1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
6 <link rel="stylesheet" href="llvm.css" type="text/css">
7 <title>LLVM 2.6 Release Notes</title>
11 <div class="doc_title">LLVM 2.6 Release Notes</div>
14 <li><a href="#intro">Introduction</a></li>
15 <li><a href="#subproj">Sub-project Status Update</a></li>
16 <li><a href="#externalproj">External Projects Using LLVM 2.6</a></li>
17 <li><a href="#whatsnew">What's New in LLVM 2.6?</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></li>
21 <li><a href="#additionalinfo">Additional Information</a></li>
24 <div class="doc_author">
25 <p>Written by the <a href="http://llvm.org">LLVM Team</a></p>
28 <!-- *********************************************************************** -->
29 <div class="doc_section">
30 <a name="intro">Introduction</a>
32 <!-- *********************************************************************** -->
34 <div class="doc_text">
36 <p>This document contains the release notes for the LLVM Compiler
37 Infrastructure, release 2.6. Here we describe the status of LLVM, including
38 major improvements from the previous release and significant known problems.
39 All LLVM releases may be downloaded from the <a
40 href="http://llvm.org/releases/">LLVM releases web site</a>.</p>
42 <p>For more information about LLVM, including information about the latest
43 release, please check out the <a href="http://llvm.org/">main LLVM
44 web site</a>. If you have questions or comments, the <a
45 href="http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM Developer's Mailing
46 List</a> is a good place to send them.</p>
48 <p>Note that if you are reading this file from a Subversion checkout or the
49 main LLVM web page, this document applies to the <i>next</i> release, not the
50 current one. To see the release notes for a specific release, please see the
51 <a href="http://llvm.org/releases/">releases page</a>.</p>
58 include/llvm/Analysis/LiveValues.h => Dan
59 lib/Transforms/IPO/MergeFunctions.cpp => consider for 2.8.
60 llvm/Analysis/PointerTracking.h => Edwin wants this, consider for 2.8.
64 <!-- Unfinished features in 2.6:
67 variable debug info for optimized code
68 postalloc scheduler: anti dependence breaking, hazard recognizer?
70 loop dependence analysis
71 ELF Writer? How stable?
72 <li>PostRA scheduler improvements, ARM adoption (David Goodwin).</li>
73 2.7 supports the GDB 7.0 jit interfaces for debug info.
76 <!-- for announcement email:
80 KLEE web page at klee.llvm.org
81 Many new papers added to /pubs/
86 <!-- *********************************************************************** -->
87 <div class="doc_section">
88 <a name="subproj">Sub-project Status Update</a>
90 <!-- *********************************************************************** -->
92 <div class="doc_text">
94 The LLVM 2.6 distribution currently consists of code from the core LLVM
95 repository (which roughly includes the LLVM optimizers, code generators
96 and supporting tools), the Clang repository and the llvm-gcc repository. In
97 addition to this code, the LLVM Project includes other sub-projects that are in
98 development. Here we include updates on these subprojects.
104 <!--=========================================================================-->
105 <div class="doc_subsection">
106 <a name="clang">Clang: C/C++/Objective-C Frontend Toolkit</a>
109 <div class="doc_text">
111 <p>The <a href="http://clang.llvm.org/">Clang project</a> is an effort to build
112 a set of new 'LLVM native' front-end technologies for the C family of languages.
113 LLVM 2.6 is the first release to officially include Clang, and it provides a
114 production quality C and Objective-C compiler. If you are interested in <a
115 href="http://clang.llvm.org/performance.html">fast compiles</a> and
116 <a href="http://clang.llvm.org/diagnostics.html">good diagnostics</a>, we
117 encourage you to try it out. Clang currently compiles typical Objective-C code
118 3x faster than GCC and compiles C code about 30% faster than GCC at -O0 -g
119 (which is when the most pressure is on the frontend).</p>
121 <p>In addition to supporting these languages, C++ support is also <a
122 href="http://clang.llvm.org/cxx_status.html">well under way</a>, and mainline
123 Clang is able to parse the libstdc++ 4.2 headers and even codegen simple apps.
124 If you are interested in Clang C++ support or any other Clang feature, we
125 strongly encourage you to get involved on the <a
126 href="http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev">Clang front-end mailing
129 <p>In the LLVM 2.6 time-frame, the Clang team has made many improvements:</p>
132 <li>C and Objective-C support are now considered production quality.</li>
133 <li>AuroraUX, FreeBSD, and OpenBSD are now supported.</li>
134 <li>Most of Objective-C 2.0 is now supported with the GNU runtime.</li>
135 <li>Many many bugs are fixed and many features have been added.</li>
139 <!--=========================================================================-->
140 <div class="doc_subsection">
141 <a name="clangsa">Clang Static Analyzer</a>
144 <div class="doc_text">
146 <p>Previously announced in the 2.4 and 2.5 LLVM releases, the Clang project also
147 includes an early stage static source code analysis tool for <a
148 href="http://clang.llvm.org/StaticAnalysis.html">automatically finding bugs</a>
149 in C and Objective-C programs. The tool performs checks to find
150 bugs that occur on a specific path within a program.</p>
152 <p>In the LLVM 2.6 time-frame, the analyzer core has undergone several important
153 improvements and cleanups and now includes a new <em>Checker</em> interface that
154 is intended to eventually serve as a basis for domain-specific checks. Further,
155 in addition to generating HTML files for reporting analysis results, the
156 analyzer can now also emit bug reports in a structured XML format that is
157 intended to be easily readable by other programs.</p>
159 <p>The set of checks performed by the static analyzer continues to expand, and
160 future plans for the tool include full source-level inter-procedural analysis
161 and deeper checks such as buffer overrun detection. There are many opportunities
162 to extend and enhance the static analyzer, and anyone interested in working on
163 this project is encouraged to get involved!</p>
167 <!--=========================================================================-->
168 <div class="doc_subsection">
169 <a name="vmkit">VMKit: JVM/CLI Virtual Machine Implementation</a>
172 <div class="doc_text">
174 The <a href="http://vmkit.llvm.org/">VMKit project</a> is an implementation of
175 a JVM and a CLI Virtual Machine (Microsoft .NET is an
176 implementation of the CLI) using LLVM for static and just-in-time
180 VMKit version 0.26 builds with LLVM 2.6 and you can find it on its
181 <a href="http://vmkit.llvm.org/releases/">web page</a>. The release includes
182 bug fixes, cleanup and new features. The major changes are:</p>
186 <li>A new llcj tool to generate shared libraries or executables of Java
188 <li>Cooperative garbage collection. </li>
189 <li>Fast subtype checking (paper from Click et al [JGI'02]). </li>
190 <li>Implementation of a two-word header for Java objects instead of the original
191 three-word header. </li>
192 <li>Better Java specification-compliance: division by zero checks, stack
193 overflow checks, finalization and references support. </li>
199 <!--=========================================================================-->
200 <div class="doc_subsection">
201 <a name="compiler-rt">compiler-rt: Compiler Runtime Library</a>
204 <div class="doc_text">
206 The new LLVM <a href="http://compiler-rt.llvm.org/">compiler-rt project</a>
207 is a simple library that provides an implementation of the low-level
208 target-specific hooks required by code generation and other runtime components.
209 For example, when compiling for a 32-bit target, converting a double to a 64-bit
210 unsigned integer is compiled into a runtime call to the "__fixunsdfdi"
211 function. The compiler-rt library provides highly optimized implementations of
212 this and other low-level routines (some are 3x faster than the equivalent
213 libgcc routines).</p>
216 All of the code in the compiler-rt project is available under the standard LLVM
217 License, a "BSD-style" license.</p>
221 <!--=========================================================================-->
222 <div class="doc_subsection">
223 <a name="klee">KLEE: Symbolic Execution and Automatic Test Case Generator</a>
226 <div class="doc_text">
228 The new LLVM <a href="http://klee.llvm.org/">KLEE project</a> is a symbolic
229 execution framework for programs in LLVM bitcode form. KLEE tries to
230 symbolically evaluate "all" paths through the application and records state
231 transitions that lead to fault states. This allows it to construct testcases
232 that lead to faults and can even be used to verify algorithms. For more
233 details, please see the <a
234 href="http://llvm.org/pubs/2008-12-OSDI-KLEE.html">OSDI 2008 paper</a> about
239 <!--=========================================================================-->
240 <div class="doc_subsection">
241 <a name="dragonegg">DragonEgg: GCC-4.5 as an LLVM frontend</a>
244 <div class="doc_text">
246 The goal of <a href="http://dragonegg.llvm.org/">DragonEgg</a> is to make
247 gcc-4.5 act like llvm-gcc without requiring any gcc modifications whatsoever.
248 DragonEgg is a shared library (llvm.so) that is loaded by gcc at runtime. It
249 uses the new gcc plugin architecture to disable the GCC optimizers and code
250 generators, and schedule the LLVM optimizers and code generators (or direct
251 output of LLVM IR) instead. Currently only Linux and Darwin are supported,
252 and only on x86-32 and x86-64. It should be easy to add additional unix-like
253 architectures and other processor families. Since gcc-4.5 has not yet been
254 released, neither has <a href="http://dragonegg.llvm.org/">DragonEgg</a>.
255 To build <a href="http://dragonegg.llvm.org/">DragonEgg</a> you will need to
256 check out the development versions of <a href="http://gcc.gnu.org/svn.html/">
257 gcc</a>, <a href="http://llvm.org/docs/GettingStarted.html#checkout">llvm</a>
258 and <a href="http://dragonegg.llvm.org/">DragonEgg</a> from their respective
259 subversion repositories.
265 <!--=========================================================================-->
266 <div class="doc_subsection">
267 <a name="mc">llvm-mc: Machine Code Toolkit</a>
270 <div class="doc_text">
272 The LLVM Machine Code (MC) Toolkit project is a (very early) effort to build
273 better tools for dealing with machine code, object file formats, etc. The idea
274 is to be able to generate most of the target specific details of assemblers and
275 disassemblers from existing LLVM target .td files (with suitable enhancements),
276 and to build infrastructure for reading and writing common object file formats.
277 One of the first deliverables is to build a full assembler and integrate it into
278 the compiler, which is predicted to substantially reduce compile time in some
282 <p>In the LLVM 2.6 timeframe, the MC framework has grown to the point where it
283 can reliably parse and pretty print (with some encoding information) a
284 darwin/x86 .s file successfully, and has the very early phases of a Mach-O
285 assembler in progress. Beyond the MC framework itself, major refactoring of the
286 LLVM code generator has started. The idea is to make the code generator reason
287 about the code it is producing in a much more semantic way, rather than a
288 textual way. For example, the code generator now uses MCSection objects to
289 represent section assignments, instead of text strings that print to .section
292 <p>MC is an early and ongoing project that will hopefully continue to lead to
293 many improvements in the code generator and build infrastructure useful for many
300 <!-- *********************************************************************** -->
301 <div class="doc_section">
302 <a name="externalproj">External Projects Using LLVM 2.6</a>
304 <!-- *********************************************************************** -->
306 <div class="doc_text">
308 <p>An exciting aspect of LLVM is that it is used as an enabling technology for
309 a lot of other language and tools projects. This section lists some of the
310 projects that have already been updated to work with LLVM 2.6.</p>
314 <!--=========================================================================-->
315 <div class="doc_subsection">
316 <a name="Rubinius">Rubinius</a>
319 <div class="doc_text">
320 <p><a href="http://github.com/evanphx/rubinius">Rubinius</a> is an environment
321 for running Ruby code which strives to write as much of the core class
322 implementation in Ruby as possible. Combined with a bytecode interpreting VM, it
323 uses LLVM to optimize and compile ruby code down to machine code. Techniques
324 such as type feedback, method inlining, and uncommon traps are all used to
325 remove dynamism from ruby execution and increase performance.</p>
327 <p>Since LLVM 2.5, Rubinius has made several major leaps forward, implementing
328 a counter based JIT, type feedback, and speculative method inlining.
333 <!--=========================================================================-->
334 <div class="doc_subsection">
335 <a name="macruby">MacRuby</a>
338 <div class="doc_text">
341 <a href="http://macruby.org">MacRuby</a> is an implementation of Ruby on top of
342 core Mac OS X technologies, such as the Objective-C common runtime and garbage
343 collector, and the CoreFoundation framework. It is principally developed by
344 Apple and aims at enabling the creation of full-fledged Mac OS X applications.
348 MacRuby uses LLVM for optimization passes, JIT and AOT compilation of Ruby
349 expressions. It also uses zero-cost DWARF exceptions to implement Ruby exception
355 <!--=========================================================================-->
356 <div class="doc_subsection">
357 <a name="pure">Pure</a>
360 <div class="doc_text">
362 <a href="http://pure-lang.googlecode.com/">Pure</a>
363 is an algebraic/functional programming language based on term rewriting.
364 Programs are collections of equations which are used to evaluate expressions in
365 a symbolic fashion. Pure offers dynamic typing, eager and lazy evaluation,
366 lexical closures, a hygienic macro system (also based on term rewriting),
367 built-in list and matrix support (including list and matrix comprehensions) and
368 an easy-to-use C interface. The interpreter uses LLVM as a backend to
369 JIT-compile Pure programs to fast native code.</p>
371 <p>Pure versions 0.31 and later have been tested and are known to work with
372 LLVM 2.6 (and continue to work with older LLVM releases >= 2.3 as well).
377 <!--=========================================================================-->
378 <div class="doc_subsection">
379 <a name="ldc">LLVM D Compiler</a>
382 <div class="doc_text">
384 <a href="http://www.dsource.org/projects/ldc">LDC</a> is an implementation of
385 the D Programming Language using the LLVM optimizer and code generator.
386 The LDC project works great with the LLVM 2.6 release. General improvements in
388 cycle have included new inline asm constraint handling, better debug info
389 support, general bug fixes, and better x86-64 support. This has allowed
390 some major improvements in LDC, getting it much closer to being as
391 fully featured as the original DMD compiler from DigitalMars.
395 <!--=========================================================================-->
396 <div class="doc_subsection">
397 <a name="RoadsendPHP">Roadsend PHP</a>
400 <div class="doc_text">
402 <a href="http://code.roadsend.com/rphp">Roadsend PHP</a> (rphp) is an open
403 source implementation of the PHP programming
404 language that uses LLVM for its optimizer, JIT, and static compiler. This is a
405 reimplementation of an earlier project that is now based on LLVM.</p>
408 <!--=========================================================================-->
409 <div class="doc_subsection">
410 <a name="UnladenSwallow">Unladen Swallow</a>
413 <div class="doc_text">
415 <a href="http://code.google.com/p/unladen-swallow/">Unladen Swallow</a> is a
416 branch of <a href="http://python.org/">Python</a> intended to be fully
417 compatible and significantly faster. It uses LLVM's optimization passes and JIT
421 <!--=========================================================================-->
422 <div class="doc_subsection">
423 <a name="llvm-lua">llvm-lua</a>
426 <div class="doc_text">
428 <a href="http://code.google.com/p/llvm-lua/">LLVM-Lua</a> uses LLVM to add JIT
429 and static compiling support to the Lua VM. Lua bytecode is analyzed to
430 remove type checks, then LLVM is used to compile the bytecode down to machine
436 <!-- *********************************************************************** -->
437 <div class="doc_section">
438 <a name="whatsnew">What's New in LLVM 2.6?</a>
440 <!-- *********************************************************************** -->
442 <div class="doc_text">
444 <p>This release includes a huge number of bug fixes, performance tweaks, and
445 minor improvements. Some of the major improvements and new features are listed
451 <!--=========================================================================-->
452 <div class="doc_subsection">
453 <a name="majorfeatures">Major New Features</a>
456 <div class="doc_text">
458 <p>LLVM 2.6 includes several major new capabilities:</p>
461 <li>New <a href="#compiler-rt">compiler-rt</a>, <A href="#klee">KLEE</a>,
462 and <a href="#mc">machine code toolkit</a> sub-projects.</li>
463 <li>Debug information now includes line numbers when optimizations are enabled.
464 This allows statistical sampling tools like oprofile and Shark to map
465 samples back to source lines.</li>
466 <li>LLVM now includes new experimental backends to support the MSP430, SystemZ,
467 and BlackFin architectures.</li>
468 <li>LLVM supports a new <a href="GoldPlugin.html">Gold Linker Plugin</a> which
469 enables support for <a href="LinkTimeOptimization.html">transparent
470 link-time optimization</a> on ELF targets when used with the Gold binutils
472 <li>LLVM now supports doing optimization and code generation on multiple
473 threads. Please see the <a href="ProgrammersManual.html#threading">LLVM
474 Programmer's Manual</a> for more information.</li>
475 <li>LLVM now has experimental support for <a
476 href="http://nondot.org/~sabre/LLVMNotes/EmbeddedMetadata.txt">embedded
477 metadata</a> in LLVM IR, though the implementation is not guaranteed to be
478 final and the .bc file format may change in future releases. Debug info
479 does not yet use this format in LLVM 2.6.</li>
484 <!--=========================================================================-->
485 <div class="doc_subsection">
486 <a name="coreimprovements">LLVM IR and Core Improvements</a>
489 <div class="doc_text">
490 <p>LLVM IR has several new features for better support of new targets and that
491 expose new optimization opportunities:</p>
494 <li>The <a href="LangRef.html#i_add">add</a>, <a
495 href="LangRef.html#i_sub">sub</a>, and <a href="LangRef.html#i_mul">mul</a>
496 instructions have been split into integer and floating point versions (like
497 divide and remainder), introducing new <a
498 href="LangRef.html#i_fadd">fadd</a>, <a href="LangRef.html#i_fsub">fsub</a>,
499 and <a href="LangRef.html#i_fmul">fmul</a> instructions.</li>
500 <li>The <a href="LangRef.html#i_add">add</a>, <a
501 href="LangRef.html#i_sub">sub</a>, and <a href="LangRef.html#i_mul">mul</a>
502 instructions now support optional "nsw" and "nuw" bits which indicate that
503 the operation is guaranteed to not overflow (in the signed or
504 unsigned case, respectively). This gives the optimizer more information and
505 can be used for things like C signed integer values, which are undefined on
507 <li>The <a href="LangRef.html#i_sdiv">sdiv</a> instruction now supports an
508 optional "exact" flag which indicates that the result of the division is
509 guaranteed to have a remainder of zero. This is useful for optimizing pointer
510 subtraction in C.</li>
511 <li>The <a href="LangRef.html#i_getelementptr">getelementptr</a> instruction now
512 supports arbitrary integer index values for array/pointer indices. This
513 allows for better code generation on 16-bit targets like PIC16.</li>
514 <li>The <a href="LangRef.html#i_getelementptr">getelementptr</a> instruction now
515 supports an "inbounds" optimization hint that tells the optimizer that the
516 pointer is guaranteed to be within its allocated object.</li>
517 <li>LLVM now support a series of new linkage types for global values which allow
518 for better optimization and new capabilities:
520 <li><a href="LangRef.html#linkage_linkonce">linkonce_odr</a> and
521 <a href="LangRef.html#linkage_weak">weak_odr</a> have the same linkage
522 semantics as the non-"odr" linkage types. The difference is that these
523 linkage types indicate that all definitions of the specified function
524 are guaranteed to have the same semantics. This allows inlining
525 templates functions in C++ but not inlining weak functions in C,
526 which previously both got the same linkage type.</li>
527 <li><a href="LangRef.html#linkage_available_externally">available_externally
528 </a> is a new linkage type that gives the optimizer visibility into the
529 definition of a function (allowing inlining and side effect analysis)
530 but that does not cause code to be generated. This allows better
531 optimization of "GNU inline" functions, extern templates, etc.</li>
532 <li><a href="LangRef.html#linkage_linker_private">linker_private</a> is a
533 new linkage type (which is only useful on Mac OS X) that is used for
534 some metadata generation and other obscure things.</li>
536 <li>Finally, target-specific intrinsics can now return multiple values, which
537 is useful for modeling target operations with multiple results.</li>
542 <!--=========================================================================-->
543 <div class="doc_subsection">
544 <a name="optimizer">Optimizer Improvements</a>
547 <div class="doc_text">
549 <p>In addition to a large array of minor performance tweaks and bug fixes, this
550 release includes a few major enhancements and additions to the optimizers:</p>
554 <li>The <a href="Passes.html#scalarrepl">Scalar Replacement of Aggregates</a>
555 pass has many improvements that allow it to better promote vector unions,
556 variables which are memset, and much more strange code that can happen to
557 do bitfield accesses to register operations. An interesting change is that
558 it now produces "unusual" integer sizes (like i1704) in some cases and lets
559 other optimizers clean things up.</li>
560 <li>The <a href="Passes.html#loop-reduce">Loop Strength Reduction</a> pass now
561 promotes small integer induction variables to 64-bit on 64-bit targets,
562 which provides a major performance boost for much numerical code. It also
563 promotes shorts to int on 32-bit hosts, etc. LSR now also analyzes pointer
564 expressions (e.g. getelementptrs), as well as integers.</li>
565 <li>The <a href="Passes.html#gvn">GVN</a> pass now eliminates partial
566 redundancies of loads in simple cases.</li>
567 <li>The <a href="Passes.html#inline">Inliner</a> now reuses stack space when
568 inlining similar arrays from multiple callees into one caller.</li>
569 <li>LLVM includes a new experimental Static Single Information (SSI)
570 construction pass.</li>
578 <!--=========================================================================-->
579 <div class="doc_subsection">
580 <a name="executionengine">Interpreter and JIT Improvements</a>
583 <div class="doc_text">
586 <li>LLVM has a new "EngineBuilder" class which makes it more obvious how to
587 set up and configure an ExecutionEngine (a JIT or interpreter).</li>
588 <li>The JIT now supports generating more than 16M of code.</li>
589 <li>When configured with --with-oprofile, the JIT can now inform oprofile about
590 JIT'd code, allowing oprofile to get line number and function name
591 information for JIT'd functions.</li>
592 <li>When "libffi" is available, the LLVM interpreter now uses it, which supports
593 calling almost arbitrary external (natively compiled) functions.</li>
594 <li>Clients of the JIT can now register a 'JITEventListener' object to receive
595 callbacks when the JIT emits or frees machine code. The OProfile support
596 uses this mechanism.</li>
601 <!--=========================================================================-->
602 <div class="doc_subsection">
603 <a name="codegen">Target Independent Code Generator Improvements</a>
606 <div class="doc_text">
608 <p>We have put a significant amount of work into the code generator
609 infrastructure, which allows us to implement more aggressive algorithms and make
614 <li>The <tt>llc -asm-verbose</tt> option (exposed from llvm-gcc and clang as
615 <tt>-fverbose-asm</tt>) now adds a lot of useful information in comments to
616 the generated .s file. This information includes location information (if
617 built with <tt>-g</tt>) and loop nest information.</li>
618 <li>The code generator now supports a new MachineVerifier pass which is useful
619 for finding bugs in targets and codegen passes.</li>
620 <li>The Machine LICM is now enabled by default. It hoists instructions out of
621 loops (such as constant pool loads, loads from read-only stubs, vector
622 constant synthesization code, etc) and is currently configured to only do so
623 when the hoisted operation can be rematerialized.</li>
624 <li>The Machine Sinking pass is now enabled by default. This pass moves
625 side-effect free operations down the CFG so that they are executed on fewer
626 paths through a function.</li>
627 <li>The code generator now performs "stack slot coloring" of register spills,
628 which allows spill slots to be reused. This leads to smaller stack frames
629 in cases where there are lots of register spills.</li>
630 <li>The register allocator has many improvements to take better advantage of
631 commutable operations, various spiller peephole optimizations, and can now
632 coalesce cross-register-class copies.</li>
633 <li>Tblgen now supports multiclass inheritance and a number of new string and
634 list operations like !(subst), !(foreach), !car, !cdr, !null, !if, !cast.
635 These make the .td files more expressive and allow more aggressive factoring
636 of duplication across instruction patterns.</li>
637 <li>Target-specific intrinsics can now be added without having to hack VMCore to
638 add them. This makes it easier to maintain out-of-tree targets.</li>
639 <li>The instruction selector is better at propagating information about values
640 (such as whether they are sign/zero extended etc) across basic block
642 <li>The SelectionDAG datastructure has new nodes for representing buildvector
643 and <a href="http://llvm.org/PR2957">vector shuffle</a> operations. This
644 makes operations and pattern matching more efficient and easier to get
646 <li>The Prolog/Epilog Insertion Pass now has experimental support for performing
647 the "shrink wrapping" optimization, which moves spills and reloads around in
648 the CFG to avoid doing saves on paths that don't need them.</li>
649 <li>LLVM includes new experimental support for writing ELF .o files directly
650 from the compiler. It works well for many simple C testcases, but doesn't
651 support exception handling, debug info, inline assembly, etc.</li>
652 <li>Targets can now specify register allocation hints through
653 MachineRegisterInfo::setRegAllocationHint. A regalloc hint consists of hint
654 type and physical register number. A hint type of zero specifies a register
655 allocation preference. Other hint type values are target specific which are
656 resolved by TargetRegisterInfo::ResolveRegAllocHint. An example is the ARM
657 target which uses register hints to request that the register allocator
658 provide an even / odd register pair to two virtual registers.</li>
662 <!--=========================================================================-->
663 <div class="doc_subsection">
664 <a name="x86">X86-32 and X86-64 Target Improvements</a>
667 <div class="doc_text">
668 <p>New features of the X86 target include:
673 <li>SSE 4.2 builtins are now supported.</li>
674 <li>GCC-compatible soft float modes are now supported, which are typically used
676 <li>X86-64 now models implicit zero extensions better, which allows the code
677 generator to remove a lot of redundant zexts. It also models the 8-bit "H"
678 registers as sugregs, which allows them to be used in some tricky
680 <li>X86-64 now supports the "local exec" and "initial exec" thread local storage
682 <li>The vector forms of the <a href="LangRef.html#i_icmp">icmp</a> and <a
683 href="LangRef.html#i_fcmp">fcmp</a> instructions now select to efficient
685 <li>The X86 backend has preliminary support for <a
686 href="CodeGenerator.html#x86_memory">mapping address spaces to segment
687 register references</a>. This allows you to write GS or FS relative memory
688 accesses directly in LLVM IR for cases where you know exactly what you're
689 doing (such as in an OS kernel). There are some known problems with this
690 support, but it works in simple cases.</li>
691 <li>The X86 code generator has been refactored to move all global variable
692 reference logic to one place
693 (<tt>X86Subtarget::ClassifyGlobalReference</tt>) which
694 makes it easier to reason about.</li>
700 <!--=========================================================================-->
701 <div class="doc_subsection">
702 <a name="pic16">PIC16 Target Improvements</a>
705 <div class="doc_text">
706 <p>New features of the PIC16 target include:
710 <li>Support for floating-point, indirect function calls, and
711 passing/returning aggregate types to functions.
712 <li>The code generator is able to generate debug info into output COFF files.
713 <li>Support for placing an object into a specific section or at a specific
714 address in memory.</li>
717 <p>Things not yet supported:</p>
720 <li>Variable arguments.</li>
721 <li>Interrupts/programs.</li>
726 <!--=========================================================================-->
727 <div class="doc_subsection">
728 <a name="ARM">ARM Target Improvements</a>
731 <div class="doc_text">
732 <p>New features of the ARM target include:
737 <li>Preliminary support for processors, such as the Cortex-A8 and Cortex-A9,
738 that implement version v7-A of the ARM architecture. The ARM backend now
739 supports both the Thumb2 and Advanced SIMD (Neon) instruction sets.</li>
741 <li>The AAPCS-VFP "hard float" calling conventions are also supported with the
742 <tt>-float-abi=hard</tt> flag.</li>
744 <li>The ARM calling convention code is now tblgen generated instead of resorting
749 <p>These features are still somewhat experimental
750 and subject to change. The Neon intrinsics, in particular, may change in future
751 releases of LLVM.</p>
757 <!--=========================================================================-->
758 <div class="doc_subsection">
759 <a name="OtherTarget">Other Target Specific Improvements</a>
762 <div class="doc_text">
763 <p>New features of other targets include:
767 <li>Mips now supports O32 Calling Convention.</li>
768 <li>Many improvements to the 32-bit PowerPC SVR4 ABI (used on powerpc-linux)
769 support, lots of bugs fixed.</li>
770 <li>Added support for the 64-bit PowerPC SVR4 ABI (used on powerpc64-linux).
771 Needs more testing.</li>
776 <!--=========================================================================-->
777 <div class="doc_subsection">
778 <a name="newapis">New Useful APIs</a>
781 <div class="doc_text">
783 <p>This release includes a number of new APIs that are used internally, which
784 may also be useful for external clients.
788 <li>New <a href="http://llvm.org/doxygen/PrettyStackTrace_8h-source.html">
789 PrettyStackTrace classes</a> allows crashes of llvm tools (and applications
790 that integrate them) to provide more detailed indication of what the
791 compiler was doing at the time of the crash (e.g. running a pass).
792 At the top level for each LLVM tool, it includes the command line arguments.
794 <li>New <a href="http://llvm.org/doxygen/StringRef_8h-source.html">StringRef</a>
795 and <a href="http://llvm.org/doxygen/Twine_8h-source.html">Twine</a> classes
796 make operations on character ranges and
797 string concatenation to be more efficient. StringRef is just a <tt>const
798 char*</tt> with a length, Twine is a light-weight rope.</li>
799 <li>LLVM has new WeakVH, AssertingVH, and CallbackVH classes, which make it
800 easier to write LLVM IR transformations. WeakVH is automatically drops to
801 null when the referenced Value is deleted, and is updated across a
802 replaceAllUsesWith operation. AssertingVH aborts the program if the
803 referenced value is destroyed while it is being referenced. CallbackVH is
804 a customizable class for handling value references. See <a
805 href="http://llvm.org/doxygen/ValueHandle_8h-source.html">ValueHandle.h</a>
806 for more information.</li>
807 <li>The new '<a href="http://llvm.org/doxygen/Triple_8h-source.html">Triple
808 </a>' class centralizes a lot of logic that reasons about target
810 <li>The new '<a href="http://llvm.org/doxygen/ErrorHandling_8h-source.html">
811 llvm_report_error()</a>' set of APIs allows tools to embed the LLVM
812 optimizer and backend and recover from previously unrecoverable errors.</li>
813 <li>LLVM has new abstractions for <a
814 href="http://llvm.org/doxygen/Atomic_8h-source.html">atomic operations</a>
815 and <a href="http://llvm.org/doxygen/RWMutex_8h-source.html">reader/writer
817 <li>LLVM has new <a href="http://llvm.org/doxygen/SourceMgr_8h-source.html">
818 SourceMgr and SMLoc classes</a> which implement caret diagnostics and basic
819 include stack processing for simple parsers. It is used by tablegen,
820 llvm-mc, the .ll parser and FileCheck.</li>
826 <!--=========================================================================-->
827 <div class="doc_subsection">
828 <a name="otherimprovements">Other Improvements and New Features</a>
831 <div class="doc_text">
832 <p>Other miscellaneous features include:</p>
835 <li>LLVM now includes a new internal '<a
836 href="http://llvm.org/cmds/FileCheck.html">FileCheck</a>' tool which allows
837 writing much more accurate regression tests that run faster. Please see the
838 <a href="TestingGuide.html#FileCheck">FileCheck section of the Testing
839 Guide</a> for more information.</li>
840 <li>LLVM profile information support has been significantly improved to produce
841 correct use counts, and has support for edge profiling with reduced runtime
842 overhead. Combined, the generated profile information is both more correct and
843 imposes about half as much overhead (2.6. from 12% to 6% overhead on SPEC
845 <li>The C bindings (in the llvm/include/llvm-c directory) include many newly
847 <li>LLVM 2.6 includes a brand new experimental LLVM bindings to the Ada2005
848 programming language.</li>
850 <li>The LLVMC driver has several new features:
852 <li>Dynamic plugins now work on Windows.</li>
853 <li>New option property: init. Makes possible to provide default values for
854 options defined in plugins (interface to cl::init).</li>
855 <li>New example: Skeleton, shows how to create a standalone LLVMC-based
857 <li>New example: mcc16, a driver for the PIC16 toolchain.</li>
866 <!--=========================================================================-->
867 <div class="doc_subsection">
868 <a name="changes">Major Changes and Removed Features</a>
871 <div class="doc_text">
873 <p>If you're already an LLVM user or developer with out-of-tree changes based
874 on LLVM 2.5, this section lists some "gotchas" that you may run into upgrading
875 from the previous release.</p>
878 <li>The Itanium (IA64) backend has been removed. It was not actively supported
879 and had bitrotted.</li>
880 <li>The BigBlock register allocator has been removed, it had also bitrotted.</li>
881 <li>The C Backend (-march=c) is no longer considered part of the LLVM release
882 criteria. We still want it to work, but no one is maintaining it and it lacks
883 support for arbitrary precision integers and other important IR features.</li>
885 <li>All LLVM tools now default to overwriting their output file, behaving more
886 like standard unix tools. Previously, this only happened with the '-f'
888 <li>LLVM build now builds all libraries as .a files instead of some
889 libraries as relinked .o files. This requires some APIs like
890 InitializeAllTargets.h.
895 <p>In addition, many APIs have changed in this release. Some of the major LLVM
899 <li>All uses of hash_set and hash_map have been removed from the LLVM tree and
900 the wrapper headers have been removed.</li>
901 <li>The llvm/Streams.h and DOUT member of Debug.h have been removed. The
902 llvm::Ostream class has been completely removed and replaced with uses of
904 <li>LLVM's global uniquing tables for <tt>Type</tt>s and <tt>Constant</tt>s have
905 been privatized into members of an <tt>LLVMContext</tt>. A number of APIs
906 now take an <tt>LLVMContext</tt> as a parameter. To smooth the transition
907 for clients that will only ever use a single context, the new
908 <tt>getGlobalContext()</tt> API can be used to access a default global
909 context which can be passed in any and all cases where a context is
911 <li>The <tt>getABITypeSize</tt> methods are now called <tt>getAllocSize</tt>.</li>
912 <li>The <tt>Add</tt>, <tt>Sub</tt>, and <tt>Mul</tt> operators are no longer
913 overloaded for floating-point types. Floating-point addition, subtraction,
914 and multiplication are now represented with new operators <tt>FAdd</tt>,
915 <tt>FSub</tt>, and <tt>FMul</tt>. In the <tt>IRBuilder</tt> API,
916 <tt>CreateAdd</tt>, <tt>CreateSub</tt>, <tt>CreateMul</tt>, and
917 <tt>CreateNeg</tt> should only be used for integer arithmetic now;
918 <tt>CreateFAdd</tt>, <tt>CreateFSub</tt>, <tt>CreateFMul</tt>, and
919 <tt>CreateFNeg</tt> should now be used for floating-point arithmetic.</li>
920 <li>The DynamicLibrary class can no longer be constructed, its functionality has
921 moved to static member functions.</li>
922 <li><tt>raw_fd_ostream</tt>'s constructor for opening a given filename now
923 takes an extra <tt>Force</tt> argument. If <tt>Force</tt> is set to
924 <tt>false</tt>, an error will be reported if a file with the given name
925 already exists. If <tt>Force</tt> is set to <tt>true</tt>, the file will
926 be silently truncated (which is the behavior before this flag was
928 <li><tt>SCEVHandle</tt> no longer exists, because reference counting is no
929 longer done for <tt>SCEV*</tt> objects, instead <tt>const SCEV*</tt> should be
932 <li>Many APIs, notably <tt>llvm::Value</tt>, now use the <tt>StringRef</tt>
933 and <tt>Twine</tt> classes instead of passing <tt>const char*</tt>
934 or <tt>std::string</tt>, as described in
935 the <a href="ProgrammersManual.html#string_apis">Programmer's Manual</a>. Most
936 clients should be unaffected by this transition, unless they are used to
937 <tt>Value::getName()</tt> returning a string. Here are some tips on updating to
940 <li><tt>getNameStr()</tt> is still available, and matches the old
941 behavior. Replacing <tt>getName()</tt> calls with this is an safe option,
942 although more efficient alternatives are now possible.</li>
944 <li>If you were just relying on <tt>getName()</tt> being able to be sent to
945 a <tt>std::ostream</tt>, consider migrating
946 to <tt>llvm::raw_ostream</tt>.</li>
948 <li>If you were using <tt>getName().c_str()</tt> to get a <tt>const
949 char*</tt> pointer to the name, you can use <tt>getName().data()</tt>.
950 Note that this string (as before), may not be the entire name if the
951 name contains embedded null characters.</li>
953 <li>If you were using operator plus on the result of <tt>getName()</tt> and
954 treating the result as an <tt>std::string</tt>, you can either
955 use <tt>Twine::str</tt> to get the result as an <tt>std::string</tt>, or
956 could move to a <tt>Twine</tt> based design.</li>
958 <li><tt>isName()</tt> should be replaced with comparison
959 against <tt>getName()</tt> (this is now efficient).
963 <li>The registration interfaces for backend Targets has changed (what was
964 previously TargetMachineRegistry). For backend authors, see the <a
965 href="WritingAnLLVMBackend.html#TargetRegistration">Writing An LLVM Backend</a>
966 guide. For clients, the notable API changes are:
968 <li><tt>TargetMachineRegistry</tt> has been renamed
969 to <tt>TargetRegistry</tt>.</li>
971 <li>Clients should move to using the <tt>TargetRegistry::lookupTarget()</tt>
972 function to find targets.</li>
981 <!-- *********************************************************************** -->
982 <div class="doc_section">
983 <a name="portability">Portability and Supported Platforms</a>
985 <!-- *********************************************************************** -->
987 <div class="doc_text">
989 <p>LLVM is known to work on the following platforms:</p>
992 <li>Intel and AMD machines (IA32, X86-64, AMD64, EMT-64) running Red Hat
993 Linux, Fedora Core, FreeBSD and AuroraUX (and probably other unix-like
995 <li>PowerPC and X86-based Mac OS X systems, running 10.3 and above in 32-bit
996 and 64-bit modes.</li>
997 <li>Intel and AMD machines running on Win32 using MinGW libraries (native).</li>
998 <li>Intel and AMD machines running on Win32 with the Cygwin libraries (limited
999 support is available for native builds with Visual C++).</li>
1000 <li>Sun UltraSPARC workstations running Solaris 10.</li>
1001 <li>Alpha-based machines running Debian GNU/Linux.</li>
1004 <p>The core LLVM infrastructure uses GNU autoconf to adapt itself
1005 to the machine and operating system on which it is built. However, minor
1006 porting may be required to get LLVM to work on new platforms. We welcome your
1007 portability patches and reports of successful builds or error messages.</p>
1011 <!-- *********************************************************************** -->
1012 <div class="doc_section">
1013 <a name="knownproblems">Known Problems</a>
1015 <!-- *********************************************************************** -->
1017 <div class="doc_text">
1019 <p>This section contains significant known problems with the LLVM system,
1020 listed by component. If you run into a problem, please check the <a
1021 href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
1022 there isn't already one.</p>
1025 <li>LLVM will not correctly compile on Solaris and/or OpenSolaris
1026 using the stock GCC 3.x.x series 'out the box',
1027 See: <a href="#brokengcc">Broken versions of GCC and other tools</a>.
1028 However, A <a href="http://pkg.auroraux.org/GCC">Modern GCC Build</a>
1029 for x86/x64 has been made available from the third party AuroraUX Project
1030 that has been meticulously tested for bootstrapping LLVM & Clang.</li>
1035 <!-- ======================================================================= -->
1036 <div class="doc_subsection">
1037 <a name="experimental">Experimental features included with this release</a>
1040 <div class="doc_text">
1042 <p>The following components of this LLVM release are either untested, known to
1043 be broken or unreliable, or are in early development. These components should
1044 not be relied on, and bugs should not be filed against them, but they may be
1045 useful to some people. In particular, if you would like to work on one of these
1046 components, please contact us on the <a
1047 href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev list</a>.</p>
1050 <li>The MSIL, Alpha, SPU, MIPS, PIC16, Blackfin, MSP430 and SystemZ backends are
1052 <li>The <tt>llc</tt> "<tt>-filetype=asm</tt>" (the default) is the only
1053 supported value for this option. The ELF writer is experimental.</li>
1054 <li>The implementation of Andersen's Alias Analysis has many known bugs.</li>
1059 <!-- ======================================================================= -->
1060 <div class="doc_subsection">
1061 <a name="x86-be">Known problems with the X86 back-end</a>
1064 <div class="doc_text">
1067 <li>The X86 backend does not yet support
1068 all <a href="http://llvm.org/PR879">inline assembly that uses the X86
1069 floating point stack</a>. It supports the 'f' and 't' constraints, but not
1071 <li>The X86 backend generates inefficient floating point code when configured
1072 to generate code for systems that don't have SSE2.</li>
1073 <li>Win64 code generation wasn't widely tested. Everything should work, but we
1074 expect small issues to happen. Also, llvm-gcc cannot build the mingw64
1075 runtime currently due
1076 to <a href="http://llvm.org/PR2255">several</a>
1077 <a href="http://llvm.org/PR2257">bugs</a> and due to lack of support for
1079 'u' inline assembly constraint and for X87 floating point inline assembly.</li>
1080 <li>The X86-64 backend does not yet support the LLVM IR instruction
1081 <tt>va_arg</tt>. Currently, the llvm-gcc and front-ends support variadic
1082 argument constructs on X86-64 by lowering them manually.</li>
1087 <!-- ======================================================================= -->
1088 <div class="doc_subsection">
1089 <a name="ppc-be">Known problems with the PowerPC back-end</a>
1092 <div class="doc_text">
1095 <li>The Linux PPC32/ABI support needs testing for the interpreter and static
1096 compilation, and lacks support for debug information.</li>
1101 <!-- ======================================================================= -->
1102 <div class="doc_subsection">
1103 <a name="arm-be">Known problems with the ARM back-end</a>
1106 <div class="doc_text">
1109 <li>Support for the Advanced SIMD (Neon) instruction set is still incomplete
1110 and not well tested. Some features may not work at all, and the code quality
1111 may be poor in some cases.</li>
1112 <li>Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6
1113 processors, thumb programs can crash or produce wrong
1114 results (<a href="http://llvm.org/PR1388">PR1388</a>).</li>
1115 <li>Compilation for ARM Linux OABI (old ABI) is supported but not fully tested.
1121 <!-- ======================================================================= -->
1122 <div class="doc_subsection">
1123 <a name="sparc-be">Known problems with the SPARC back-end</a>
1126 <div class="doc_text">
1129 <li>The SPARC backend only supports the 32-bit SPARC ABI (-m32); it does not
1130 support the 64-bit SPARC ABI (-m64).</li>
1135 <!-- ======================================================================= -->
1136 <div class="doc_subsection">
1137 <a name="mips-be">Known problems with the MIPS back-end</a>
1140 <div class="doc_text">
1143 <li>64-bit MIPS targets are not supported yet.</li>
1148 <!-- ======================================================================= -->
1149 <div class="doc_subsection">
1150 <a name="alpha-be">Known problems with the Alpha back-end</a>
1153 <div class="doc_text">
1157 <li>On 21164s, some rare FP arithmetic sequences which may trap do not have the
1158 appropriate nops inserted to ensure restartability.</li>
1163 <!-- ======================================================================= -->
1164 <div class="doc_subsection">
1165 <a name="c-be">Known problems with the C back-end</a>
1168 <div class="doc_text">
1171 <li><a href="http://llvm.org/PR802">The C backend has only basic support for
1172 inline assembly code</a>.</li>
1173 <li><a href="http://llvm.org/PR1658">The C backend violates the ABI of common
1174 C++ programs</a>, preventing intermixing between C++ compiled by the CBE and
1175 C++ code compiled with <tt>llc</tt> or native compilers.</li>
1176 <li>The C backend does not support all exception handling constructs.</li>
1177 <li>The C backend does not support arbitrary precision integers.</li>
1183 <!-- ======================================================================= -->
1184 <div class="doc_subsection">
1185 <a name="c-fe">Known problems with the llvm-gcc C front-end</a>
1188 <div class="doc_text">
1190 <p>The only major language feature of GCC not supported by llvm-gcc is
1191 the <tt>__builtin_apply</tt> family of builtins. However, some extensions
1192 are only supported on some targets. For example, trampolines are only
1193 supported on some targets (these are used when you take the address of a
1194 nested function).</p>
1196 <p>If you run into GCC extensions which are not supported, please let us know.
1201 <!-- ======================================================================= -->
1202 <div class="doc_subsection">
1203 <a name="c++-fe">Known problems with the llvm-gcc C++ front-end</a>
1206 <div class="doc_text">
1208 <p>The C++ front-end is considered to be fully
1209 tested and works for a number of non-trivial programs, including LLVM
1210 itself, Qt, Mozilla, etc.</p>
1213 <li>Exception handling works well on the X86 and PowerPC targets. Currently
1214 only Linux and Darwin targets are supported (both 32 and 64 bit).</li>
1219 <!-- ======================================================================= -->
1220 <div class="doc_subsection">
1221 <a name="fortran-fe">Known problems with the llvm-gcc Fortran front-end</a>
1224 <div class="doc_text">
1226 <li>Fortran support generally works, but there are still several unresolved bugs
1227 in Bugzilla. Please see the tools/gfortran component for details.</li>
1231 <!-- ======================================================================= -->
1232 <div class="doc_subsection">
1233 <a name="ada-fe">Known problems with the llvm-gcc Ada front-end</a>
1236 <div class="doc_text">
1237 The llvm-gcc 4.2 Ada compiler works fairly well; however, this is not a mature
1238 technology, and problems should be expected.
1240 <li>The Ada front-end currently only builds on X86-32. This is mainly due
1241 to lack of trampoline support (pointers to nested functions) on other platforms.
1242 However, it <a href="http://llvm.org/PR2006">also fails to build on X86-64</a>
1243 which does support trampolines.</li>
1244 <li>The Ada front-end <a href="http://llvm.org/PR2007">fails to bootstrap</a>.
1245 This is due to lack of LLVM support for <tt>setjmp</tt>/<tt>longjmp</tt> style
1246 exception handling, which is used internally by the compiler.
1247 Workaround: configure with --disable-bootstrap.</li>
1248 <li>The c380004, <a href="http://llvm.org/PR2010">c393010</a>
1249 and <a href="http://llvm.org/PR2421">cxg2021</a> ACATS tests fail
1250 (c380004 also fails with gcc-4.2 mainline).
1251 If the compiler is built with checks disabled then <a href="http://llvm.org/PR2010">c393010</a>
1252 causes the compiler to go into an infinite loop, using up all system memory.</li>
1253 <li>Some GCC specific Ada tests continue to crash the compiler.</li>
1254 <li>The -E binder option (exception backtraces)
1255 <a href="http://llvm.org/PR1982">does not work</a> and will result in programs
1256 crashing if an exception is raised. Workaround: do not use -E.</li>
1257 <li>Only discrete types <a href="http://llvm.org/PR1981">are allowed to start
1258 or finish at a non-byte offset</a> in a record. Workaround: do not pack records
1259 or use representation clauses that result in a field of a non-discrete type
1260 starting or finishing in the middle of a byte.</li>
1261 <li>The <tt>lli</tt> interpreter <a href="http://llvm.org/PR2009">considers
1262 'main' as generated by the Ada binder to be invalid</a>.
1263 Workaround: hand edit the file to use pointers for <tt>argv</tt> and
1264 <tt>envp</tt> rather than integers.</li>
1265 <li>The <tt>-fstack-check</tt> option <a href="http://llvm.org/PR2008">is
1270 <!-- ======================================================================= -->
1271 <div class="doc_subsection">
1272 <a name="ocaml-bindingse">Known problems with the O'Caml bindings</a>
1275 <div class="doc_text">
1277 <p>The Llvm.Linkage module is broken, and has incorrect values. Only
1278 Llvm.Linkage.External, Llvm.Linkage.Available_externally, and
1279 Llvm.Linkage.Link_once will be correct. If you need any of the other linkage
1280 modes, you'll have to write an external C library in order to expose the
1281 functionality. This has been fixed in the trunk.</p>
1284 <!-- *********************************************************************** -->
1285 <div class="doc_section">
1286 <a name="additionalinfo">Additional Information</a>
1288 <!-- *********************************************************************** -->
1290 <div class="doc_text">
1292 <p>A wide variety of additional information is available on the <a
1293 href="http://llvm.org">LLVM web page</a>, in particular in the <a
1294 href="http://llvm.org/docs/">documentation</a> section. The web page also
1295 contains versions of the API documentation which is up-to-date with the
1296 Subversion version of the source code.
1297 You can access versions of these documents specific to this release by going
1298 into the "<tt>llvm/doc/</tt>" directory in the LLVM tree.</p>
1300 <p>If you have any questions or comments about LLVM, please feel free to contact
1301 us via the <a href="http://llvm.org/docs/#maillist"> mailing
1306 <!-- *********************************************************************** -->
1310 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1311 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1312 <a href="http://validator.w3.org/check/referer"><img
1313 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1315 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
1316 Last modified: $Date$