Gitiles
Code Review Sign In
gerrit-public.fairphone.software / fp2-dev / platform / external / llvm / 0437b2c059439411449e58f23dd173fd7f40abaf / . / docs / ReleaseNotes.html
blob: 97227446427aa62f732b5832ed1d9bf1ce38038d [file] [log] [blame]
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
3<html>
4<head>
5 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
6 <link rel="stylesheet" href="llvm.css" type="text/css">
Chris Lattner2c3ef982008-10-11 18:18:28 +0000[diff] [blame]7 <title>LLVM 2.4 Release Notes</title>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]8</head>
9<body>
10
Chris Lattner2c3ef982008-10-11 18:18:28 +0000[diff] [blame]11<div class="doc_title">LLVM 2.4 Release Notes</div>
Mikhail Glushenkov808cd3e2008-10-13 02:08:34 +0000[diff] [blame]12
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]13<ol>
14 <li><a href="#intro">Introduction</a></li>
Chris Lattner78a130b2008-10-13 18:01:01 +0000[diff] [blame]15 <li><a href="#subproj">Sub-project Status Update</a></li>
16 <li><a href="#whatsnew">What's New in LLVM?</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]17 <li><a href="GettingStarted.html">Installation Instructions</a></li>
18 <li><a href="#portability">Portability and Supported Platforms</a></li>
19 <li><a href="#knownproblems">Known Problems</a>
20 <li><a href="#additionalinfo">Additional Information</a></li>
21</ol>
22
23<div class="doc_author">
24 <p>Written by the <a href="http://llvm.org">LLVM Team</a><p>
25</div>
26
27<!-- *********************************************************************** -->
28<div class="doc_section">
29 <a name="intro">Introduction</a>
30</div>
31<!-- *********************************************************************** -->
32
33<div class="doc_text">
34
Chris Lattner92eacf82008-10-13 17:57:36 +0000[diff] [blame]35<p>This document contains the release notes for the LLVM Compiler
36Infrastructure, release 2.4. Here we describe the status of LLVM, including
37major improvements from the previous release and significant known problems.
38All LLVM releases may be downloaded from the <a
39href="http://llvm.org/releases/">LLVM releases web site</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]40
41<p>For more information about LLVM, including information about the latest
42release, please check out the <a href="http://llvm.org/">main LLVM
43web site</a>. If you have questions or comments, the <a
Chris Lattner92eacf82008-10-13 17:57:36 +0000[diff] [blame]44href="http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM Developer's Mailing
45List</a> is a good place to send them.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]46
Mikhail Glushenkov808cd3e2008-10-13 02:08:34 +0000[diff] [blame]47<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]48main LLVM web page, this document applies to the <i>next</i> release, not the
49current one. To see the release notes for a specific releases, please see the
50<a href="http://llvm.org/releases/">releases page</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]51
52</div>
53
Chris Lattner2c3ef982008-10-11 18:18:28 +0000[diff] [blame]54<!-- Unfinished features in 2.4:
Chris Lattner8fda4952008-02-10 07:46:44 +0000[diff] [blame]55 Machine LICM
56 Machine Sinking
57 LegalizeDAGTypes
Chris Lattner4f076402008-10-12 19:47:48 +0000[diff] [blame]58 llc -enable-value-prop, propagation of value info (sign/zero ext info) from
59 one MBB to another
Chris Lattner8fda4952008-02-10 07:46:44 +0000[diff] [blame]60 -->
Mikhail Glushenkov808cd3e2008-10-13 02:08:34 +0000[diff] [blame]61
Chris Lattner4f076402008-10-12 19:47:48 +0000[diff] [blame]62 <!-- for announcement email:
63 mention dev mtg
Chris Lattner548493b2008-10-12 22:57:58 +0000[diff] [blame]64 Xcode 3.1 and 3.1.1.
Chris Lattner4f076402008-10-12 19:47:48 +0000[diff] [blame]65 -->
Chris Lattner8fda4952008-02-10 07:46:44 +0000[diff] [blame]66
Chris Lattner78a130b2008-10-13 18:01:01 +0000[diff] [blame]67<!-- *********************************************************************** -->
68<div class="doc_section">
69 <a name="subproj">Sub-project Status Update</a>
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]70</div>
Chris Lattner78a130b2008-10-13 18:01:01 +0000[diff] [blame]71<!-- *********************************************************************** -->
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]72
73<div class="doc_text">
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]74<p>
Chris Lattner78a130b2008-10-13 18:01:01 +0000[diff] [blame]75The LLVM 2.4 distribution currently consists of code from the core LLVM
Chris Lattnera3c683b2008-10-13 18:11:54 +0000[diff] [blame]76repository (which roughly includes the LLVM optimizers, code generators and
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]77supporting tools) and the llvm-gcc repository. In addition to this code, the
78LLVM Project includes other sub-projects that are in development. The two which
Chris Lattner2c3ef982008-10-11 18:18:28 +0000[diff] [blame]79are the most actively developed are the <a href="#clang">Clang Project</a> and
Duncan Sands275d1e02008-10-13 18:33:28 +0000[diff] [blame]80the <a href="#vmkit">vmkit Project</a>.
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]81</p>
Chris Lattner78a130b2008-10-13 18:01:01 +0000[diff] [blame]82
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]83</div>
84
Chris Lattner78a130b2008-10-13 18:01:01 +0000[diff] [blame]85
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]86<!--=========================================================================-->
Chris Lattner78a130b2008-10-13 18:01:01 +0000[diff] [blame]87<div class="doc_subsection">
Chris Lattnera3c683b2008-10-13 18:11:54 +0000[diff] [blame]88<a name="clang">Clang: C/C++/Objective-C Frontend Toolkit</a>
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]89</div>
90
91<div class="doc_text">
92
93<p>The <a href="http://clang.llvm.org/">Clang project</a> is an effort to build
94a set of new 'LLVM native' front-end technologies for the LLVM optimizer
95and code generator. Clang is continuing to make major strides forward in all
96areas. Its C and Objective-C parsing support is very solid, and the code
97generation support is far enough along to build many C applications. While not
98yet production quality, it is progressing very nicely. In addition, C++
99front-end work has started to make significant progress.</p>
100
Chris Lattnerc2d84672008-10-14 00:52:49 +0000[diff] [blame]101<p>Codegen progress/state [DANIEL]</p>
Chris Lattner2c3ef982008-10-11 18:18:28 +0000[diff] [blame]102
Chris Lattnera3c683b2008-10-13 18:11:54 +0000[diff] [blame]103</div>
104
105<!--=========================================================================-->
106<div class="doc_subsection">
107<a name="clangsa">Clang Static Analyzer</a>
108</div>
109
110<div class="doc_text">
111
Ted Kremenek0437b2c2008-10-14 05:14:21 +0000[diff] [blame^]112<p>The Clang project also includes an early stage static source code analysis
113tool for <a href="http://clang.llvm.org/StaticAnalysis.html">automatically
114finding bugs</a> in C and Objective-C programs. The tool performs a growing set
115of checks to find bugs that occur on a specific path within a program. Examples
116of bugs the tool finds include logic errors such as null dereferences,
117violations of various API rules, dead code, and potential memory leaks in
118Objective-C programs. Since its public announcement at Apple's WWDC 2008
119conference, the public feedback on the tool has been extremely positive, and
120conservative estimates put the number of real bugs it has found in
121industrial-quality software on the order of thousands.</p>
Chris Lattnerc2d84672008-10-14 00:52:49 +0000[diff] [blame]122
Ted Kremenek0437b2c2008-10-14 05:14:21 +0000[diff] [blame^]123<p>The tool also provides a simple web GUI to inspect potential bugs found by
124the tool. While still early in development, the GUI illustrates some of the key
125features of Clang: accurate source location information, which is used by the
126GUI to highlight specific code expressions that relate to a bug (including those
127that span multiple lines) and built-in knowledge of macros, which is used to
128perform inline expansion of macros within the GUI itself.</p>
129
130<p>The set of checks perform by the static analyzer is gradually expanding, and
131future plans for the tool include full source-level inter-procedural analysis
132and deeper checks such as buffer overrun detection. There are many opportunities
133to extend and enhance the static analyzer, and anyone interested in working on
134this project is encouraged to get involved!</p>
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]135
136</div>
137
Chris Lattner78a130b2008-10-13 18:01:01 +0000[diff] [blame]138<!--=========================================================================-->
139<div class="doc_subsection">
Chris Lattnera3c683b2008-10-13 18:11:54 +0000[diff] [blame]140<a name="vmkit">vmkit: JVM/CLI Virtual Machine Implementation</a>
Chris Lattner78a130b2008-10-13 18:01:01 +0000[diff] [blame]141</div>
142
143<div class="doc_text">
144<p>
Chris Lattnera3c683b2008-10-13 18:11:54 +0000[diff] [blame]145The <a href="http://vmkit.llvm.org/">vmkit project</a> is an implementation of
Chris Lattner78a130b2008-10-13 18:01:01 +0000[diff] [blame]146a JVM and a CLI Virtual Machines (Microsoft .NET is an
147implementation of the CLI) using the Just-In-Time compiler of LLVM.</p>
148
Chris Lattnerc2d84672008-10-14 00:52:49 +0000[diff] [blame]149<p>[NICOLAS]</p>
Chris Lattner78a130b2008-10-13 18:01:01 +0000[diff] [blame]150
151</div>
152
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]153
154<!-- *********************************************************************** -->
155<div class="doc_section">
Chris Lattner78a130b2008-10-13 18:01:01 +0000[diff] [blame]156 <a name="whatsnew">What's New in LLVM?</a>
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]157</div>
158<!-- *********************************************************************** -->
159
Chris Lattner5cce7a52008-06-08 22:59:35 +0000[diff] [blame]160<div class="doc_text">
161
Chris Lattner92eacf82008-10-13 17:57:36 +0000[diff] [blame]162<p>This release includes a huge number of bug fixes, performance tweaks and
163minor improvements. Some of the major improvements and new features are listed
164in this section.
Chris Lattner5cce7a52008-06-08 22:59:35 +0000[diff] [blame]165</p>
166</div>
167
Chris Lattnerf6a0d382008-06-08 21:58:17 +0000[diff] [blame]168<!--=========================================================================-->
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]169<div class="doc_subsection">
Chris Lattnere0798082008-02-06 18:00:06 +0000[diff] [blame]170<a name="majorfeatures">Major New Features</a>
171</div>
172
173<div class="doc_text">
174
Chris Lattner2c3ef982008-10-11 18:18:28 +0000[diff] [blame]175<p>LLVM 2.4 includes several major new capabilities:</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]176
177<ul>
Chris Lattner4352cc82008-10-13 20:47:20 +0000[diff] [blame]178<li><p>The most visible end-user change in LLVM 2.4 is that it includes many
Chris Lattnerea9c2242008-10-13 21:50:36 +0000[diff] [blame]179optimizations and changes to make -O0 compile times much faster. You should see
Chris Lattner4352cc82008-10-13 20:47:20 +0000[diff] [blame]180improvements on the order of 30% or more faster than LLVM 2.3. There are many
181pieces to this change, described in more detail below. The speedups and new
182components can also be used for JIT compilers that want fast compilation as
183well.</p></li>
Chris Lattner2c3ef982008-10-11 18:18:28 +0000[diff] [blame]184
Chris Lattner4352cc82008-10-13 20:47:20 +0000[diff] [blame]185<li><p>The biggest change to the LLVM IR is that Multiple Return Values (which
186were introduced in LLVM 2.3) have been generalized to full support for "First
187Class Aggregate" values in LLVM 2.4. This means that LLVM IR supports using
188structs and arrays as values in a function. This capability is mostly useful
189for front-end authors, who prefer to treat things like complex numbers, simple
190tuples, dope vectors, etc as Value*'s instead of as a tuple of Value*'s or as
191memory values.</p></li>
Mikhail Glushenkov808cd3e2008-10-13 02:08:34 +0000[diff] [blame]192
Chris Lattner4352cc82008-10-13 20:47:20 +0000[diff] [blame]193<li><p>LLVM 2.4 also includes an initial port for the PIC16 microprocessor. This
194is the LLVM targer that only has support for 8 bit registers, and a number of
195other crazy constraints. While the port is still in early development stages,
196it shows some interesting things you can do with LLVM.</p></li>
Chris Lattnerba948522008-06-05 06:57:39 +0000[diff] [blame]197
Chris Lattner5ba16862008-02-10 08:18:42 +0000[diff] [blame]198</ul>
199
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]200</div>
201
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]202
203<!--=========================================================================-->
204<div class="doc_subsection">
Chris Lattnerf65224f2008-06-08 21:34:41 +0000[diff] [blame]205<a name="llvm-gcc">llvm-gcc 4.2 Improvements</a>
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]206</div>
207
208<div class="doc_text">
209
Chris Lattner4352cc82008-10-13 20:47:20 +0000[diff] [blame]210<p>LLVM fully supports the llvm-gcc 4.2 front-end, which marries the GCC
211front-ends and driver with the LLVM optimizer and code generator. It currently
212includes support for the C, C++, Objective-C, Ada, and Fortran front-ends.</p>
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]213
Chris Lattner5cce7a52008-06-08 22:59:35 +0000[diff] [blame]214<ul>
Chris Lattnerf7bae6d2008-10-13 20:59:32 +0000[diff] [blame]215<li>LLVM 2.4 supports the full set of atomic <tt>__sync_*</tt> builtins. LLVM
2162.3 only supported those used by OpenMP, but 2.4 supports them all. While
Chris Lattnerea9c2242008-10-13 21:50:36 +0000[diff] [blame]217llvm-gcc supports all of these builtins, note that not all targets do. X86
218support them all in both 32-bit and 64-bit mode and PowerPC supports them all
219except for the 64-bit operations when in 32-bit mode.</li>
Chris Lattner5cce7a52008-06-08 22:59:35 +0000[diff] [blame]220
Chris Lattnerf7bae6d2008-10-13 20:59:32 +0000[diff] [blame]221<li>llvm-gcc now supports an <tt>-flimited-precision</tt> option, which tells
222the compiler that it is ok to use low-precision approximations of certain libm
Chris Lattner4352cc82008-10-13 20:47:20 +0000[diff] [blame]223functions (like tan, log, etc). This allows you to get high performance if you
224only need (say) 14-bits of precision.</li>
225
226<li>llvm-gcc now supports a C language extension known as "<a
227href="http://lists.cs.uiuc.edu/pipermail/cfe-dev/2008-August/002670.html">Blocks
228</a>. This feature is similar to nested functions and closures, but does not
229require stack trampolines (with most ABIs) and supports returning closures
230from functions that define them. Note that actually <em>using</em> Blocks
231requires a small runtime that is not included with llvm-gcc.</li>
232
233<li>llvm-gcc now supports a new <tt>-flto</tt> option. On systems that support
234transparent Link Time Optimization (currently Darwin systems with Xcode 3.1 and
235later) this allows the use of LTO with other optimization levels like -Os.
236Previously, LTO could only be used with -O4, which implied optimizations in
237-O3 that can increase code size.</li>
Gabor Greif0e49cc82008-06-09 06:06:18 +0000[diff] [blame]238</ul>
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]239
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]240</div>
241
242
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]243<!--=========================================================================-->
244<div class="doc_subsection">
Chris Lattnerc3649452008-02-10 08:17:19 +0000[diff] [blame]245<a name="coreimprovements">LLVM Core Improvements</a>
246</div>
247
248<div class="doc_text">
249<p>New features include:
250</p>
251
Chris Lattnerc5a4e312008-06-05 06:25:56 +0000[diff] [blame]252<ul>
Chris Lattnerf7bae6d2008-10-13 20:59:32 +0000[diff] [blame]253<li>A major change to the "Use" class landed, which shrank it by 25%. Since
254this is a pervasive part of the LLVM, it ended up reducing the memory use of
255LLVM IR in general by 15% for most programs.</li>
256
257<li>Values with no names are now pretty printed by <tt>llvm-dis</tt> more
258nicely. They now print as "<tt>%3 = add i32 %A, 4</tt>" instead of
259"<tt>add i32 %A, 4 ; &lt;i32&gt;:3</tt>", which makes it much easier to read.
260</li>
261
262<li>LLVM 2.4 includes some changes for better vector support. First, the shift
263operations (<tt>shl</tt>, <tt>ashr</tt>, <tt>lshr</tt>) now all support vectors
264and do an element-by-element shift (shifts of the whole vector can be
265accomplished by bitcasting the vector to &lt;1 x i128&gt; for example). Second,
Chris Lattnerc2d84672008-10-14 00:52:49 +0000[diff] [blame]266there is initial support in development for vector comparisons with the
267<a href="LangRef.html#i_fcmp">fcmp</a>/<a href="LangRef.html#i_icmp">icmp</a>
268instructions. These instructions compare two vectors and return a vector of
269i1's for each result. Note that there is very little codegen support available
270for any of these IR features though.</li>
Chris Lattnerf7bae6d2008-10-13 20:59:32 +0000[diff] [blame]271
272<li>A new <tt>DebugInfoBuilder</tt> class is available, which makes it much
273easier for front-ends to create debug info descriptors, similar to the way that
274IRBuilder makes it easier to create LLVM IR.</li>
275
276<li>LLVM now supports "function attributes", which allows us to seperate return
277value attributes from function attributes. LLVM now supports attributes on a
278function itself, a return value, and its parameters. New supported function
279attributes include noinline/alwaysinline and the "opt-size" flag which says the
280function should be optimized for code size.</li>
281
Chris Lattnera5a49382008-10-12 18:30:33 +0000[diff] [blame]282<li>LLVM IR now directly represents "common" linkage, instead of
283 representing it as a form of weak linkage.</li>
Chris Lattnerf7bae6d2008-10-13 20:59:32 +0000[diff] [blame]284
Chris Lattnerc3649452008-02-10 08:17:19 +0000[diff] [blame]285</ul>
Mikhail Glushenkov808cd3e2008-10-13 02:08:34 +0000[diff] [blame]286
Chris Lattnerc3649452008-02-10 08:17:19 +0000[diff] [blame]287</div>
288
289<!--=========================================================================-->
290<div class="doc_subsection">
Chris Lattner8fda4952008-02-10 07:46:44 +0000[diff] [blame]291<a name="optimizer">Optimizer Improvements</a>
292</div>
293
294<div class="doc_text">
295
Chris Lattnerea9c2242008-10-13 21:50:36 +0000[diff] [blame]296<p>In addition to a huge array of bug fixes and minor performance tweaks, this
297release includes a few major enhancements and additions to the optimizers:</p>
Chris Lattner8fda4952008-02-10 07:46:44 +0000[diff] [blame]298
299<ul>
300
Chris Lattnerea9c2242008-10-13 21:50:36 +0000[diff] [blame]301<li>The Global Value Numbering (GVN) pass now does local Partial Redundancy
302Elimination (PRE) to eliminate some partially redundant expressions in cases
303where doing so won't grow code size.</li>
Chris Lattner4f076402008-10-12 19:47:48 +0000[diff] [blame]304
Chris Lattnerea9c2242008-10-13 21:50:36 +0000[diff] [blame]305<li>LLVM 2.4 includes a new loop deletion pass (which removes output-free
306provably-finite loops) and a rewritten Aggressive Dead Code Elimination (ADCE)
307pass that no longer uses control dependence information. These changes speed up
308the optimizer and also prevents it from deleting output-free infinite
309loops.</li>
Chris Lattner4f076402008-10-12 19:47:48 +0000[diff] [blame]310
Duncan Sands2aecc3f2008-10-13 18:42:43 +0000[diff] [blame]311<li>The new AddReadAttrs pass works out which functions are read-only or
312read-none (these correspond to 'pure' and 'const' in C) and marks them
313with the appropriate attribute.</li>
Chris Lattnerea9c2242008-10-13 21:50:36 +0000[diff] [blame]314
315<li>LLVM 2.4 now includes a new SparsePropagation framework, which makes it
316trivial to build lattice-based dataflow solvers that operate over LLVM IR. Using
317this interface means that you just define objects to represent your lattice
318values and the transfer functions that operate on them. It handles the
319mechanics of worklist processing, liveness tracking, handling PHI nodes,
320etc.</li>
321
322<li>Various helper functions (ComputeMaskedBits, ComputeNumSignBits, etc) were
323pulled out of the Instruction Combining pass and put into a new
324<tt>ValueTracking.h</tt> header, where they can be reused by other passes.</li>
325
326<li>The tail duplication pass has been removed from the standard optimizer
327sequence used by llvm-gcc. This pass still exists, but the benefits it once
328provided are now achieved by other passes.</li>
329
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]330</ul>
331
332</div>
333
334<!--=========================================================================-->
335<div class="doc_subsection">
336<a name="codegen">Code Generator Improvements</a>
337</div>
338
339<div class="doc_text">
340
341<p>We put a significant amount of work into the code generator infrastructure,
342which allows us to implement more aggressive algorithms and make it run
343faster:</p>
344
345<ul>
Chris Lattnerea9c2242008-10-13 21:50:36 +0000[diff] [blame]346<li>The target-independent code generator supports (and the X86 backend
347 currently implements) a new interface for "fast" instruction selection. This
348 interface is optimized to produce code as quickly as possible, sacrificing
349 code quality to do it. This is used by default at -O0 or when using
350 "llc -fast" on X86. It is straight-forward to add support for
351 other targets if faster -O0 compilation is desired.</li>
352
353<li>In addition to the new 'fast' instruction selection path, many existing
354 pieces of the code generator have been optimized in significant ways.
355 SelectionDAG's are now pool allocated and use better algorithms in many
356 places, the ".s" file printers now use raw_ostream to emit text much faster,
357 etc. The end result of these improvements is that the compiler also takes
358 substantially less time to generate code that is just as good (and often
359 better) than before.</li>
360
361<li>Each target has been split to separate the .s file printing logic from the
362 rest of the target. This enables JIT compilers that don't link in the
363 (somewhat large) code and data tables used for printing a .s file.</li>
364
365<li>The code generator now includes a "stack slot coloring" pass, which packs
366 together individual spilled values into common stack slots. This reduces
367 the size of stack frames with many spills, which tends to increase L1 cache
368 effectiveness.</li>
369
370<li>Various pieces of the register allocator (e.g. the coalescer and two-address
371 operation elimination pass) now know how to rematerialize trivial operations
372 to avoid copies and include several other optimizations.</li>
373
374<li>The <a href="CodeGenerator.html#selectiondag_process">graphs</a> produced by
375 the <tt>llc -view-*-dags</tt> options are now significantly prettier and
376 easier to read.</li>
377
378<li>LLVM 2.4 includes a new register allocator based on Partitioned Boolean
379 Quadratic Programming (PBQP). This register allocator is still in
380 development, but is very simple and clean.</li>
Chris Lattner0eef6e42008-06-08 02:45:07 +0000[diff] [blame]381
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]382</ul>
383
384</div>
385
386
387<!--=========================================================================-->
388<div class="doc_subsection">
Chris Lattnercc64a322008-10-13 22:06:31 +0000[diff] [blame]389<a name="targetspecific">Target Specific Improvements</a>
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]390</div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]391
392<div class="doc_text">
Chris Lattnerc3649452008-02-10 08:17:19 +0000[diff] [blame]393<p>New target-specific features include:
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]394</p>
395
396<ul>
Chris Lattner4f076402008-10-12 19:47:48 +0000[diff] [blame]397<li>Exception handling is supported by default on Linux/x86-64.</li>
Duncan Sandsd8745a92008-10-13 17:32:15 +0000[diff] [blame]398<li>Position Independent Code (PIC) is now supported on Linux/x86-64.</li>
Chris Lattnercc64a322008-10-13 22:06:31 +0000[diff] [blame]399<li>MIPS floating point support? [BRUNO]</li>
400<li>The PowerPC backend now supports trampolines.</li>
Chris Lattnere45f18b2008-06-05 08:02:49 +0000[diff] [blame]401</ul>
Chris Lattnere45f18b2008-06-05 08:02:49 +0000[diff] [blame]402
Mikhail Glushenkov808cd3e2008-10-13 02:08:34 +0000[diff] [blame]403</div>
Chris Lattnere45f18b2008-06-05 08:02:49 +0000[diff] [blame]404
405
406<!--=========================================================================-->
407<div class="doc_subsection">
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]408<a name="otherimprovements">Other Improvements</a>
409</div>
410
411<div class="doc_text">
412<p>New features include:
413</p>
414
415<ul>
Mikhail Glushenkovcc2d3012008-10-13 02:46:01 +0000[diff] [blame]416<li><tt>llvmc2</tt> (the generic compiler driver) gained plugin
Chris Lattnercc64a322008-10-13 22:06:31 +0000[diff] [blame]417 support. It is now easier to experiment with <tt>llvmc2</tt> and
418 build your own tools based on it.</li>
419
420<li>LLVM 2.4 includes a number of new generic algorithms and data structures,
421 include a scoped hash table, 'immutable' data structures, and a
422 <tt>raw_ostream</tt> class. The <tt>raw_ostream</tt> class and
423 <tt>format</tt> allow for efficient file output, and various pieces of LLVM
424 have switched over to use it. The eventual goal is to eliminate
425 std::ostream in favor of it.</li>
426
427<li>Recycler + pool allocation stuff? [DAN]</li>
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]428</ul>
Mikhail Glushenkov808cd3e2008-10-13 02:08:34 +0000[diff] [blame]429
Chris Lattnerb0d6c502007-09-21 03:54:09 +0000[diff] [blame]430</div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]431
Chris Lattner92eacf82008-10-13 17:57:36 +0000[diff] [blame]432<!--=========================================================================-->
433<div class="doc_subsection">
434<a name="changes">Major Changes and Removed Features</a>
435</div>
436
437<div class="doc_text">
438
Chris Lattnercc64a322008-10-13 22:06:31 +0000[diff] [blame]439<p>If you're already an LLVM user or developer with out-of-tree changes based
440on LLVM 2.3, this section lists some "gotchas" that you may run into upgrading
441from the previous release.</p>
Chris Lattner92eacf82008-10-13 17:57:36 +0000[diff] [blame]442
443<ul>
444
Chris Lattnercc64a322008-10-13 22:06:31 +0000[diff] [blame]445<li>The LLVM IR generated by llvm-gcc no longer names all instructions. This
446 makes it run faster, but may be more confusing to some people. If you
447 prefer to have names, the '<tt>opt -instnamer</tt>' pass will add names to
448 all instructions.</li>
449
450<li>The LoadVN and GCSE passes have been removed from the tree. They are
451 obsolete and have been replaced with the GVN and MemoryDependence passes.
452 </li>
453</ul>
454
455
456<p>In addition, many APIs have changed in this release. Some of the major LLVM
457API changes are:</p>
458
459<ul>
460
461<li>Attributes changes [DEVANG] </li>
Chris Lattner92eacf82008-10-13 17:57:36 +0000[diff] [blame]462
463<li>The <tt>DbgStopPointInst</tt> methods <tt>getDirectory</tt> and
464<tt>getFileName</tt> now return <tt>Value*</tt> instead of strings. These can be
465converted to strings using <tt>llvm::GetConstantStringInfo</tt> defined via
466"llvm/Analysis/ValueTracking.h".</li>
467
Chris Lattnercc64a322008-10-13 22:06:31 +0000[diff] [blame]468<li>The APIs to create various instructions have changed from lower case
469 "create" methods to upper case "Create" methods (e.g.
470 <tt>BinaryOperator::create</tt>). LLVM 2.4 includes both cases, but the
471 lower case ones are removed in mainline, please migrate.</li>
472
Chris Lattner92eacf82008-10-13 17:57:36 +0000[diff] [blame]473<li>Various header files like "llvm/ADT/iterator" were given a .h suffix.
474 Change your code to #include "llvm/ADT/iterator.h" instead.</li>
475
476</ul>
477
478</div>
479
480
481
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]482<!-- *********************************************************************** -->
483<div class="doc_section">
484 <a name="portability">Portability and Supported Platforms</a>
485</div>
486<!-- *********************************************************************** -->
487
488<div class="doc_text">
489
490<p>LLVM is known to work on the following platforms:</p>
491
492<ul>
Mikhail Glushenkov808cd3e2008-10-13 02:08:34 +0000[diff] [blame]493<li>Intel and AMD machines (IA32) running Red Hat Linux, Fedora Core and FreeBSD
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]494 (and probably other unix-like systems).</li>
Chris Lattnerbd633de92008-02-06 06:30:34 +0000[diff] [blame]495<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]496 64-bit modes.</li>
Chris Lattnere0798082008-02-06 18:00:06 +0000[diff] [blame]497<li>Intel and AMD machines running on Win32 using MinGW libraries (native).</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]498<li>Intel and AMD machines running on Win32 with the Cygwin libraries (limited
499 support is available for native builds with Visual C++).</li>
Gabor Greif5e75e652008-06-05 18:39:01 +0000[diff] [blame]500<li>Sun UltraSPARC workstations running Solaris 10.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]501<li>Alpha-based machines running Debian GNU/Linux.</li>
Gabor Greif5e75e652008-06-05 18:39:01 +0000[diff] [blame]502<li>Itanium-based (IA64) machines running Linux and HP-UX.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]503</ul>
504
Chris Lattnerba948522008-06-05 06:57:39 +0000[diff] [blame]505<p>The core LLVM infrastructure uses GNU autoconf to adapt itself
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]506to the machine and operating system on which it is built. However, minor
507porting may be required to get LLVM to work on new platforms. We welcome your
508portability patches and reports of successful builds or error messages.</p>
509
510</div>
511
512<!-- *********************************************************************** -->
513<div class="doc_section">
514 <a name="knownproblems">Known Problems</a>
515</div>
516<!-- *********************************************************************** -->
517
518<div class="doc_text">
519
520<p>This section contains all known problems with the LLVM system, listed by
521component. As new problems are discovered, they will be added to these
522sections. If you run into a problem, please check the <a
523href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
524there isn't already one.</p>
525
526</div>
527
528<!-- ======================================================================= -->
529<div class="doc_subsection">
530 <a name="experimental">Experimental features included with this release</a>
531</div>
532
533<div class="doc_text">
534
535<p>The following components of this LLVM release are either untested, known to
536be broken or unreliable, or are in early development. These components should
537not be relied on, and bugs should not be filed against them, but they may be
538useful 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]539components, please contact us on the <a
540href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev list</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]541
542<ul>
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]543<li>The MSIL, IA64, Alpha, SPU, and MIPS backends are experimental.</li>
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]544<li>The llc "<tt>-filetype=asm</tt>" (the default) is the only supported
Chris Lattnerbd633de92008-02-06 06:30:34 +0000[diff] [blame]545 value for this option.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]546</ul>
547
548</div>
549
550<!-- ======================================================================= -->
551<div class="doc_subsection">
552 <a name="x86-be">Known problems with the X86 back-end</a>
553</div>
554
555<div class="doc_text">
556
557<ul>
Anton Korobeynikov77d15e22008-06-08 10:24:13 +0000[diff] [blame]558 <li>The X86 backend does not yet support
559 all <a href="http://llvm.org/PR879">inline assembly that uses the X86
560 floating point stack</a>. It supports the 'f' and 't' constraints, but not
561 'u'.</li>
Chris Lattnerd95b5962008-06-08 21:19:07 +0000[diff] [blame]562 <li>The X86 backend generates inefficient floating point code when configured
563 to generate code for systems that don't have SSE2.</li>
Duncan Sands8b4f3772008-06-08 19:38:43 +0000[diff] [blame]564 <li>Win64 code generation wasn't widely tested. Everything should work, but we
Anton Korobeynikov77d15e22008-06-08 10:24:13 +0000[diff] [blame]565 expect small issues to happen. Also, llvm-gcc cannot build mingw64 runtime
566 currently due
567 to <a href="http://llvm.org/PR2255">several</a>
Chris Lattnerf06141a2008-06-08 23:12:47 +0000[diff] [blame]568 <a href="http://llvm.org/PR2257">bugs</a> due to lack of support for the
569 'u' inline assembly constraint and X87 floating point inline assembly.</li>
Dan Gohman3882ecb2008-06-08 23:05:11 +0000[diff] [blame]570 <li>The X86-64 backend does not yet support the LLVM IR instruction
571 <tt>va_arg</tt>. Currently, the llvm-gcc front-end supports variadic
572 argument constructs on X86-64 by lowering them manually.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]573</ul>
574
575</div>
576
577<!-- ======================================================================= -->
578<div class="doc_subsection">
579 <a name="ppc-be">Known problems with the PowerPC back-end</a>
580</div>
581
582<div class="doc_text">
583
584<ul>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]585<li>The Linux PPC32/ABI support needs testing for the interpreter and static
586compilation, and lacks support for debug information.</li>
587</ul>
588
589</div>
590
591<!-- ======================================================================= -->
592<div class="doc_subsection">
593 <a name="arm-be">Known problems with the ARM back-end</a>
594</div>
595
596<div class="doc_text">
597
598<ul>
599<li>Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6
Duncan Sands1827fa62007-09-26 15:59:54 +0000[diff] [blame]600processors, thumb programs can crash or produce wrong
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]601results (<a href="http://llvm.org/PR1388">PR1388</a>).</li>
602<li>Compilation for ARM Linux OABI (old ABI) is supported, but not fully tested.
603</li>
Chris Lattnerd95b5962008-06-08 21:19:07 +0000[diff] [blame]604<li>There is a bug in QEMU-ARM (&lt;= 0.9.0) which causes it to incorrectly
605 execute
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]606programs compiled with LLVM. Please use more recent versions of QEMU.</li>
607</ul>
608
609</div>
610
611<!-- ======================================================================= -->
612<div class="doc_subsection">
613 <a name="sparc-be">Known problems with the SPARC back-end</a>
614</div>
615
616<div class="doc_text">
617
618<ul>
619<li>The SPARC backend only supports the 32-bit SPARC ABI (-m32), it does not
620 support the 64-bit SPARC ABI (-m64).</li>
621</ul>
622
623</div>
624
625<!-- ======================================================================= -->
626<div class="doc_subsection">
627 <a name="alpha-be">Known problems with the Alpha back-end</a>
628</div>
629
630<div class="doc_text">
631
632<ul>
633
634<li>On 21164s, some rare FP arithmetic sequences which may trap do not have the
635appropriate nops inserted to ensure restartability.</li>
636
637</ul>
638</div>
639
640<!-- ======================================================================= -->
641<div class="doc_subsection">
642 <a name="ia64-be">Known problems with the IA64 back-end</a>
643</div>
644
645<div class="doc_text">
646
647<ul>
Chris Lattnerf06141a2008-06-08 23:12:47 +0000[diff] [blame]648<li>The Itanium backend is highly experimental, and has a number of known
649 issues. We are looking for a maintainer for the Itanium backend. If you
650 are interested, please contact the llvmdev mailing list.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]651</ul>
652
653</div>
654
655<!-- ======================================================================= -->
656<div class="doc_subsection">
657 <a name="c-be">Known problems with the C back-end</a>
658</div>
659
660<div class="doc_text">
661
662<ul>
Chris Lattner499c07c2008-06-05 06:35:40 +0000[diff] [blame]663<li><a href="http://llvm.org/PR802">The C backend has only basic support for
664 inline assembly code</a>.</li>
Chris Lattnere0194462007-09-26 06:01:35 +0000[diff] [blame]665<li><a href="http://llvm.org/PR1658">The C backend violates the ABI of common
666 C++ programs</a>, preventing intermixing between C++ compiled by the CBE and
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]667 C++ code compiled with llc or native compilers.</li>
Duncan Sandsbd0ad6e2008-02-10 13:40:55 +0000[diff] [blame]668<li>The C backend does not support all exception handling constructs.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]669</ul>
670
671</div>
672
673
674<!-- ======================================================================= -->
675<div class="doc_subsection">
Chris Lattner499c07c2008-06-05 06:35:40 +0000[diff] [blame]676 <a name="c-fe">Known problems with the llvm-gcc C front-end</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]677</div>
678
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]679<div class="doc_text">
680
Chris Lattnere0798082008-02-06 18:00:06 +0000[diff] [blame]681<p>llvm-gcc does not currently support <a href="http://llvm.org/PR869">Link-Time
682Optimization</a> on most platforms "out-of-the-box". Please inquire on the
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]683llvmdev mailing list if you are interested.</p>
684
Chris Lattner499c07c2008-06-05 06:35:40 +0000[diff] [blame]685<p>The only major language feature of GCC not supported by llvm-gcc is
686 the <tt>__builtin_apply</tt> family of builtins. However, some extensions
687 are only supported on some targets. For example, trampolines are only
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]688 supported on some targets (these are used when you take the address of a
689 nested function).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]690
Chris Lattner499c07c2008-06-05 06:35:40 +0000[diff] [blame]691<p>If you run into GCC extensions which are not supported, please let us know.
692</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]693
694</div>
695
696<!-- ======================================================================= -->
697<div class="doc_subsection">
Chris Lattner499c07c2008-06-05 06:35:40 +0000[diff] [blame]698 <a name="c++-fe">Known problems with the llvm-gcc C++ front-end</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]699</div>
700
701<div class="doc_text">
702
703<p>The C++ front-end is considered to be fully
704tested and works for a number of non-trivial programs, including LLVM
705itself, Qt, Mozilla, etc.</p>
706
707<ul>
Anton Korobeynikovd9a17932008-10-11 18:27:16 +0000[diff] [blame]708<li>Exception handling works well on the X86 and PowerPC targets. Currently
709 only linux and darwin targets are supported (both 32 and 64 bit).</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]710</ul>
711
712</div>
713
714
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]715<!-- ======================================================================= -->
716<div class="doc_subsection">
Chris Lattner499c07c2008-06-05 06:35:40 +0000[diff] [blame]717 <a name="ada-fe">Known problems with the llvm-gcc Ada front-end</a>
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]718</div>
719
720<div class="doc_text">
721The llvm-gcc 4.2 Ada compiler works fairly well, however this is not a mature
722technology and problems should be expected.
723<ul>
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]724<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]725to lack of trampoline support (pointers to nested functions) on other platforms,
Duncan Sandsa73561c2008-06-08 20:18:35 +0000[diff] [blame]726however 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]727which does support trampolines.</li>
728<li>The Ada front-end <a href="http://llvm.org/PR2007">fails to bootstrap</a>.
729Workaround: configure with --disable-bootstrap.</li>
Duncan Sands2dbc2e82008-10-13 17:27:23 +0000[diff] [blame]730<li>The c380004, <a href="http://llvm.org/PR2010">c393010</a>
731and <a href="http://llvm.org/PR2421">cxg2021</a> ACATS tests fail
732(c380004 also fails with gcc-4.2 mainline).</li>
733<li>Some gcc specific Ada tests continue to crash the compiler.</li>
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]734<li>The -E binder option (exception backtraces)
735<a href="http://llvm.org/PR1982">does not work</a> and will result in programs
736crashing if an exception is raised. Workaround: do not use -E.</li>
737<li>Only discrete types <a href="http://llvm.org/PR1981">are allowed to start
738or finish at a non-byte offset</a> in a record. Workaround: do not pack records
739or use representation clauses that result in a field of a non-discrete type
740starting or finishing in the middle of a byte.</li>
Chris Lattnerd95b5962008-06-08 21:19:07 +0000[diff] [blame]741<li>The <tt>lli</tt> interpreter <a href="http://llvm.org/PR2009">considers
742'main' as generated by the Ada binder to be invalid</a>.
743Workaround: hand edit the file to use pointers for <tt>argv</tt> and
744<tt>envp</tt> rather than integers.</li>
745<li>The <tt>-fstack-check</tt> option <a href="http://llvm.org/PR2008">is
746ignored</a>.</li>
Chris Lattnerdaeb3232008-02-12 06:29:45 +0000[diff] [blame]747</ul>
748</div>
749
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]750<!-- *********************************************************************** -->
751<div class="doc_section">
752 <a name="additionalinfo">Additional Information</a>
753</div>
754<!-- *********************************************************************** -->
755
756<div class="doc_text">
757
758<p>A wide variety of additional information is available on the <a
759href="http://llvm.org">LLVM web page</a>, in particular in the <a
760href="http://llvm.org/docs/">documentation</a> section. The web page also
761contains versions of the API documentation which is up-to-date with the
762Subversion version of the source code.
763You can access versions of these documents specific to this release by going
764into the "<tt>llvm/doc/</tt>" directory in the LLVM tree.</p>
765
766<p>If you have any questions or comments about LLVM, please feel free to contact
767us via the <a href="http://llvm.org/docs/#maillist"> mailing
768lists</a>.</p>
769
770</div>
771
772<!-- *********************************************************************** -->
773
774<hr>
775<address>
776 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
777 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
778 <a href="http://validator.w3.org/check/referer"><img
Chris Lattner5ba16862008-02-10 08:18:42 +0000[diff] [blame]779 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]780
781 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
782 Last modified: $Date$
783</address>
784
785</body>
786</html>