc_text">
481<p>
482<a href="http://code.google.com/p/crack-language/">Crack</a> aims to provide
483the ease of development of a scripting language with the performance of a
484compiled language. The language derives concepts from C++, Java and Python,
485incorporating object-oriented programming, operator overloading and strong
486typing. Crack 0.2 works with LLVM 2.7, and the forthcoming Crack 0.2.1 release
487builds on LLVM 2.8.</p>
488
489</div>
490
491<!--=========================================================================-->
492<div class="doc_subsection">
493<a name="DresdenTM">Dresden TM Compiler (DTMC)</a>
494</div>
495
496<div class="doc_text">
497<p>
498<a href="http://tm.inf.tu-dresden.de">DTMC</a> provides support for
499Transactional Memory, which is an easy-to-use and efficient way to synchronize
500accesses to shared memory. Transactions can contain normal C/C++ code (e.g.,
Gabor Greif9f459132010-10-04 17:03:49 +0000501<code>__transaction { list.remove(x); x.refCount--; }</code>) and will be executed
Chris Lattner2fb6e5c2010-10-03 23:09:03 +0000502virtually atomically and isolated from other transactions.</p>
503
504</div>
505
506<!--=========================================================================-->
507<div class="doc_subsection">
Chris Lattnerc5fd1562010-10-04 16:06:37 +0000508<a name="Kai">Kai Programming Language</a>
Chris Lattner2fb6e5c2010-10-03 23:09:03 +0000509</div>
510
511<div class="doc_text">
512<p>
513<a href="http://www.oriontransfer.co.nz/research/kai">Kai</a> (Japanese 会 for
514meeting/gathering) is an experimental interpreter that provides a highly
515extensible runtime environment and explicit control over the compilation
516process. Programs are defined using nested symbolic expressions, which are all
517parsed into first-class values with minimal intrinsic semantics. Kai can
518generate optimised code at run-time (using LLVM) in order to exploit the nature
519of the underlying hardware and to integrate with external software libraries.
520It is a unique exploration into world of dynamic code compilation, and the
521interaction between high level and low level semantics.</p>
522
523</div>
524
Chris Lattner75547712010-10-03 23:49:06 +0000525<!--=========================================================================-->
526<div class="doc_subsection">
527<a name="OSL">OSL: Open Shading Language</a>
528</div>
529
530<div class="doc_text">
531<p>
532<a href="http://code.google.com/p/openshadinglanguage/">OSL</a> is a shading
533language designed for use in physically based renderers and in particular
534production rendering. By using LLVM instead of the interpreter, it was able to
535meet its performance goals (&gt;= C-code) while retaining the benefits of
536runtime specialization and a portable high-level language.
537</p>
538
539</div>
540
541
Chris Lattnere0518442010-10-01 06:34:49 +0000542
Chris Lattner8348b472008-06-08 21:58:17 +0000543<!-- *********************************************************************** -->
544<div class="doc_section">
Chris Lattner922d00f2010-07-21 15:57:40 +0000545 <a name="whatsnew">What's New in LLVM 2.8?</a>
Chris Lattner8348b472008-06-08 21:58:17 +0000546</div>
547<!-- *********************************************************************** -->
548
Chris Lattnerf8e0b4e2008-06-08 22:59:35 +0000549<div class="doc_text">
550
Chris Lattner61358ab2009-10-13 17:48:04 +0000551<p>This release includes a huge number of bug fixes, performance tweaks and
Chris Lattnerf6662f92008-10-13 17:57:36 +0000552minor improvements. Some of the major improvements and new features are listed
553in this section.
Chris Lattnerf8e0b4e2008-06-08 22:59:35 +0000554</p>
Chris Lattner61358ab2009-10-13 17:48:04 +0000555
Chris Lattner914ce462010-04-22 06:28:20 +0000556</div>
557
558<!--=========================================================================-->
559<div class="doc_subsection">
Chris Lattner252b83d2008-02-06 18:00:06 +0000560<a name="majorfeatures">Major New Features</a>
561</div>
562
563<div class="doc_text">
564
Chris Lattner922d00f2010-07-21 15:57:40 +0000565<p>LLVM 2.8 includes several major new capabilities:</p>
Chris Lattnerf3e5bc62007-05-14 06:56:09 +0000566
567<ul>
Chris Lattner11b66112010-10-04 02:42:39 +0000568<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>
573</ul>
574<li>A new 'llvm-diff' tool is available that does a semantic diff of .ll
575 files.</li>
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>
Chris Lattner8170c102008-02-10 08:18:42 +0000579</ul>
Chris Lattnerdc910082010-03-17 06:41:58 +0000580
Chris Lattnerf3e5bc62007-05-14 06:56:09 +0000581</div>
582
Chris Lattner84977642007-09-21 03:54:09 +0000583<!--=========================================================================-->
584<div class="doc_subsection">
Chris Lattner917cc712009-03-02 02:37:32 +0000585<a name="coreimprovements">LLVM IR and Core Improvements</a>
Chris Lattnerf304ffc2008-02-10 08:17:19 +0000586</div>
587
588<div class="doc_text">
Chris Lattner61358ab2009-10-13 17:48:04 +0000589<p>LLVM IR has several new features for better support of new targets and that
590expose new optimization opportunities:</p>
Chris Lattnerf304ffc2008-02-10 08:17:19 +0000591
Chris Lattner791f77b2008-06-05 06:25:56 +0000592<ul>
Chris Lattner11b66112010-10-04 02:42:39 +0000593<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.
596</li>
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>
Chris Lattnerf304ffc2008-02-10 08:17:19 +0000606</ul>
Mikhail Glushenkovea65d7d2008-10-13 02:08:34 +0000607
Chris Lattnerf304ffc2008-02-10 08:17:19 +0000608</div>
609
610<!--=========================================================================-->
611<div class="doc_subsection">
Chris Lattneracce85d2008-02-10 07:46:44 +0000612<a name="optimizer">Optimizer Improvements</a>
613</div>
614
615<div class="doc_text">
616
Chris Lattner61358ab2009-10-13 17:48:04 +0000617<p>In addition to a large array of minor performance tweaks and bug fixes, this
Chris Lattnerf3013872008-10-13 21:50:36 +0000618release includes a few major enhancements and additions to the optimizers:</p>
Chris Lattneracce85d2008-02-10 07:46:44 +0000619
620<ul>
Chris Lattner11b66112010-10-04 02:42:39 +0000621<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>
Chris Lattneracce85d2008-02-10 07:46:44 +0000626
Chris Lattner11b66112010-10-04 02:42:39 +0000627<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>
Duncan Sands30be9e42010-10-04 10:04:14 +0000634<li>The loop optimizer has significantly improved strength reduction and analysis
Chris Lattner11b66112010-10-04 02:42:39 +0000635 capabilities. Notably it is able to build on the trap value and signed
636 integer overflow information to optimize &lt;= and &gt;= 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
644 environment.</li>
645</ul>
646
647<!--
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
Chris Lattner7d9b6b42010-10-02 21:59:30 +0000650 Preliminary work on TBAA but not usable in 2.8.
651 New CorrelatedValuePropagation pass, not on by default in 2.8 yet.
Chris Lattner11b66112010-10-04 02:42:39 +0000652-->
Chris Lattner917cc712009-03-02 02:37:32 +0000653
Chris Lattner61358ab2009-10-13 17:48:04 +0000654</div>
Chris Lattner917cc712009-03-02 02:37:32 +0000655
Chris Lattner0b832202008-06-08 02:45:07 +0000656<!--=========================================================================-->
657<div class="doc_subsection">
Chris Lattner4ba2b652010-09-30 16:31:33 +0000658<a name="mc">MC Level Improvements</a>
659</div>
660
661<div class="doc_text">
662<p>
Chris Lattner7d9b6b42010-10-02 21:59:30 +0000663The LLVM Machine Code (aka MC) subsystem was created to solve a number
Chris Lattner4ba2b652010-09-30 16:31:33 +0000664of problems in the realm of assembly, disassembly, object file format handling,
665and a number of other related areas that CPU instruction-set level tools work
Chris Lattner7d9b6b42010-10-02 21:59:30 +0000666in.</p>
667
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>
673
674<ul>
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>
Chris Lattner3bdcda12010-10-04 03:58:12 +0000685<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>
Chris Lattner7d9b6b42010-10-02 21:59:30 +0000688</ul>
689
690<p>For more information, please see the <a
Chris Lattner4ba2b652010-09-30 16:31:33 +0000691href="http://blog.llvm.org/2010/04/intro-to-llvm-mc-project.html">Intro to the
692LLVM MC Project Blog Post</a>.
693</p>
694
Chris Lattner4ba2b652010-09-30 16:31:33 +0000695</div>
696
697
Chris Lattner4ba2b652010-09-30 16:31:33 +0000698<!--=========================================================================-->
699<div class="doc_subsection">
Chris Lattner511433e2009-03-02 03:24:11 +0000700<a name="codegen">Target Independent Code Generator Improvements</a>
Chris Lattner0b832202008-06-08 02:45:07 +0000701</div>
702
703<div class="doc_text">
704
Mikhail Glushenkovf795ef02009-03-01 18:09:47 +0000705<p>We have put a significant amount of work into the code generator
706infrastructure, which allows us to implement more aggressive algorithms and make
707it run faster:</p>
Chris Lattner0b832202008-06-08 02:45:07 +0000708
709<ul>
Chris Lattner3bdcda12010-10-04 03:58:12 +0000710<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, &lt;3 x float&gt; 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
728 and simplified.</li>
729<li>The code generator now has a "LocalStackSlotPass", which optimizes stack
730 slot access for targets (like ARM) that have limited stack displacement
731 addressing.</li>
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
Duncan Sands30be9e42010-10-04 10:04:14 +0000736 natively supported, easing the implementation burden on targets.</li>
Chris Lattnerb822f652010-10-04 16:46:07 +0000737<li>We have added two new bottom-up pre-allocation register pressure aware schedulers:
738<ol>
739<li>The hybrid scheduler schedules aggressively to minimize schedule length when registers are available and avoid overscheduling in high pressure situations.</li>
740<li>The instruction-level-parallelism scheduler schedules for maximum ILP when registers are available and avoid overscheduling in high pressure situations.</li>
741</ol></li>
Chris Lattner3bdcda12010-10-04 03:58:12 +0000742<li>The tblgen type inference algorithm was rewritten to be more consistent and
743 diagnose more target bugs. If you have an out-of-tree backend, you may
744 find that it finds bugs in your target description. This support also
745 allows limited support for writing patterns for instructions that return
746 multiple results (e.g. a virtual register and a flag result). The
747 'parallel' modifier in tblgen was removed, you should use the new support
748 for multiple results instead.</li>
749<li>A new (experimental) "-rendermf" pass is available which renders a
750 MachineFunction into HTML, showing live ranges and other useful
751 details.</li>
Chris Lattner7714c912010-10-04 04:39:25 +0000752<li>The new SubRegIndex tablegen class allows subregisters to be indexed
753 symbolically instead of numerically. If your target uses subregisters you
754 will need to adapt to use SubRegIndex when you upgrade to 2.8.</li>
Chris Lattner3bdcda12010-10-04 03:58:12 +0000755<!-- SplitKit -->
Chris Lattner7d9b6b42010-10-02 21:59:30 +0000756
Chris Lattner3bdcda12010-10-04 03:58:12 +0000757<li>The -fast-isel instruction selection path (used at -O0 on X86) was rewritten
758 to work bottom-up on basic blocks instead of top down. This makes it
759 slightly faster (because the MachineDCE pass is not needed any longer) and
760 allows it to generate better code in some cases.</li>
Chris Lattner7d9b6b42010-10-02 21:59:30 +0000761
Chris Lattner84977642007-09-21 03:54:09 +0000762</ul>
Chris Lattner84977642007-09-21 03:54:09 +0000763</div>
764
Chris Lattner917cc712009-03-02 02:37:32 +0000765<!--=========================================================================-->
766<div class="doc_subsection">
767<a name="x86">X86-32 and X86-64 Target Improvements</a>
768</div>
769
770<div class="doc_text">
Chris Lattner7714c912010-10-04 04:39:25 +0000771<p>New features and major changes in the X86 target include:
Chris Lattner917cc712009-03-02 02:37:32 +0000772</p>
773
774<ul>
Chris Lattner922d00f2010-07-21 15:57:40 +0000775<li>The X86 backend now supports holding X87 floating point stack values
776 in registers across basic blocks, dramatically improving performance of code
Gabor Greif9f459132010-10-04 17:03:49 +0000777 that uses long double, and when targeting CPUs that don't support SSE.</li>
Chris Lattner511433e2009-03-02 03:24:11 +0000778
Chris Lattner7714c912010-10-04 04:39:25 +0000779<li>The X86 backend now uses a SSEDomainFix pass to optimize SSE operations. On
780 Nehalem ("Core i7") and newer CPUs there is a 2 cycle latency penalty on
781 using a register in a different domain than where it was defined. This pass
782 optimizes away these stalls.</li>
Chris Lattner7d9b6b42010-10-02 21:59:30 +0000783
Duncan Sands30be9e42010-10-04 10:04:14 +0000784<li>The X86 backend now promotes 16-bit integer operations to 32-bits when
Chris Lattner7714c912010-10-04 04:39:25 +0000785 possible. This avoids 0x66 prefixes, which are slow on some
786 microarchitectures and bloat the code on all of them.</li>
Chris Lattner7d9b6b42010-10-02 21:59:30 +0000787
Chris Lattner7714c912010-10-04 04:39:25 +0000788<li>The X86 backend now supports the Microsoft "thiscall" calling convention,
789 and a <a href="LangRef.html#callingconv">calling convention</a> to support
790 <a href="#GHC">ghc</a>.</li>
Chris Lattner7d9b6b42010-10-02 21:59:30 +0000791
Chris Lattner7714c912010-10-04 04:39:25 +0000792<li>The X86 backend supports a new "llvm.x86.int" intrinsic, which maps onto
793 the X86 "int $42" and "int3" instructions.</li>
Chris Lattner7d9b6b42010-10-02 21:59:30 +0000794
Chris Lattner7714c912010-10-04 04:39:25 +0000795<li>At the IR level, the &lt;2 x float&gt; datatype is now promoted and passed
Duncan Sands30be9e42010-10-04 10:04:14 +0000796 around as a &lt;4 x float&gt; instead of being passed and returned as an MMX
Chris Lattner7714c912010-10-04 04:39:25 +0000797 vector. If you have a frontend that uses this, please pass and return a
798 &lt;2 x i32&gt; instead (using bitcasts).</li>
799
800<li>When printing .s files in verbose assembly mode (the default for clang -S),
801 the X86 backend now decodes X86 shuffle instructions and prints human
Gabor Greif9f459132010-10-04 17:03:49 +0000802 readable comments after the most inscrutable of them, e.g.:
Chris Lattner7714c912010-10-04 04:39:25 +0000803
804<pre>
805 insertps $113, %xmm3, %xmm0 <i># xmm0 = zero,xmm0[1,2],xmm3[1]</i>
806 unpcklps %xmm1, %xmm0 <i># xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]</i>
807 pshufd $1, %xmm1, %xmm1 <i># xmm1 = xmm1[1,0,0,0]</i>
808</pre>
809</li>
Chris Lattner7d9b6b42010-10-02 21:59:30 +0000810
Chris Lattner917cc712009-03-02 02:37:32 +0000811</ul>
812
Chris Lattner917cc712009-03-02 02:37:32 +0000813</div>
Chris Lattner84977642007-09-21 03:54:09 +0000814
815<!--=========================================================================-->
816<div class="doc_subsection">
Chris Lattner61358ab2009-10-13 17:48:04 +0000817<a name="ARM">ARM Target Improvements</a>
Chris Lattnerc441fb82009-03-01 02:30:21 +0000818</div>
819
820<div class="doc_text">
Chris Lattner61358ab2009-10-13 17:48:04 +0000821<p>New features of the ARM target include:
822</p>
Chris Lattnerc441fb82009-03-01 02:30:21 +0000823
824<ul>
Chris Lattner7714c912010-10-04 04:39:25 +0000825<li>The ARM backend now optimizes tail calls into jumps.</li>
826<li>Scheduling is improved through the new list-hybrid scheduler as well
827 as through better modeling of structural hazards.</li>
828<li><a href="LangRef.html#int_fp16">Half float</a> instructions are now
829 supported.</li>
830<li>NEON support has been improved to model instructions which operate onto
Duncan Sands30be9e42010-10-04 10:04:14 +0000831 multiple consecutive registers more aggressively. This avoids lots of
Chris Lattner7714c912010-10-04 04:39:25 +0000832 extraneous register copies.</li>
833<li>The ARM backend now uses a new "ARMGlobalMerge" pass, which merges several
834 global variables into one, saving extra address computation (all the global
835 variables can be accessed via same base address) and potentially reducing
836 register pressure.</li>
Chris Lattnerc441fb82009-03-01 02:30:21 +0000837
Chris Lattner7714c912010-10-04 04:39:25 +0000838<li>The ARM has received many minor improvements and tweaks which lead to
839substantially better performance in a wide range of different scenarios.</li>
Chris Lattner7d9b6b42010-10-02 21:59:30 +0000840
Chris Lattner4eac9242010-10-02 22:44:15 +0000841<li>The ARM NEON intrinsics have been substantially reworked to reduce
842 redundancy and improve code generation. Some of the major changes are:
843 <ol>
844 <li>
845 All of the NEON load and store intrinsics (llvm.arm.neon.vld* and
846 llvm.arm.neon.vst*) take an extra parameter to specify the alignment in bytes
847 of the memory being accessed.
848 </li>
849 <li>
850 The llvm.arm.neon.vaba intrinsic (vector absolute difference and
851 accumulate) has been removed. This operation is now represented using
852 the llvm.arm.neon.vabd intrinsic (vector absolute difference) followed by a
853 vector add.
854 </li>
855 <li>
856 The llvm.arm.neon.vabdl and llvm.arm.neon.vabal intrinsics (lengthening
Gabor Greif9f459132010-10-04 17:03:49 +0000857 vector absolute difference with and without accumulation) have been removed.
Chris Lattner4eac9242010-10-02 22:44:15 +0000858 They are represented using the llvm.arm.neon.vabd intrinsic (vector absolute
859 difference) followed by a vector zero-extend operation, and for vabal,
860 a vector add.
861 </li>
862 <li>
863 The llvm.arm.neon.vmovn intrinsic has been removed. Calls of this intrinsic
864 are now replaced by vector truncate operations.
865 </li>
866 <li>
867 The llvm.arm.neon.vmovls and llvm.arm.neon.vmovlu intrinsics have been
868 removed. They are now represented as vector sign-extend (vmovls) and
869 zero-extend (vmovlu) operations.
870 </li>
871 <li>
872 The llvm.arm.neon.vaddl*, llvm.arm.neon.vaddw*, llvm.arm.neon.vsubl*, and
873 llvm.arm.neon.vsubw* intrinsics (lengthening vector add and subtract) have
874 been removed. They are replaced by vector add and vector subtract operations
875 where one (vaddw, vsubw) or both (vaddl, vsubl) of the operands are either
876 sign-extended or zero-extended.
877 </li>
878 <li>
879 The llvm.arm.neon.vmulls, llvm.arm.neon.vmullu, llvm.arm.neon.vmlal*, and
880 llvm.arm.neon.vmlsl* intrinsics (lengthening vector multiply with and without
881 accumulation and subtraction) have been removed. These operations are now
882 represented as vector multiplications where the operands are either
883 sign-extended or zero-extended, followed by a vector add for vmlal or a
884 vector subtract for vmlsl. Note that the polynomial vector multiply
885 intrinsic, llvm.arm.neon.vmullp, remains unchanged.
886 </li>
887 </ol>
Bob Wilson5b2fb952010-09-13 17:37:55 +0000888</li>
Chris Lattner7714c912010-10-04 04:39:25 +0000889
Bob Wilsone8472772010-09-13 17:39:35 +0000890</ul>
Chris Lattner61358ab2009-10-13 17:48:04 +0000891</div>
Chris Lattnerc441fb82009-03-01 02:30:21 +0000892
Chris Lattner77d29b12008-06-05 08:02:49 +0000893
894<!--=========================================================================-->
895<div class="doc_subsection">
Chris Lattnerf6662f92008-10-13 17:57:36 +0000896<a name="changes">Major Changes and Removed Features</a>
897</div>
898
899<div class="doc_text">
900
Chris Lattnereeb4da02008-10-13 22:06:31 +0000901<p>If you're already an LLVM user or developer with out-of-tree changes based
Chris Lattner922d00f2010-07-21 15:57:40 +0000902on LLVM 2.7, this section lists some "gotchas" that you may run into upgrading
Chris Lattnereeb4da02008-10-13 22:06:31 +0000903from the previous release.</p>
Chris Lattnerf6662f92008-10-13 17:57:36 +0000904
905<ul>
Chris Lattnerbb117712010-10-04 01:29:06 +0000906<li>The build configuration machinery changed the output directory names. It
Duncan Sands30be9e42010-10-04 10:04:14 +0000907 wasn't clear to many people that a "Release-Asserts" build was a release build
Chris Lattnerbb117712010-10-04 01:29:06 +0000908 without asserts. To make this more clear, "Release" does not include
909 assertions and "Release+Asserts" does (likewise, "Debug" and
910 "Debug+Asserts").</li>
911<li>The MSIL Backend was removed, it was unsupported and broken.</li>
912<li>The ABCD, SSI, and SCCVN passes were removed. These were not fully
913 functional and their behavior has been or will be subsumed by the
914 LazyValueInfo pass.</li>
915<li>The LLVM IR 'Union' feature was removed. While this is a desirable feature
916 for LLVM IR to support, the existing implementation was half baked and
917 barely useful. We'd really like anyone interested to resurrect the work and
918 finish it for a future release.</li>
919<li>If you're used to reading .ll files, you'll probably notice that .ll file
920 dumps don't produce #uses comments anymore. To get them, run a .bc file
921 through "llvm-dis --show-annotations".</li>
Chris Lattner11b66112010-10-04 02:42:39 +0000922<li>Target triples are now stored in a normalized form, and all inputs from
923 humans are expected to be normalized by Triple::normalize before being
924 stored in a module triple or passed to another library.</li>
Chris Lattnerb7c85b42010-04-21 05:17:40 +0000925</ul>
Chris Lattnereeb4da02008-10-13 22:06:31 +0000926
Chris Lattnerbb117712010-10-04 01:29:06 +0000927
928
Chris Lattnereeb4da02008-10-13 22:06:31 +0000929<p>In addition, many APIs have changed in this release. Some of the major LLVM
930API changes are:</p>
Chris Lattnereeb4da02008-10-13 22:06:31 +0000931<ul>
Chris Lattner4eac9242010-10-02 22:44:15 +0000932<li>LLVM 2.8 changes the internal order of operands in <a
933 href="http://llvm.org/doxygen/classllvm_1_1InvokeInst.html"><tt>InvokeInst</tt></a>
934 and <a href="http://llvm.org/doxygen/classllvm_1_1CallInst.html"><tt>CallInst</tt></a>.
Chris Lattnerbb117712010-10-04 01:29:06 +0000935 To be portable across releases, please use the <tt>CallSite</tt> class and the
936 high-level accessors, such as <tt>getCalledValue</tt> and
937 <tt>setUnwindDest</tt>.
Chris Lattner4eac9242010-10-02 22:44:15 +0000938</li>
939<li>
Chris Lattnerbb117712010-10-04 01:29:06 +0000940 You can no longer pass use_iterators directly to cast&lt;&gt; (and similar),
941 because these routines tend to perform costly dereference operations more
942 than once. You have to dereference the iterators yourself and pass them in.
Chris Lattner4eac9242010-10-02 22:44:15 +0000943</li>
944<li>
Chris Lattnerbb117712010-10-04 01:29:06 +0000945 llvm.memcpy.*, llvm.memset.*, llvm.memmove.* intrinsics take an extra
946 parameter now ("i1 isVolatile"), totaling 5 parameters, and the pointer
947 operands are now address-space qualified.
Chris Lattner4eac9242010-10-02 22:44:15 +0000948 If you were creating these intrinsic calls and prototypes yourself (as opposed
Chris Lattnerbb117712010-10-04 01:29:06 +0000949 to using Intrinsic::getDeclaration), you can use
Gabor Greif9f459132010-10-04 17:03:49 +0000950 UpgradeIntrinsicFunction/UpgradeIntrinsicCall to be portable across releases.
Chris Lattner4eac9242010-10-02 22:44:15 +0000951</li>
952<li>
953 SetCurrentDebugLocation takes a DebugLoc now instead of a MDNode.
954 Change your code to use
955 SetCurrentDebugLocation(DebugLoc::getFromDILocation(...)).
956</li>
957<li>
Chris Lattner4eac9242010-10-02 22:44:15 +0000958 The <tt>RegisterPass</tt> and <tt>RegisterAnalysisGroup</tt> templates are
959 considered deprecated, but continue to function in LLVM 2.8. Clients are
960 strongly advised to use the upcoming <tt>INITIALIZE_PASS()</tt> and
961 <tt>INITIALIZE_AG_PASS()</tt> macros instead.
Chris Lattner4eac9242010-10-02 22:44:15 +0000962</li>
963<li>
964 The constructor for the Triple class no longer tries to understand odd triple
965 specifications. Frontends should ensure that they only pass valid triples to
966 LLVM. The Triple::normalize utility method has been added to help front-ends
967 deal with funky triples.
Chris Lattnerbb117712010-10-04 01:29:06 +0000968</li>
969
Chris Lattner4eac9242010-10-02 22:44:15 +0000970<li>
971 Some APIs got renamed:
972 <ul>
Chris Lattnerbb117712010-10-04 01:29:06 +0000973 <li>llvm_report_error -&gt; report_fatal_error</li>
974 <li>llvm_install_error_handler -&gt; install_fatal_error_handler</li>
975 <li>llvm::DwarfExceptionHandling -&gt; llvm::JITExceptionHandling</li>
976 <li>VISIBILITY_HIDDEN -&gt; LLVM_LIBRARY_VISIBILITY</li>
Chris Lattner4eac9242010-10-02 22:44:15 +0000977 </ul>
978</li>
979
Devang Patelb34dd132008-10-14 20:03:43 +0000980</ul>
Chris Lattnerf6662f92008-10-13 17:57:36 +0000981
Chris Lattnerf6662f92008-10-13 17:57:36 +0000982</div>
983
984
Chris Lattner79c3fe12003-10-02 04:57:28 +0000985<!-- *********************************************************************** -->
Misha Brukman500bc302003-11-22 00:38:41 +0000986<div class="doc_section">
987 <a name="knownproblems">Known Problems</a>
988</div>
Chris Lattner79c3fe12003-10-02 04:57:28 +0000989<!-- *********************************************************************** -->
990
Misha Brukman500bc302003-11-22 00:38:41 +0000991<div class="doc_text">
992
Mikhail Glushenkovf795ef02009-03-01 18:09:47 +0000993<p>This section contains significant known problems with the LLVM system,
Chris Lattnere18b32e2008-11-10 05:40:34 +0000994listed by component. If you run into a problem, please check the <a
Chris Lattnerc463b272005-10-29 07:07:09 +0000995href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
Chris Lattner5eccca42003-12-12 21:22:16 +0000996there isn't already one.</p>
Chris Lattner79c3fe12003-10-02 04:57:28 +0000997
Misha Brukman500bc302003-11-22 00:38:41 +0000998</div>
Chris Lattner79c3fe12003-10-02 04:57:28 +0000999
Chris Lattnerf5ee1702004-03-14 02:03:02 +00001000<!-- ======================================================================= -->
1001<div class="doc_subsection">
1002 <a name="experimental">Experimental features included with this release</a>
1003</div>
Chris Lattner79c3fe12003-10-02 04:57:28 +00001004
Chris Lattnerf5ee1702004-03-14 02:03:02 +00001005<div class="doc_text">
1006
Misha Brukman6df9e2c2004-05-12 21:46:05 +00001007<p>The following components of this LLVM release are either untested, known to
1008be broken or unreliable, or are in early development. These components should
1009not be relied on, and bugs should not be filed against them, but they may be
1010useful to some people. In particular, if you would like to work on one of these
Chris Lattner2b659ef2008-02-12 06:29:45 +00001011components, please contact us on the <a
1012href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev list</a>.</p>
Chris Lattnerf5ee1702004-03-14 02:03:02 +00001013
1014<ul>
Chris Lattnerbb117712010-10-04 01:29:06 +00001015<li>The Alpha, Blackfin, CellSPU, MicroBlaze, MSP430, MIPS, PIC16, SystemZ
1016 and XCore backends are experimental.</li>
Chris Lattner7d9b6b42010-10-02 21:59:30 +00001017<li><tt>llc</tt> "<tt>-filetype=obj</tt>" is experimental on all targets
1018 other than darwin-i386 and darwin-x86_64.</li>
Chris Lattnerf5ee1702004-03-14 02:03:02 +00001019</ul>
1020
1021</div>
Chris Lattner79c3fe12003-10-02 04:57:28 +00001022
Misha Brukman500bc302003-11-22 00:38:41 +00001023<!-- ======================================================================= -->
1024<div class="doc_subsection">
Chris Lattner26299222006-11-18 07:51:14 +00001025 <a name="x86-be">Known problems with the X86 back-end</a>
John Criswellc0c186d2005-11-08 21:11:33 +00001026</div>
1027
1028<div class="doc_text">
1029
1030<ul>
Anton Korobeynikova6094be2008-06-08 10:24:13 +00001031 <li>The X86 backend does not yet support
1032 all <a href="http://llvm.org/PR879">inline assembly that uses the X86
1033 floating point stack</a>. It supports the 'f' and 't' constraints, but not
1034 'u'.</li>
Duncan Sands47eff2b2008-06-08 19:38:43 +00001035 <li>Win64 code generation wasn't widely tested. Everything should work, but we
John Criswell524a5dd2009-03-02 15:28:15 +00001036 expect small issues to happen. Also, llvm-gcc cannot build the mingw64
Chris Lattner914ce462010-04-22 06:28:20 +00001037 runtime currently due to lack of support for the 'u' inline assembly
1038 constraint and for X87 floating point inline assembly.</li>
Dan Gohman8207ba92008-06-08 23:05:11 +00001039 <li>The X86-64 backend does not yet support the LLVM IR instruction
Chris Lattner914ce462010-04-22 06:28:20 +00001040 <tt>va_arg</tt>. Currently, front-ends support variadic
Dan Gohman8207ba92008-06-08 23:05:11 +00001041 argument constructs on X86-64 by lowering them manually.</li>
Chris Lattner26299222006-11-18 07:51:14 +00001042</ul>
1043
1044</div>
1045
1046<!-- ======================================================================= -->
1047<div class="doc_subsection">
1048 <a name="ppc-be">Known problems with the PowerPC back-end</a>
1049</div>
1050
1051<div class="doc_text">
1052
1053<ul>
Nicolas Geoffraye4285dc2007-05-15 09:21:28 +00001054<li>The Linux PPC32/ABI support needs testing for the interpreter and static
Chris Lattner57a460e2007-05-23 04:39:32 +00001055compilation, and lacks support for debug information.</li>
Chris Lattner26299222006-11-18 07:51:14 +00001056</ul>
1057
1058</div>
1059
1060<!-- ======================================================================= -->
1061<div class="doc_subsection">
Chris Lattnerf3e5bc62007-05-14 06:56:09 +00001062 <a name="arm-be">Known problems with the ARM back-end</a>
1063</div>
1064
1065<div class="doc_text">
1066
1067<ul>
Chris Lattner57a460e2007-05-23 04:39:32 +00001068<li>Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6
Duncan Sandsc90d68b2007-09-26 15:59:54 +00001069processors, thumb programs can crash or produce wrong
Chris Lattner57a460e2007-05-23 04:39:32 +00001070results (<a href="http://llvm.org/PR1388">PR1388</a>).</li>
John Criswell524a5dd2009-03-02 15:28:15 +00001071<li>Compilation for ARM Linux OABI (old ABI) is supported but not fully tested.
Chris Lattnerf3e5bc62007-05-14 06:56:09 +00001072</li>
Chris Lattnerf3e5bc62007-05-14 06:56:09 +00001073</ul>
1074
1075</div>
1076
1077<!-- ======================================================================= -->
1078<div class="doc_subsection">
Chris Lattner26299222006-11-18 07:51:14 +00001079 <a name="sparc-be">Known problems with the SPARC back-end</a>
1080</div>
1081
1082<div class="doc_text">
1083
1084<ul>
John Criswell524a5dd2009-03-02 15:28:15 +00001085<li>The SPARC backend only supports the 32-bit SPARC ABI (-m32); it does not
Chris Lattner26299222006-11-18 07:51:14 +00001086 support the 64-bit SPARC ABI (-m64).</li>
1087</ul>
1088
1089</div>
1090
1091<!-- ======================================================================= -->
1092<div class="doc_subsection">
Bruno Cardoso Lopesb7e1a4f2008-10-25 14:56:26 +00001093 <a name="mips-be">Known problems with the MIPS back-end</a>
1094</div>
1095
1096<div class="doc_text">
1097
1098<ul>
Bruno Cardoso Lopesb7e1a4f2008-10-25 14:56:26 +00001099<li>64-bit MIPS targets are not supported yet.</li>
1100</ul>
1101
1102</div>
1103
1104<!-- ======================================================================= -->
1105<div class="doc_subsection">
Chris Lattner26299222006-11-18 07:51:14 +00001106 <a name="alpha-be">Known problems with the Alpha back-end</a>
1107</div>
1108
1109<div class="doc_text">
1110
1111<ul>
1112
1113<li>On 21164s, some rare FP arithmetic sequences which may trap do not have the
1114appropriate nops inserted to ensure restartability.</li>
1115
John Criswellc0c186d2005-11-08 21:11:33 +00001116</ul>
1117</div>
1118
Chris Lattner26299222006-11-18 07:51:14 +00001119<!-- ======================================================================= -->
1120<div class="doc_subsection">
Chris Lattnerf3e5bc62007-05-14 06:56:09 +00001121 <a name="c-be">Known problems with the C back-end</a>
Chris Lattner26299222006-11-18 07:51:14 +00001122</div>
1123
1124<div class="doc_text">
1125
Chris Lattner3016ee92010-09-29 05:34:42 +00001126<p>The C backend has numerous problems and is not being actively maintained.
1127Depending on it for anything serious is not advised.</p>
1128
Chris Lattner26299222006-11-18 07:51:14 +00001129<ul>
Chris Lattner5733b272008-06-05 06:35:40 +00001130<li><a href="http://llvm.org/PR802">The C backend has only basic support for
1131 inline assembly code</a>.</li>
Chris Lattner725a0d82007-09-26 06:01:35 +00001132<li><a href="http://llvm.org/PR1658">The C backend violates the ABI of common
1133 C++ programs</a>, preventing intermixing between C++ compiled by the CBE and
Gabor Greif4906abe2009-03-02 12:02:51 +00001134 C++ code compiled with <tt>llc</tt> or native compilers.</li>
Duncan Sandsf74c0cc2008-02-10 13:40:55 +00001135<li>The C backend does not support all exception handling constructs.</li>
Duncan Sands50723a92009-02-25 11:51:54 +00001136<li>The C backend does not support arbitrary precision integers.</li>
Chris Lattner26299222006-11-18 07:51:14 +00001137</ul>
1138
1139</div>
John Criswellc0c186d2005-11-08 21:11:33 +00001140
Misha Brukman500bc302003-11-22 00:38:41 +00001141
1142<!-- ======================================================================= -->
1143<div class="doc_subsection">
Chris Lattner3016ee92010-09-29 05:34:42 +00001144 <a name="llvm-gcc">Known problems with the llvm-gcc front-end</a>
Misha Brukman500bc302003-11-22 00:38:41 +00001145</div>
Chris Lattner47588f92003-10-02 05:07:23 +00001146
Misha Brukman500bc302003-11-22 00:38:41 +00001147<div class="doc_text">
Chris Lattnerc5d658a2006-03-03 00:34:26 +00001148
Chris Lattner3016ee92010-09-29 05:34:42 +00001149<p>llvm-gcc is generally very stable for the C family of languages. The only
1150 major language feature of GCC not supported by llvm-gcc is the
1151 <tt>__builtin_apply</tt> family of builtins. However, some extensions
1152 are only supported on some targets. For example, trampolines are only
1153 supported on some targets (these are used when you take the address of a
1154 nested function).</p>
Misha Brukman500bc302003-11-22 00:38:41 +00001155
Chris Lattner3016ee92010-09-29 05:34:42 +00001156<p>Fortran support generally works, but there are still several unresolved bugs
1157 in <a href="http://llvm.org/bugs/">Bugzilla</a>. Please see the
1158 tools/gfortran component for details. Note that llvm-gcc is missing major
1159 Fortran performance work in the frontend and library that went into GCC after
1160 4.2. If you are interested in Fortran, we recommend that you consider using
1161 <a href="#dragonegg">dragonegg</a> instead.</p>
Chris Lattner79c3fe12003-10-02 04:57:28 +00001162
Duncan Sands3af96332010-10-04 10:06:56 +00001163<p>The llvm-gcc 4.2 Ada compiler has basic functionality, but is no longer being
1164actively maintained. If you are interested in Ada, we recommend that you
1165consider using <a href="#dragonegg">dragonegg</a> instead.</p>
Chris Lattner2b659ef2008-02-12 06:29:45 +00001166</div>
1167
Chris Lattner79c3fe12003-10-02 04:57:28 +00001168<!-- *********************************************************************** -->
Misha Brukman500bc302003-11-22 00:38:41 +00001169<div class="doc_section">
1170 <a name="additionalinfo">Additional Information</a>
1171</div>
Chris Lattner79c3fe12003-10-02 04:57:28 +00001172<!-- *********************************************************************** -->
1173
Misha Brukman500bc302003-11-22 00:38:41 +00001174<div class="doc_text">
1175
Chris Lattner416db102005-05-16 17:13:10 +00001176<p>A wide variety of additional information is available on the <a
Chris Lattnerb4b0ce72007-05-18 00:44:29 +00001177href="http://llvm.org">LLVM web page</a>, in particular in the <a
1178href="http://llvm.org/docs/">documentation</a> section. The web page also
Reid Spencer669ed452007-07-09 08:04:31 +00001179contains versions of the API documentation which is up-to-date with the
1180Subversion version of the source code.
Misha Brukman109d9e82005-03-30 19:14:24 +00001181You can access versions of these documents specific to this release by going
1182into the "<tt>llvm/doc/</tt>" directory in the LLVM tree.</p>
Chris Lattner79c3fe12003-10-02 04:57:28 +00001183
Misha Brukman500bc302003-11-22 00:38:41 +00001184<p>If you have any questions or comments about LLVM, please feel free to contact
Chris Lattnerc463b272005-10-29 07:07:09 +00001185us via the <a href="http://llvm.org/docs/#maillist"> mailing
Chris Lattner5eccca42003-12-12 21:22:16 +00001186lists</a>.</p>
Chris Lattner79c3fe12003-10-02 04:57:28 +00001187
Misha Brukman500bc302003-11-22 00:38:41 +00001188</div>
Chris Lattner79c3fe12003-10-02 04:57:28 +00001189
1190<!-- *********************************************************************** -->
Chris Lattner79c3fe12003-10-02 04:57:28 +00001191
Misha Brukman500bc302003-11-22 00:38:41 +00001192<hr>
Misha Brukman2061e892003-11-22 01:23:39 +00001193<address>
Misha Brukman38847d52003-12-21 22:53:21 +00001194 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
Misha Brukman44408702008-12-11 17:34:48 +00001195 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
Misha Brukman2061e892003-11-22 01:23:39 +00001196 <a href="http://validator.w3.org/check/referer"><img
Misha Brukman44408702008-12-11 17:34:48 +00001197 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
Misha Brukman2061e892003-11-22 01:23:39 +00001198
Chris Lattnerb4b0ce72007-05-18 00:44:29 +00001199 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
Misha Brukman500bc302003-11-22 00:38:41 +00001200 Last modified: $Date$
Misha Brukman2061e892003-11-22 01:23:39 +00001201</address>
Chris Lattner79c3fe12003-10-02 04:57:28 +00001202
Misha Brukman500bc302003-11-22 00:38:41 +00001203</body>
1204</html>