Gitiles
Code Review Sign In
gerrit-public.fairphone.software / fp2-dev / platform / external / llvm / 1c0030e243ac5e16a8c1d825f831cfd2e77fa26a / . / docs / ReleaseNotes.html
blob: c63f6f48f8fa8e8279dbe596f770e3758d69b311 [file] [log] [blame]
Anton Korobeynikov877b73b2008-06-08 10:23:46 +0000[diff] [blame]1
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
3 "http://www.w3.org/TR/html4/strict.dtd">
4<html>
5<head>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <link rel="stylesheet" href="llvm.css" type="text/css">
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]8 <title>LLVM 2.3 Release Notes</title>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]9</head>
10<body>
11
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]12<div class="doc_title">LLVM 2.3 Release Notes</div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]13
14<ol>
15 <li><a href="#intro">Introduction</a></li>
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]16 <li><a href="#changes">Major Changes and Sub-project Status</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]17 <li><a href="#whatsnew">What's New?</a></li>
18 <li><a href="GettingStarted.html">Installation Instructions</a></li>
19 <li><a href="#portability">Portability and Supported Platforms</a></li>
20 <li><a href="#knownproblems">Known Problems</a>
21 <li><a href="#additionalinfo">Additional Information</a></li>
22</ol>
23
24<div class="doc_author">
25 <p>Written by the <a href="http://llvm.org">LLVM Team</a><p>
26</div>
27
Chris Lattnere45f18b2008-06-05 08:02:49 +0000[diff] [blame]28<!-- Done through Week-of-Mon-20080324.txt -->
29
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]30<!-- *********************************************************************** -->
31<div class="doc_section">
32 <a name="intro">Introduction</a>
33</div>
34<!-- *********************************************************************** -->
35
36<div class="doc_text">
37
38<p>This document contains the release notes for the LLVM compiler
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]39infrastructure, release 2.3. Here we describe the status of LLVM, including
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]40major improvements from the previous release and any known problems. All LLVM
41releases may be downloaded from the <a href="http://llvm.org/releases/">LLVM
42releases web site</a>.</p>
43
44<p>For more information about LLVM, including information about the latest
45release, please check out the <a href="http://llvm.org/">main LLVM
46web site</a>. If you have questions or comments, the <a
47href="http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM developer's mailing
48list</a> is a good place to send them.</p>
49
50<p>Note that if you are reading this file from a Subversion checkout or the
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]51main LLVM web page, this document applies to the <i>next</i> release, not the
52current one. To see the release notes for a specific releases, please see the
53<a href="http://llvm.org/releases/">releases page</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]54
55</div>
56
57<!-- *********************************************************************** -->
58<div class="doc_section">
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]59 <a name="changes">Major Changes and Sub-project Status</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]60</div>
61<!-- *********************************************************************** -->
62
63<div class="doc_text">
64
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]65<p>This is the fourteenth public release of the LLVM Compiler Infrastructure.
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]66It includes a large number of features and refinements from LLVM 2.2.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]67
68</div>
69
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]70<!-- Unfinished features in 2.3:
Chris Lattner8fda4952008-02-10 07:46:44 +0000[diff] [blame]71 Machine LICM
72 Machine Sinking
73 LegalizeDAGTypes
74 -->
75
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]76<!--=========================================================================-->
77<div class="doc_subsection">
Chris Lattnere45f18b2008-06-05 08:02:49 +0000[diff] [blame]78<a name="changes">Major Changes in LLVM 2.3</a>
Chris Lattner41bf8e92008-02-10 07:04:35 +0000[diff] [blame]79</div>
80
81<div class="doc_text">
82
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]83<p>LLVM 2.3 no longer supports llvm-gcc 4.0, it has been replaced with
84 llvm-gcc 4.2.</p>
Chris Lattnere0798082008-02-06 18:00:06 +0000[diff] [blame]85
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]86<p>LLVM 2.3 no longer includes the <tt>llvm-upgrade</tt> tool. It was useful
87 for upgrading LLVM 1.9 files to LLVM 2.x syntax, but you can always use a
88 previous LLVM release to do this. One nice impact of this is that the LLVM
89 regression test suite no longer depends on llvm-upgrade, which makes it run
90 faster.</p>
91
92<p>The <tt>llvm2cpp</tt> tool has been folded into llc, use
Chris Lattnerd95b5962008-06-08 21:19:07 +0000[diff] [blame]93 <tt>llc -march=cpp</tt> instead of <tt>llvm2cpp</tt>.</p>
94
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]95<p>LLVM API Changes:</p>
Chris Lattnere45f18b2008-06-05 08:02:49 +0000[diff] [blame]96
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]97<ul>
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]98<li>Several core LLVM IR classes have migrated to use the
99 '<tt>FOOCLASS::Create(...)</tt>' pattern instead of '<tt>new
100 FOOCLASS(...)</tt>' (e.g. where FOOCLASS=<tt>BasicBlock</tt>). We hope to
101 standardize on <tt>FOOCLASS::Create</tt> for all IR classes in the future,
102 but not all of them have been moved over yet.</li>
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]103<li>LLVM 2.3 renames the LLVMBuilder and LLVMFoldingBuilder classes to
104 IRBuilder.</li>
105<li>MRegisterInfo was renamed to TargetRegisterInfo.</li>
106<li>The MappedFile class is gone, please use MemoryBuffer instead.</li>
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]107<li>The '<tt>-enable-eh</tt>' flag to llc has been removed. Now code should
108 encode whether it is safe to omit unwind information for a function by
109 tagging the Function object with the '<tt>nounwind</tt>' attribute.</li>
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]110
111</ul>
Chris Lattnere0798082008-02-06 18:00:06 +0000[diff] [blame]112</div>
113
114<!--=========================================================================-->
115<div class="doc_subsection">
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]116<a name="otherprojects">Other LLVM Sub-Projects</a>
117</div>
118
119<div class="doc_text">
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]120<p>
121The core LLVM 2.3 distribution currently consists of code from the core LLVM
122repository (which roughly contains the LLVM optimizer, code generators and
123supporting tools) and the llvm-gcc repository. In addition to this code, the
124LLVM Project includes other sub-projects that are in development. The two which
125are the most actively developed are the new <a href="#vmkit">vmkit Project</a>
126and the <a href="#clang">Clang Project</a>.
127</p>
128</div>
129
130<!--=========================================================================-->
131<div class="doc_subsubsection">
132<a name="vmkit">vmkit</a>
133</div>
134
135<div class="doc_text">
136<p>
137The "vmkit" project is a new addition to the LLVM family. It is an
138implementation of a JVM and a CLI Virtual Machines (Microsoft .NET is an
139implementation of the CLI) using the Just-In-Time compiler of LLVM.</p>
140
141<p>The JVM, called JnJVM, executes real-world applications such as Apache
142projects (e.g. Felix and Tomcat) and the SpecJVM98 benchmark. It uses the GNU
143Classpath project for the base classes. The CLI implementation, called N3, is
144its in early stages but can execute simple applications and the "pnetmark"
145benchmark. It uses the pnetlib project as its core library.</p>
146
147<p>The 'vmkit' VMs compare in performance with industrial and top open-source
148VMs on scientific applications. Besides the JIT, the VMs use many features of
149the LLVM framework, including the standard set of optimizations, atomic
150operations, custom function provider and memory manager for JITed methods, and
151specific virtual machine optimizations. vmkit is not an official part of LLVM
1522.3 release. It is publicly available under the LLVM license and can be
153downloaded from:
154</p>
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]155
156<p>
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]157<tt>svn co http://llvm.org/svn/llvm-project/vmkit/trunk vmkit</tt>
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]158</p>
159
160</div>
161
162<!--=========================================================================-->
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]163<div class="doc_subsubsection">
164<a name="clang">Clang</a>
165</div>
166
167<div class="doc_text">
168
169<p>The <a href="http://clang.llvm.org/">Clang project</a> is an effort to build
170a set of new 'LLVM native' front-end technologies for the LLVM optimizer
171and code generator. Clang is continuing to make major strides forward in all
172areas. Its C and Objective-C parsing support is very solid, and the code
173generation support is far enough along to build many C applications. While not
174yet production quality, it is progressing very nicely. In addition, C++
175front-end work has started to make significant progress.</p>
176
177<p>At this point, Clang is most useful if you are interested in source-to-source
178transformations (such as refactoring) and other source-level tools for C and
179Objective-C. Clang now also includes tools for turning C code into pretty HTML,
180and includes a new <a href="http://clang.llvm.org/StaticAnalysis.html">static
181analysis tool</a> in development. This tool is automatically focused on finding
182bugs in C and Objective-C code.</p>
183
184</div>
185
186
187<!-- *********************************************************************** -->
188<div class="doc_section">
189 <a name="whatsnew">What's New?</a>
190</div>
191<!-- *********************************************************************** -->
192
193<!--=========================================================================-->
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]194<div class="doc_subsection">
Chris Lattnere0798082008-02-06 18:00:06 +0000[diff] [blame]195<a name="majorfeatures">Major New Features</a>
196</div>
197
198<div class="doc_text">
199
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]200<p>LLVM 2.3 includes several major new capabilities:</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]201
202<ul>
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]203<li><p>The biggest change in LLVM 2.3 is Multiple Return Value (MRV) support.
204 MRVs allow LLVM IR to directly represent functions that return multiple
205 values without having to pass them "by reference" in the LLVM IR. This
206 allows a front-end to generate more efficient code, as MRVs are generally
207 returned in registers if a target supports them. See the <a
208 href="LangRef.html#i_getresult">LLVM IR Reference</a> for more details.</p>
209
210 <p>MRVs are fully supported in the LLVM IR, but are not yet fully supported in
211 on all targets. However, it is generally safe to return up to 2 values from
212 a function: most targets should be able to handle at least that. MRV
213 support is a critical requirement for X86-64 ABI support, as X86-64 requires
214 the ability to return multiple registers from functions, and we use MRVs to
215 accomplish this in a direct way.</p></li>
Chris Lattnerba948522008-06-05 06:57:39 +0000[diff] [blame]216
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]217<li><p>LLVM 2.3 includes a complete reimplementation of the "<tt>llvmc</tt>"
218 tool. It is designed to overcome several problems with the original
219 <tt>llvmc</tt> and to provide a superset of the features of the
220 '<tt>gcc</tt>' driver.</p>
Chris Lattnerba948522008-06-05 06:57:39 +0000[diff] [blame]221
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]222<p>The main features of <tt>llvmc2</tt> are:
223 <ul>
224 <li>Extended handling of command line options and smart rules for
225 dispatching them to different tools.</li>
226 <li>Flexible (and extensible) rules for defining different tools.</li>
227 <li>The different intermediate steps performed by tools are represented
228 as edges in the abstract graph.</li>
229 <li>The 'language' for driver behavior definition is tablegen and thus
230 it's relatively easy to add new features.</li>
231 <li>The definition of driver is transformed into set of C++ classes, thus
232 no runtime interpretation is needed.</li>
233 </ul></p>
234 </li>
Chris Lattnerba948522008-06-05 06:57:39 +0000[diff] [blame]235
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]236<li><p>LLVM 2.3 includes a completely rewritten interface for <a
237 href="LinkTimeOptimization.html">Link Time Optimization</a>. This interface
238 is written in C, which allows for easier integration with C code bases, and
239 incorporates improvements we learned about from the first incarnation of the
240 interface.</p></li>
Chris Lattnerba948522008-06-05 06:57:39 +0000[diff] [blame]241
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]242<li><p>The <a href="tutorial/LangImpl1.html">Kaleidoscope tutorial</a> now
243 includes a "port" of the tutorial that <a
244 href="tutorial/OCamlLangImpl1.html">uses the Ocaml bindings</a> to implement
245 the Kaleidoscope language.</p></li>
Chris Lattnerba948522008-06-05 06:57:39 +0000[diff] [blame]246
Chris Lattner5ba16862008-02-10 08:18:42 +0000[diff] [blame]247</ul>
248
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]249</div>
250
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]251
252<!--=========================================================================-->
253<div class="doc_subsection">
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]254<a name="llvm-gcc">llvm-gcc 4.2 Improvements</a>
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]255</div>
256
257<div class="doc_text">
258
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]259<p>LLVM 2.3 fully supports the llvm-gcc 4.2 front-end.</p>
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]260
261<p>llvm-gcc 4.2 includes numerous fixes to better support the Objective-C
262front-end. Objective-C now works very well on Mac OS/X.</p>
263
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]264<p>Fortran EQUIVALENCEs are now supported by the gfortran front-end.</p>
265
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]266<p>llvm-gcc 4.2 includes many other fixes which improve conformance with the
267relevant parts of the GCC testsuite.</p>
268
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]269</div>
270
271
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]272<!--=========================================================================-->
273<div class="doc_subsection">
Chris Lattnerc3649452008-02-10 08:17:19 +0000[diff] [blame]274<a name="coreimprovements">LLVM Core Improvements</a>
275</div>
276
277<div class="doc_text">
278<p>New features include:
279</p>
280
Chris Lattnerc3649452008-02-10 08:17:19 +0000[diff] [blame]281
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]282Common linkage?
283
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]284Atomic operation support, Alpha, X86, X86-64, PowerPC. "__sync_synchronize",
Anton Korobeynikov877b73b2008-06-08 10:23:46 +0000[diff] [blame]285"__sync_val_compare_and_swap", etc
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]286
287<ul>
Gordon Henriksenf9b229b2008-06-05 12:51:50 +0000[diff] [blame]288<li>The C and Ocaml bindings have received additional improvements. The
289bindings now cover pass managers, several transformation passes, iteration
290over the LLVM IR, target data, and parameter attribute lists.</li>
Chris Lattnerc3649452008-02-10 08:17:19 +0000[diff] [blame]291</ul>
292
293</div>
294
295<!--=========================================================================-->
296<div class="doc_subsection">
Chris Lattner8fda4952008-02-10 07:46:44 +0000[diff] [blame]297<a name="optimizer">Optimizer Improvements</a>
298</div>
299
300<div class="doc_text">
301
Chris Lattnerc3649452008-02-10 08:17:19 +0000[diff] [blame]302<p>In addition to a huge array of bug fixes and minor performance tweaks, the
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]303LLVM 2.3 optimizers support a few major enhancements:</p>
Chris Lattner8fda4952008-02-10 07:46:44 +0000[diff] [blame]304
305<ul>
306
Devang Patel67e46712008-06-05 21:44:00 +0000[diff] [blame]307<li>Loop index set splitting on by default.<p>
308This transformation hoists conditions from loop bodies and reduces loop's
309iteration space to improve performance. For example, <p>
310<pre>
Chris Lattnerabc3b352008-06-08 20:25:30 +0000[diff] [blame]311for (i = LB; i &lt; UB; ++i)
312 if (i &lt;= NV)
Devang Patel67e46712008-06-05 21:44:00 +0000[diff] [blame]313 LOOP_BODY
314</pre>
315is transformed into
316<pre>
317NUB = min(NV+1, UB)
Chris Lattnerabc3b352008-06-08 20:25:30 +0000[diff] [blame]318for (i = LB; i &lt; NUB; ++i)
Devang Patel67e46712008-06-05 21:44:00 +0000[diff] [blame]319 LOOP_BODY
320</pre>
321</li>
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]322
Gordon Henriksena2468472008-06-08 22:05:11 +0000[diff] [blame]323<li>LLVM includes a new <tt>memcpy</tt> optimization pass which removes
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]324dead <tt>memcpy</tt> calls, unneeded copies of aggregates, and performs
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]325return slot optimization. The LLVM optimizer now notices long sequences of
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]326consecutive stores and merges them into <tt>memcpy</tt>'s where profitable.</li>
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]327
Gabor Greif5e75e652008-06-05 18:39:01 +0000[diff] [blame]328<li>Alignment detection for vector memory references and for <tt>memcpy</tt> and
329<tt>memset</tt> is now more aggressive.</li>
Owen Anderson7f8ea062008-06-06 16:23:15 +0000[diff] [blame]330
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]331<li>The aggressive dead code elimination (ADCE) optimization has been rewritten
332to make it both faster and safer in the presence of code containing infinite
333loops. Some of its prior functionality has been factored out into the loop
334deletion pass, which <em>is</em> safe for infinite loops.</li>
335
336<li>Several optimizations have been sped up, leading to faster code generation
337 with the same code quality.</li>
338
339<li>The 'SimplifyLibCalls' pass, which optimizes calls to libc and libm
340 functions for C-based languages, has been rewritten to be a FunctionPass
341 instead a ModulePass. This allows it to be run more often and to be
342 included at -O1 in llvm-gcc. It was also extended to include more
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]343 optimizations and several corner case bugs were fixed.</li>
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]344
Duncan Sands8b4f3772008-06-08 19:38:43 +0000[diff] [blame]345<li>LLVM now includes a simple 'Jump Threading' pass, which attempts to simplify
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]346 conditional branches using information about predecessor blocks, simplifying
Chris Lattnerd95b5962008-06-08 21:19:07 +0000[diff] [blame]347 the control flow graph. This pass is pretty basic at this point, but
348 catches some important cases and provides a foundation to build on.</li>
349
350<li>Several corner case bugs which could lead to deleting volatile memory
351 accesses have been fixed.</li>
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]352</ul>
353
354</div>
355
356<!--=========================================================================-->
357<div class="doc_subsection">
358<a name="codegen">Code Generator Improvements</a>
359</div>
360
361<div class="doc_text">
362
363<p>We put a significant amount of work into the code generator infrastructure,
364which allows us to implement more aggressive algorithms and make it run
365faster:</p>
366
367<ul>
Dan Gohman1c0030e2008-06-08 22:49:27 +0000[diff] [blame^]368<li>The code generator now has support for carrying information about memory
369 references throughout the entire code generation process, via the
370 <a href="http://llvm.org/doxygen/classllvm_1_1MachineMemOperand.html">
371 MachineMemOperand</a> class. In the future this will be used to improve
372 both pre-pass and post-pass scheduling, and to improve compiler-debugging
373 output.</li>
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]374
375<li>The target-independent code generator infrastructure now uses LLVM's APInt
376 class to handle integer values, which allows it to support integer types
377 larger than 64 bits. Note that support for such types is also dependent on
378 target-specific support. Use of APInt is also a step toward support for
379 non-power-of-2 integer sizes.</li>
380
Chris Lattnerd95b5962008-06-08 21:19:07 +0000[diff] [blame]381<li>LLVM 2.3 includes several compile time speedups for code with large basic
Dan Gohmanf3d8f8d2008-06-08 22:44:43 +0000[diff] [blame]382 blocks, particularly in the instruction selection phase, register
383 allocation, scheduling, and tail merging/jump threading.</li>
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]384
385<li>Several improvements which make llc's <tt>--view-sunit-dags</tt>
386 visualization of scheduling dependency graphs easier to understand.</li>
387
388<li>The code generator allows targets to write patterns that generate subreg
389 references directly in .td files now.</li>
390
391<li><tt>memcpy</tt> lowering in the backend is more aggressive, particularly for
392 <tt>memcpy</tt> calls introduced by the code generator when handling
393 pass-by-value structure argument copies.</li>
Chris Lattnerd95b5962008-06-08 21:19:07 +0000[diff] [blame]394
395<li>Inline assembly with multiple register results now returns those results
396 directly in the appropriate registers, rather than going through memory.
397 Inline assembly that uses constraints like "ir" with immediates now use the
398 'i' form when possible instead of always loading the value in a register.
399 This saves an instruction and reduces register use.</li>
400
401<li>Added support for PIC/GOT style tail calls on x86/32 and initial support
402 for tail calls on PowerPC 32 (it may also work on ppc64 but not
403 thoroughly tested).</li>
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]404</ul>
405
406</div>
407
408
409<!--=========================================================================-->
410<div class="doc_subsection">
Chris Lattnere45f18b2008-06-05 08:02:49 +0000[diff] [blame]411<a name="x86specific">X86/X86-64 Specific Improvements</a>
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]412</div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]413
414<div class="doc_text">
Chris Lattnerc3649452008-02-10 08:17:19 +0000[diff] [blame]415<p>New target-specific features include:
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]416</p>
417
418<ul>
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]419<li>llvm-gcc's X86-64 ABI conformance is far improved, particularly in the
Gabor Greif5e75e652008-06-05 18:39:01 +0000[diff] [blame]420 area of passing and returning structures by value. llvm-gcc compiled code
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]421 now interoperates very well on X86-64 systems with other compilers.</li>
422
Duncan Sands8b4f3772008-06-08 19:38:43 +0000[diff] [blame]423<li>Support for Win64 was added. This includes code generation itself, JIT
Anton Korobeynikov77d15e22008-06-08 10:24:13 +0000[diff] [blame]424 support and necessary changes to llvm-gcc.</li>
425
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]426<li>The LLVM X86 backend now supports the support SSE 4.1 instruction set, and
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]427 the llvm-gcc 4.2 front-end supports the SSE 4.1 compiler builtins. Various
428 generic vector operations (insert/extract/shuffle) are much more efficient
429 when SSE 4.1 is enabled. The JIT automatically takes advantage of these
430 instructions, but llvm-gcc must be explicitly told to use them, e.g. with
431 <tt>-march=penryn</tt>.</li>
Anton Korobeynikov77d15e22008-06-08 10:24:13 +0000[diff] [blame]432
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]433<li>The X86 backend now does a number of optimizations that aim to avoid
434 converting numbers back and forth from SSE registers to the X87 floating
435 point stack.</li>
Anton Korobeynikov77d15e22008-06-08 10:24:13 +0000[diff] [blame]436
Chris Lattnerba948522008-06-05 06:57:39 +0000[diff] [blame]437<li>The X86 backend supports stack realignment, which is particularly useful for
438 vector code on OS's without 16-byte aligned stacks.</li>
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]439
440<li>The X86 backend now supports the "sseregparm" options in GCC, which allow
441 functions to be tagged as passing floating point values in SSE
442 registers.</li>
443
444<li>Trampolines (taking the address of a nested function) now work on
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]445 Linux/X86-64.</li>
Anton Korobeynikov77d15e22008-06-08 10:24:13 +0000[diff] [blame]446
Chris Lattnere45f18b2008-06-05 08:02:49 +0000[diff] [blame]447<li><tt>__builtin_prefetch</tt> is now compiled into the appropriate prefetch
448 instructions instead of being ignored.</li>
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]449
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]450<li>128-bit integers are now supported on X86-64 targets.</li>
Dan Gohman74e1ff52008-06-05 16:15:39 +0000[diff] [blame]451
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]452<li>The register allocator can now rematerialize PIC-base computations.</li>
453
Chris Lattnerd95b5962008-06-08 21:19:07 +0000[diff] [blame]454<li>The "t" and "f" inline assembly constraints for the X87 floating point stack
455 now work. However, the "u" constraint is still not fully supported.</li>
456
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]457</ul>
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]458
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]459</div>
460
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]461<!--=========================================================================-->
462<div class="doc_subsection">
Chris Lattnere45f18b2008-06-05 08:02:49 +0000[diff] [blame]463<a name="targetspecific">Other Target Specific Improvements</a>
464</div>
465
466<div class="doc_text">
467<p>New target-specific features include:
468</p>
469
470<ul>
Gabor Greif5e75e652008-06-05 18:39:01 +0000[diff] [blame]471<li>The LLVM C backend now supports vector code.</li>
Chris Lattnere45f18b2008-06-05 08:02:49 +0000[diff] [blame]472</ul>
473
474</div>
475
476
477
478<!--=========================================================================-->
479<div class="doc_subsection">
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]480<a name="otherimprovements">Other Improvements</a>
481</div>
482
483<div class="doc_text">
484<p>New features include:
485</p>
486
487<ul>
Chris Lattnerba948522008-06-05 06:57:39 +0000[diff] [blame]488<li>LLVM now builds with GCC 4.3.</li>
Chris Lattnerd95b5962008-06-08 21:19:07 +0000[diff] [blame]489<li>Bugpoint now supports running custom scripts (with the <tt>-run-custom</tt>
490 option) to determine how to execute the command and whether it is making
491 forward process.</li>
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]492</ul>
493
494</div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]495
496<!-- *********************************************************************** -->
497<div class="doc_section">
498 <a name="portability">Portability and Supported Platforms</a>
499</div>
500<!-- *********************************************************************** -->
501
502<div class="doc_text">
503
504<p>LLVM is known to work on the following platforms:</p>
505
506<ul>
Gabor Greif5e75e652008-06-05 18:39:01 +0000[diff] [blame]507<li>Intel and AMD machines (IA32) running Red Hat Linux, Fedora Core and FreeBSD
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]508 (and probably other unix-like systems).</li>
Chris Lattnerbd633de92008-02-06 06:30:34 +0000[diff] [blame]509<li>PowerPC and X86-based Mac OS X systems, running 10.3 and above in 32-bit and
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]510 64-bit modes.</li>
Chris Lattnere0798082008-02-06 18:00:06 +0000[diff] [blame]511<li>Intel and AMD machines running on Win32 using MinGW libraries (native).</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]512<li>Intel and AMD machines running on Win32 with the Cygwin libraries (limited
513 support is available for native builds with Visual C++).</li>
Gabor Greif5e75e652008-06-05 18:39:01 +0000[diff] [blame]514<li>Sun UltraSPARC workstations running Solaris 10.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]515<li>Alpha-based machines running Debian GNU/Linux.</li>
Gabor Greif5e75e652008-06-05 18:39:01 +0000[diff] [blame]516<li>Itanium-based (IA64) machines running Linux and HP-UX.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]517</ul>
518
Chris Lattnerba948522008-06-05 06:57:39 +0000[diff] [blame]519<p>The core LLVM infrastructure uses GNU autoconf to adapt itself
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]520to the machine and operating system on which it is built. However, minor
521porting may be required to get LLVM to work on new platforms. We welcome your
522portability patches and reports of successful builds or error messages.</p>
523
524</div>
525
526<!-- *********************************************************************** -->
527<div class="doc_section">
528 <a name="knownproblems">Known Problems</a>
529</div>
530<!-- *********************************************************************** -->
531
532<div class="doc_text">
533
534<p>This section contains all known problems with the LLVM system, listed by
535component. As new problems are discovered, they will be added to these
536sections. If you run into a problem, please check the <a
537href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
538there isn't already one.</p>
539
540</div>
541
542<!-- ======================================================================= -->
543<div class="doc_subsection">
544 <a name="experimental">Experimental features included with this release</a>
545</div>
546
547<div class="doc_text">
548
549<p>The following components of this LLVM release are either untested, known to
550be broken or unreliable, or are in early development. These components should
551not be relied on, and bugs should not be filed against them, but they may be
552useful to some people. In particular, if you would like to work on one of these
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]553components, please contact us on the <a
554href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev list</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]555
556<ul>
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]557<li>The MSIL, IA64, Alpha, SPU, and MIPS backends are experimental.</li>
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]558<li>The llc "<tt>-filetype=asm</tt>" (the default) is the only supported
Chris Lattnerbd633de92008-02-06 06:30:34 +0000[diff] [blame]559 value for this option.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]560</ul>
561
562</div>
563
564<!-- ======================================================================= -->
565<div class="doc_subsection">
566 <a name="x86-be">Known problems with the X86 back-end</a>
567</div>
568
569<div class="doc_text">
570
571<ul>
Anton Korobeynikov77d15e22008-06-08 10:24:13 +0000[diff] [blame]572 <li>The X86 backend does not yet support
573 all <a href="http://llvm.org/PR879">inline assembly that uses the X86
574 floating point stack</a>. It supports the 'f' and 't' constraints, but not
575 'u'.</li>
Chris Lattnerd95b5962008-06-08 21:19:07 +0000[diff] [blame]576 <li>The X86 backend generates inefficient floating point code when configured
577 to generate code for systems that don't have SSE2.</li>
Duncan Sands8b4f3772008-06-08 19:38:43 +0000[diff] [blame]578 <li>Win64 code generation wasn't widely tested. Everything should work, but we
Anton Korobeynikov77d15e22008-06-08 10:24:13 +0000[diff] [blame]579 expect small issues to happen. Also, llvm-gcc cannot build mingw64 runtime
580 currently due
581 to <a href="http://llvm.org/PR2255">several</a>
582 <a href="http://llvm.org/PR2257">bugs</a> in FP stackifier
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]583</ul>
584
585</div>
586
587<!-- ======================================================================= -->
588<div class="doc_subsection">
589 <a name="ppc-be">Known problems with the PowerPC back-end</a>
590</div>
591
592<div class="doc_text">
593
594<ul>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]595<li>The Linux PPC32/ABI support needs testing for the interpreter and static
596compilation, and lacks support for debug information.</li>
597</ul>
598
599</div>
600
601<!-- ======================================================================= -->
602<div class="doc_subsection">
603 <a name="arm-be">Known problems with the ARM back-end</a>
604</div>
605
606<div class="doc_text">
607
608<ul>
609<li>Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6
Duncan Sands1827fa62007-09-26 15:59:54 +0000[diff] [blame]610processors, thumb programs can crash or produce wrong
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]611results (<a href="http://llvm.org/PR1388">PR1388</a>).</li>
612<li>Compilation for ARM Linux OABI (old ABI) is supported, but not fully tested.
613</li>
Chris Lattnerd95b5962008-06-08 21:19:07 +0000[diff] [blame]614<li>There is a bug in QEMU-ARM (&lt;= 0.9.0) which causes it to incorrectly
615 execute
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]616programs compiled with LLVM. Please use more recent versions of QEMU.</li>
617</ul>
618
619</div>
620
621<!-- ======================================================================= -->
622<div class="doc_subsection">
623 <a name="sparc-be">Known problems with the SPARC back-end</a>
624</div>
625
626<div class="doc_text">
627
628<ul>
629<li>The SPARC backend only supports the 32-bit SPARC ABI (-m32), it does not
630 support the 64-bit SPARC ABI (-m64).</li>
631</ul>
632
633</div>
634
635<!-- ======================================================================= -->
636<div class="doc_subsection">
637 <a name="alpha-be">Known problems with the Alpha back-end</a>
638</div>
639
640<div class="doc_text">
641
642<ul>
643
644<li>On 21164s, some rare FP arithmetic sequences which may trap do not have the
645appropriate nops inserted to ensure restartability.</li>
646
647</ul>
648</div>
649
650<!-- ======================================================================= -->
651<div class="doc_subsection">
652 <a name="ia64-be">Known problems with the IA64 back-end</a>
653</div>
654
655<div class="doc_text">
656
657<ul>
658
659<li>C++ programs are likely to fail on IA64, as calls to <tt>setjmp</tt> are
660made where the argument is not 16-byte aligned, as required on IA64. (Strictly
661speaking this is not a bug in the IA64 back-end; it will also be encountered
662when building C++ programs using the C back-end.)</li>
663
664<li>The C++ front-end does not use <a href="http://llvm.org/PR406">IA64
665ABI compliant layout of v-tables</a>. In particular, it just stores function
666pointers instead of function descriptors in the vtable. This bug prevents
667mixing C++ code compiled with LLVM with C++ objects compiled by other C++
668compilers.</li>
669
670<li>There are a few ABI violations which will lead to problems when mixing LLVM
671output with code built with other compilers, particularly for floating-point
672programs.</li>
673
Duncan Sands8b4f3772008-06-08 19:38:43 +0000[diff] [blame]674<li>Defining vararg functions is not supported (but calling them is OK).</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]675
676<li>The Itanium backend has bitrotted somewhat.</li>
677</ul>
678
679</div>
680
681<!-- ======================================================================= -->
682<div class="doc_subsection">
683 <a name="c-be">Known problems with the C back-end</a>
684</div>
685
686<div class="doc_text">
687
688<ul>
Chris Lattner499c07c2008-06-05 06:35:40 +0000[diff] [blame]689<li><a href="http://llvm.org/PR802">The C backend has only basic support for
690 inline assembly code</a>.</li>
Chris Lattnere0194462007-09-26 06:01:35 +0000[diff] [blame]691<li><a href="http://llvm.org/PR1658">The C backend violates the ABI of common
692 C++ programs</a>, preventing intermixing between C++ compiled by the CBE and
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]693 C++ code compiled with llc or native compilers.</li>
Duncan Sandsbd0ad6e2008-02-10 13:40:55 +0000[diff] [blame]694<li>The C backend does not support all exception handling constructs.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]695</ul>
696
697</div>
698
699
700<!-- ======================================================================= -->
701<div class="doc_subsection">
Chris Lattner499c07c2008-06-05 06:35:40 +0000[diff] [blame]702 <a name="c-fe">Known problems with the llvm-gcc C front-end</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]703</div>
704
705<!-- _______________________________________________________________________ -->
706<div class="doc_subsubsection">Bugs</div>
707
708<div class="doc_text">
709
Chris Lattnere0798082008-02-06 18:00:06 +0000[diff] [blame]710<p>llvm-gcc does not currently support <a href="http://llvm.org/PR869">Link-Time
711Optimization</a> on most platforms "out-of-the-box". Please inquire on the
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]712llvmdev mailing list if you are interested.</p>
713
Chris Lattner499c07c2008-06-05 06:35:40 +0000[diff] [blame]714<p>The only major language feature of GCC not supported by llvm-gcc is
715 the <tt>__builtin_apply</tt> family of builtins. However, some extensions
716 are only supported on some targets. For example, trampolines are only
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]717 supported on some targets (these are used when you take the address of a
718 nested function).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]719
Chris Lattner499c07c2008-06-05 06:35:40 +0000[diff] [blame]720<p>If you run into GCC extensions which are not supported, please let us know.
721</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]722
723</div>
724
725<!-- ======================================================================= -->
726<div class="doc_subsection">
Chris Lattner499c07c2008-06-05 06:35:40 +0000[diff] [blame]727 <a name="c++-fe">Known problems with the llvm-gcc C++ front-end</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]728</div>
729
730<div class="doc_text">
731
732<p>The C++ front-end is considered to be fully
733tested and works for a number of non-trivial programs, including LLVM
734itself, Qt, Mozilla, etc.</p>
735
736<ul>
Chris Lattner499c07c2008-06-05 06:35:40 +0000[diff] [blame]737<li>Exception handling works well on the X86 and PowerPC targets, including
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]738X86-64 darwin. This works when linking to a libstdc++ compiled by GCC. It is
739supported on X86-64 linux, but that is disabled by default in this release.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]740</ul>
741
742</div>
743
744
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]745<!-- ======================================================================= -->
746<div class="doc_subsection">
Chris Lattner499c07c2008-06-05 06:35:40 +0000[diff] [blame]747 <a name="ada-fe">Known problems with the llvm-gcc Ada front-end</a>
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]748</div>
749
750<div class="doc_text">
751The llvm-gcc 4.2 Ada compiler works fairly well, however this is not a mature
752technology and problems should be expected.
753<ul>
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]754<li>The Ada front-end currently only builds on X86-32. This is mainly due
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]755to lack of trampoline support (pointers to nested functions) on other platforms,
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]756however it <a href="http://llvm.org/PR2006">also fails to build on X86-64</a>
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]757which does support trampolines.</li>
758<li>The Ada front-end <a href="http://llvm.org/PR2007">fails to bootstrap</a>.
759Workaround: configure with --disable-bootstrap.</li>
760<li>The c380004 and <a href="http://llvm.org/PR2010">c393010</a> ACATS tests
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]761fail (c380004 also fails with gcc-4.2 mainline). When built at -O3, the
762<a href="http://llvm.org/PR2421">cxg2021</a> ACATS test also fails.</li>
763<li>Some gcc specific Ada tests continue to crash the compiler. The testsuite
764reports most tests as having failed even though they pass.</li>
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]765<li>The -E binder option (exception backtraces)
766<a href="http://llvm.org/PR1982">does not work</a> and will result in programs
767crashing if an exception is raised. Workaround: do not use -E.</li>
768<li>Only discrete types <a href="http://llvm.org/PR1981">are allowed to start
769or finish at a non-byte offset</a> in a record. Workaround: do not pack records
770or use representation clauses that result in a field of a non-discrete type
771starting or finishing in the middle of a byte.</li>
Chris Lattnerd95b5962008-06-08 21:19:07 +0000[diff] [blame]772<li>The <tt>lli</tt> interpreter <a href="http://llvm.org/PR2009">considers
773'main' as generated by the Ada binder to be invalid</a>.
774Workaround: hand edit the file to use pointers for <tt>argv</tt> and
775<tt>envp</tt> rather than integers.</li>
776<li>The <tt>-fstack-check</tt> option <a href="http://llvm.org/PR2008">is
777ignored</a>.</li>
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]778</ul>
779</div>
780
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]781<!-- *********************************************************************** -->
782<div class="doc_section">
783 <a name="additionalinfo">Additional Information</a>
784</div>
785<!-- *********************************************************************** -->
786
787<div class="doc_text">
788
789<p>A wide variety of additional information is available on the <a
790href="http://llvm.org">LLVM web page</a>, in particular in the <a
791href="http://llvm.org/docs/">documentation</a> section. The web page also
792contains versions of the API documentation which is up-to-date with the
793Subversion version of the source code.
794You can access versions of these documents specific to this release by going
795into the "<tt>llvm/doc/</tt>" directory in the LLVM tree.</p>
796
797<p>If you have any questions or comments about LLVM, please feel free to contact
798us via the <a href="http://llvm.org/docs/#maillist"> mailing
799lists</a>.</p>
800
801</div>
802
803<!-- *********************************************************************** -->
804
805<hr>
806<address>
807 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
808 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
809 <a href="http://validator.w3.org/check/referer"><img
Chris Lattner5ba16862008-02-10 08:18:42 +0000[diff] [blame]810 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]811
812 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
813 Last modified: $Date$
814</address>
815
816</body>
817</html>