1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <link rel="stylesheet" href="llvm.css" type="text/css">
8 <title>LLVM 2.8 Release Notes</title>
12 <div class="doc_title">LLVM 2.8 Release Notes</div>
14 <img align=right src="http://llvm.org/img/DragonSmall.png"
15 width="136" height="136" alt="LLVM Dragon Logo">
18 <li><a href="#intro">Introduction</a></li>
19 <li><a href="#subproj">Sub-project Status Update</a></li>
20 <li><a href="#externalproj">External Projects Using LLVM 2.8</a></li>
21 <li><a href="#whatsnew">What's New in LLVM 2.8?</a></li>
22 <li><a href="GettingStarted.html">Installation Instructions</a></li>
23 <li><a href="#knownproblems">Known Problems</a></li>
24 <li><a href="#additionalinfo">Additional Information</a></li>
27 <div class="doc_author">
28 <p>Written by the <a href="http://llvm.org">LLVM Team</a></p>
32 <h1 style="color:red">These are in-progress notes for the upcoming LLVM 2.8
35 <a href="http://llvm.org/releases/2.7/docs/ReleaseNotes.html">LLVM 2.7
36 Release Notes</a>.</h1>
39 <!-- *********************************************************************** -->
40 <div class="doc_section">
41 <a name="intro">Introduction</a>
43 <!-- *********************************************************************** -->
45 <div class="doc_text">
47 <p>This document contains the release notes for the LLVM Compiler
48 Infrastructure, release 2.8. Here we describe the status of LLVM, including
49 major improvements from the previous release and significant known problems.
50 All LLVM releases may be downloaded from the <a
51 href="http://llvm.org/releases/">LLVM releases web site</a>.</p>
53 <p>For more information about LLVM, including information about the latest
54 release, please check out the <a href="http://llvm.org/">main LLVM
55 web site</a>. If you have questions or comments, the <a
56 href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM Developer's
57 Mailing List</a> is a good place to send them.</p>
59 <p>Note that if you are reading this file from a Subversion checkout or the
60 main LLVM web page, this document applies to the <i>next</i> release, not the
61 current one. To see the release notes for a specific release, please see the
62 <a href="http://llvm.org/releases/">releases page</a>.</p>
69 include/llvm/Analysis/LiveValues.h => Dan
70 lib/Transforms/IPO/MergeFunctions.cpp => consider for 2.8.
75 <!-- Features that need text if they're finished for 2.9:
78 loop dependence analysis
80 CorrelatedValuePropagation
83 <!-- Announcement, lldb, libc++ -->
86 <!-- *********************************************************************** -->
87 <div class="doc_section">
88 <a name="subproj">Sub-project Status Update</a>
90 <!-- *********************************************************************** -->
92 <div class="doc_text">
94 The LLVM 2.8 distribution currently consists of code from the core LLVM
95 repository (which roughly includes the LLVM optimizers, code generators
96 and supporting tools), the Clang repository and the llvm-gcc repository. In
97 addition to this code, the LLVM Project includes other sub-projects that are in
98 development. Here we include updates on these subprojects.
104 <!--=========================================================================-->
105 <div class="doc_subsection">
106 <a name="clang">Clang: C/C++/Objective-C Frontend Toolkit</a>
109 <div class="doc_text">
111 <p><a href="http://clang.llvm.org/">Clang</a> is an LLVM front end for the C,
112 C++, and Objective-C languages. Clang aims to provide a better user experience
113 through expressive diagnostics, a high level of conformance to language
114 standards, fast compilation, and low memory use. Like LLVM, Clang provides a
115 modular, library-based architecture that makes it suitable for creating or
116 integrating with other development tools. Clang is considered a
117 production-quality compiler for C, Objective-C, C++ and Objective-C++ on x86
118 (32- and 64-bit), and for darwin-arm targets.</p>
120 <p>In the LLVM 2.8 time-frame, the Clang team has made many improvements:</p>
123 <li>Surely these guys have done something</li>
124 <li>X86-64 abi improvements? Did they make it in?</li>
128 <!--=========================================================================-->
129 <div class="doc_subsection">
130 <a name="clangsa">Clang Static Analyzer</a>
133 <div class="doc_text">
135 <p>The <a href="http://clang-analyzer.llvm.org/">Clang Static Analyzer</a>
136 project is an effort to use static source code analysis techniques to
137 automatically find bugs in C and Objective-C programs (and hopefully <a
138 href="http://clang-analyzer.llvm.org/dev_cxx.html">C++ in the
139 future</a>!). The tool is very good at finding bugs that occur on specific
140 paths through code, such as on error conditions.</p>
142 <p>The LLVM 2.8 release fixes a number of bugs and slightly improves precision
143 over 2.7, but there are no major new features in the release.
148 <!--=========================================================================-->
149 <div class="doc_subsection">
150 <a name="dragonegg">DragonEgg: llvm-gcc ported to gcc-4.5</a>
153 <div class="doc_text">
155 <a href="http://dragonegg.llvm.org/">DragonEgg</a> is a port of llvm-gcc to
156 gcc-4.5. Unlike llvm-gcc, dragonegg in theory does not require any gcc-4.5
157 modifications whatsoever (currently one small patch is needed) thanks to the
158 new <a href="http://gcc.gnu.org/wiki/plugins">gcc plugin architecture</a>.
159 DragonEgg is a gcc plugin that makes gcc-4.5 use the LLVM optimizers and code
160 generators instead of gcc's, just like with llvm-gcc.
164 DragonEgg is still a work in progress, but it is able to compile a lot of code,
165 for example all of gcc, LLVM and clang. Currently Ada, C, C++ and Fortran work
166 well, while all other languages either don't work at all or only work poorly.
167 For the moment only the x86-32 and x86-64 targets are supported, and only on
168 linux and darwin (darwin may need additional gcc patches).
172 The 2.8 release has the following notable changes:
174 <li>The plugin loads faster due to exporting fewer symbols.</li>
175 <li>Additional vector operations such as addps256 are now supported.</li>
176 <li>Ada global variables with no initial value are no longer zero initialized,
177 resulting in better optimization.</li>
178 <li>The '-fplugin-arg-dragonegg-enable-gcc-optzns' flag now runs all gcc
179 optimizers, rather than just a handful.</li>
180 <li>Fortran programs using common variables now link correctly.</li>
181 <li>GNU OMP constructs no longer crash the compiler.</li>
187 <!--=========================================================================-->
188 <div class="doc_subsection">
189 <a name="vmkit">VMKit: JVM/CLI Virtual Machine Implementation</a>
192 <div class="doc_text">
194 The <a href="http://vmkit.llvm.org/">VMKit project</a> is an implementation of
195 a Java Virtual Machine (Java VM or JVM) that uses LLVM for static and
196 just-in-time compilation. As of LLVM 2.8, VMKit now supports copying garbage
197 collectors, and can be configured to use MMTk's copy mark-sweep garbage
198 collector. In LLVM 2.8, the VMKit .NET VM is no longer being maintained.
202 <!--=========================================================================-->
203 <div class="doc_subsection">
204 <a name="compiler-rt">compiler-rt: Compiler Runtime Library</a>
207 <div class="doc_text">
209 The new LLVM <a href="http://compiler-rt.llvm.org/">compiler-rt project</a>
210 is a simple library that provides an implementation of the low-level
211 target-specific hooks required by code generation and other runtime components.
212 For example, when compiling for a 32-bit target, converting a double to a 64-bit
213 unsigned integer is compiled into a runtime call to the "__fixunsdfdi"
214 function. The compiler-rt library provides highly optimized implementations of
215 this and other low-level routines (some are 3x faster than the equivalent
216 libgcc routines).</p>
219 All of the code in the compiler-rt project is available under the standard LLVM
220 License, a "BSD-style" license. New in LLVM 2.8, compiler_rt now supports
221 soft floating point (for targets that don't have a real floating point unit),
222 and includes an extensive testsuite for the "blocks" language feature and the
223 blocks runtime included in compiler_rt.</p>
227 <!--=========================================================================-->
228 <div class="doc_subsection">
229 <a name="lldb">LLDB: Low Level Debugger</a>
232 <div class="doc_text">
234 <a href="http://lldb.llvm.org/">LLDB</a> is a brand new member of the LLVM
235 umbrella of projects. LLDB is a next generation, high-performance debugger. It
236 is built as a set of reusable components which highly leverage existing
237 libraries in the larger LLVM Project, such as the Clang expression parser, the
238 LLVM disassembler and the LLVM JIT.</p>
241 LLDB is in early development and not included as part of the LLVM 2.8 release,
242 but is mature enough to support basic debugging scenarios on Mac OS X in C,
243 Objective-C and C++. We'd really like help extending and expanding LLDB to
244 support new platforms, new languages, new architectures, and new features.
249 <!--=========================================================================-->
250 <div class="doc_subsection">
251 <a name="libc++">libc++: C++ Standard Library</a>
254 <div class="doc_text">
256 <a href="http://libc++.llvm.org/">libc++</a> is another new member of the LLVM
257 family. It is an implementation of the C++ standard library, written from the
258 ground up to specifically target the forthcoming C++'0X standard and focus on
259 delivering great performance.</p>
262 As of the LLVM 2.8 release, libc++ is virtually feature complete, but would
263 benefit from more testing and better integration with Clang++. It is also
264 looking forward to the C++ committee finalizing the C++'0x standard.
270 <!-- *********************************************************************** -->
271 <div class="doc_section">
272 <a name="externalproj">External Open Source Projects Using LLVM 2.8</a>
274 <!-- *********************************************************************** -->
276 <div class="doc_text">
278 <p>An exciting aspect of LLVM is that it is used as an enabling technology for
279 a lot of other language and tools projects. This section lists some of the
280 projects that have already been updated to work with LLVM 2.8.</p>
283 <!--=========================================================================-->
284 <div class="doc_subsection">
285 <a name="tce">TTA-based Codesign Environment (TCE)</a>
288 <div class="doc_text">
290 <a href="http://tce.cs.tut.fi/">TCE</a> is a toolset for designing
291 application-specific processors (ASP) based on the Transport triggered
292 architecture (TTA). The toolset provides a complete co-design flow from C/C++
293 programs down to synthesizable VHDL and parallel program binaries. Processor
294 customization points include the register files, function units, supported
295 operations, and the interconnection network.</p>
297 <p>TCE uses llvm-gcc/Clang and LLVM for C/C++ language support, target
298 independent optimizations and also for parts of code generation. It generates
299 new LLVM-based code generators "on the fly" for the designed TTA processors and
300 loads them in to the compiler backend as runtime libraries to avoid per-target
301 recompilation of larger parts of the compiler chain.</p>
305 <!--=========================================================================-->
306 <div class="doc_subsection">
307 <a name="Horizon">Horizon Bytecode Compiler</a>
310 <div class="doc_text">
312 <a href="http://www.quokforge.org/projects/horizon">Horizon</a> is a bytecode
313 language and compiler written on top of LLVM, intended for producing
314 single-address-space managed code operating systems that
315 run faster than the equivalent multiple-address-space C systems.
316 More in-depth blurb is available on <a
317 href="http://www.quokforge.org/projects/horizon/wiki/Wiki">the wiki</a>.</p>
321 <!--=========================================================================-->
322 <div class="doc_subsection">
323 <a name="clamav">Clam AntiVirus</a>
326 <div class="doc_text">
328 <a href=http://www.clamav.net>Clam AntiVirus</a> is an open source (GPL)
329 anti-virus toolkit for UNIX, designed especially for e-mail scanning on mail
330 gateways. Since version 0.96 it has <a
331 href="http://vrt-sourcefire.blogspot.com/2010/09/introduction-to-clamavs-low-level.html">bytecode
332 signatures</a> that allow writing detections for complex malware. It
333 uses LLVM's JIT to speed up the execution of bytecode on
334 X86,X86-64,PPC32/64, falling back to its own interpreter otherwise.
335 The git version was updated to work with LLVM 2.8
339 href="http://git.clamav.net/gitweb?p=clamav-bytecode-compiler.git;a=blob_plain;f=docs/user/clambc-user.pdf">
340 ClamAV bytecode compiler</a> uses Clang and LLVM to compile a C-like
341 language, insert runtime checks, and generate ClamAV bytecode.</p>
345 <!--=========================================================================-->
346 <div class="doc_subsection">
347 <a name="pure">Pure</a>
350 <div class="doc_text">
352 <a href="http://pure-lang.googlecode.com/">Pure</a>
353 is an algebraic/functional
354 programming language based on term rewriting. Programs are collections
355 of equations which are used to evaluate expressions in a symbolic
356 fashion. Pure offers dynamic typing, eager and lazy evaluation, lexical
357 closures, a hygienic macro system (also based on term rewriting),
358 built-in list and matrix support (including list and matrix
359 comprehensions) and an easy-to-use C interface. The interpreter uses
360 LLVM as a backend to JIT-compile Pure programs to fast native code.</p>
362 <p>Pure versions 0.44 and later have been tested and are known to work with
363 LLVM 2.8 (and continue to work with older LLVM releases >= 2.5).</p>
367 <!--=========================================================================-->
368 <div class="doc_subsection">
369 <a name="GHC">Glasgow Haskell Compiler (GHC)</a>
372 <div class="doc_text">
374 <a href="http://www.haskell.org/ghc/">GHC</a> is an open source,
375 state-of-the-art programming suite for
376 Haskell, a standard lazy functional programming language. It includes
377 an optimizing static compiler generating good code for a variety of
378 platforms, together with an interactive system for convenient, quick
381 <p>In addition to the existing C and native code generators, GHC 7.0 now
383 href="http://hackage.haskell.org/trac/ghc/wiki/Commentary/Compiler/Backends/LLVM">LLVM
384 code generator</a>. GHC supports LLVM 2.7 and later.</p>
388 <!--=========================================================================-->
389 <div class="doc_subsection">
390 <a name="Clay">Clay Programming Language</a>
393 <div class="doc_text">
395 <a href="http://tachyon.in/clay/">Clay</a> is a new systems programming
396 language that is specifically designed for generic programming. It makes
397 generic programming very concise thanks to whole program type propagation. It
398 uses LLVM as its backend.</p>
402 <!--=========================================================================-->
403 <div class="doc_subsection">
404 <a name="llvm-py">llvm-py Python Bindings for LLVM</a>
407 <div class="doc_text">
409 <a href="http://www.mdevan.org/llvm-py/">llvm-py</a> has been updated to work
410 with LLVM 2.8. llvm-py provides Python bindings for LLVM, allowing you to write a
411 compiler backend or a VM in Python.</p>
416 <!--=========================================================================-->
417 <div class="doc_subsection">
418 <a name="FAUST">FAUST Real-Time Audio Signal Processing Language</a>
421 <div class="doc_text">
423 <a href="http://faust.grame.fr">FAUST</a> is a compiled language for real-time
424 audio signal processing. The name FAUST stands for Functional AUdio STream. Its
425 programming model combines two approaches: functional programming and block
426 diagram composition. In addition with the C, C++, JAVA output formats, the
427 Faust compiler can now generate LLVM bitcode, and works with LLVM 2.7 and
432 <!--=========================================================================-->
433 <div class="doc_subsection">
434 <a name="jade">Jade Just-in-time Adaptive Decoder Engine</a>
437 <div class="doc_text">
439 href="http://sourceforge.net/apps/trac/orcc/wiki/JadeDocumentation">Jade</a>
440 (Just-in-time Adaptive Decoder Engine) is a generic video decoder engine using
441 LLVM for just-in-time compilation of video decoder configurations. Those
442 configurations are designed by MPEG Reconfigurable Video Coding (RVC) committee.
443 MPEG RVC standard is built on a stream-based dataflow representation of
444 decoders. It is composed of a standard library of coding tools written in
445 RVC-CAL language and a dataflow configuration &emdash; block diagram &emdash;
448 <p>Jade project is hosted as part of the <a href="http://orcc.sf.net">Open
449 RVC-CAL Compiler</a> and requires it to translate the RVC-CAL standard library
450 of video coding tools into an LLVM assembly code.</p>
454 <!--=========================================================================-->
455 <div class="doc_subsection">
456 <a name="neko_llvm_jit">LLVM JIT for Neko VM</a>
459 <div class="doc_text">
460 <p><a href="http://github.com/vava/neko_llvm_jit">Neko LLVM JIT</a>
461 replaces the standard Neko JIT with an LLVM-based implementation. While not
462 fully complete, it is already providing a 1.5x speedup on 64-bit systems.
463 Neko LLVM JIT requires LLVM 2.8 or later.</p>
467 <!--=========================================================================-->
468 <div class="doc_subsection">
469 <a name="crack">Crack Scripting Language</a>
472 <div class="doc_text">
474 <a href="http://code.google.com/p/crack-language/">Crack</a> aims to provide
475 the ease of development of a scripting language with the performance of a
476 compiled language. The language derives concepts from C++, Java and Python,
477 incorporating object-oriented programming, operator overloading and strong
478 typing. Crack 0.2 works with LLVM 2.7, and the forthcoming Crack 0.2.1 release
479 builds on LLVM 2.8.</p>
483 <!--=========================================================================-->
484 <div class="doc_subsection">
485 <a name="DresdenTM">Dresden TM Compiler (DTMC)</a>
488 <div class="doc_text">
490 <a href="http://tm.inf.tu-dresden.de">DTMC</a> provides support for
491 Transactional Memory, which is an easy-to-use and efficient way to synchronize
492 accesses to shared memory. Transactions can contain normal C/C++ code (e.g.,
493 __transaction { list.remove(x); x.refCount--; }) and will be executed
494 virtually atomically and isolated from other transactions.</p>
498 <!--=========================================================================-->
499 <div class="doc_subsection">
500 <a name="Kai">Kai Interpreter</a>
503 <div class="doc_text">
505 <a href="http://www.oriontransfer.co.nz/research/kai">Kai</a> (Japanese 会 for
506 meeting/gathering) is an experimental interpreter that provides a highly
507 extensible runtime environment and explicit control over the compilation
508 process. Programs are defined using nested symbolic expressions, which are all
509 parsed into first-class values with minimal intrinsic semantics. Kai can
510 generate optimised code at run-time (using LLVM) in order to exploit the nature
511 of the underlying hardware and to integrate with external software libraries.
512 It is a unique exploration into world of dynamic code compilation, and the
513 interaction between high level and low level semantics.</p>
517 <!--=========================================================================-->
518 <div class="doc_subsection">
519 <a name="OSL">OSL: Open Shading Language</a>
522 <div class="doc_text">
524 <a href="http://code.google.com/p/openshadinglanguage/">OSL</a> is a shading
525 language designed for use in physically based renderers and in particular
526 production rendering. By using LLVM instead of the interpreter, it was able to
527 meet its performance goals (>= C-code) while retaining the benefits of
528 runtime specialization and a portable high-level language.
535 <!-- *********************************************************************** -->
536 <div class="doc_section">
537 <a name="whatsnew">What's New in LLVM 2.8?</a>
539 <!-- *********************************************************************** -->
541 <div class="doc_text">
543 <p>This release includes a huge number of bug fixes, performance tweaks and
544 minor improvements. Some of the major improvements and new features are listed
550 <!--=========================================================================-->
551 <div class="doc_subsection">
552 <a name="majorfeatures">Major New Features</a>
555 <div class="doc_text">
557 <p>LLVM 2.8 includes several major new capabilities:</p>
560 <li>As mentioned above, <a href="#libc++">libc++</a> and <a
561 href="#lldb">LLDB</a> are major new additions to the LLVM collective.</li>
562 <li>LLVM 2.8 now has pretty decent support for debugging optimized code. You
563 should be able to reliably get debug info for function arguments, assuming
564 that the value is actually available where you have stopped.</li>
566 <li>A new 'llvm-diff' tool is available that does a semantic diff of .ll
568 <li>The <a href="#mc">MC subproject</a> has made major progress in this release.
569 Direct .o file writing support for darwin/x86[-64] is now reliable and
570 support for other targets and object file formats are in progress.</li>
575 <!--=========================================================================-->
576 <div class="doc_subsection">
577 <a name="coreimprovements">LLVM IR and Core Improvements</a>
580 <div class="doc_text">
581 <p>LLVM IR has several new features for better support of new targets and that
582 expose new optimization opportunities:</p>
585 <li>The <a href="LangRef.html#int_libc">memcpy, memmove, and memset</a>
586 intrinsics now take address space qualified pointers and a bit to indicate
587 whether the transfer is "<a href="LangRef.html#volatile">volatile</a>" or not.
589 <li>Per-instruction debug info metadata is much faster and uses less memory by
590 using the new DebugLoc class.</li>
591 <li>LLVM IR now has a more formalized concept of "<a
592 href="LangRef.html#trapvalues">trap values</a>", which allow the optimizer
593 to optimize more aggressively in the presence of undefined behavior, while
594 still producing predictable results.</li>
595 <li>LLVM IR now supports two new <a href="LangRef.html#linkage">linkage
596 types</a> (linker_private_weak and linker_private_weak_def_auto) which map
597 onto some obscure MachO concepts.</li>
602 <!--=========================================================================-->
603 <div class="doc_subsection">
604 <a name="optimizer">Optimizer Improvements</a>
607 <div class="doc_text">
609 <p>In addition to a large array of minor performance tweaks and bug fixes, this
610 release includes a few major enhancements and additions to the optimizers:</p>
613 <li>As mentioned above, the optimizer now has support for updating debug
614 information as it goes. A key aspect of this is the new <a
615 href="SourceLevelDebugging.html#format_common_value">llvm.dbg.value</a>
616 intrinsic. This intrinsic represents debug info for variables that are
617 promoted to SSA values (typically by mem2reg or the -scalarrepl passes).</li>
619 <li>The JumpThreading pass is now much more aggressive about implied value
620 relations, allowing it to thread conditions like "a == 4" when a is known to
621 be 13 in one of the predecessors of a block. It does this in conjunction
622 with the new LazyValueInfo analysis pass.</li>
623 <li>The new RegionInfo analysis pass identifies single-entry single-exit regions
624 in the CFG. You can play with it with the "opt -regions analyze" or
625 "opt -view-regions" commands.</li>
626 <li>The loop optimizer has significantly improve strength reduction and analysis
627 capabilities. Notably it is able to build on the trap value and signed
628 integer overflow information to optimize <= and >= loops.</li>
629 <li>The CallGraphSCCPassManager now has some basic support for iterating within
630 an SCC when a optimizer devirtualizes a function call. This allows inlining
631 through indirect call sites that are devirtualized by store-load forwarding
632 and other optimizations.</li>
633 <li>The new <A href="Passes.html#loweratomic">-loweratomic</a> pass is available
634 to lower atomic instructions into their non-atomic form. This can be useful
635 to optimize generic code that expects to run in a single-threaded
640 <p>In addition to these features that are done in 2.8, there is preliminary
641 support in the release for Type Based Alias Analysis
642 Preliminary work on TBAA but not usable in 2.8.
643 New CorrelatedValuePropagation pass, not on by default in 2.8 yet.
648 <!--=========================================================================-->
649 <div class="doc_subsection">
650 <a name="mc">MC Level Improvements</a>
653 <div class="doc_text">
655 The LLVM Machine Code (aka MC) subsystem was created to solve a number
656 of problems in the realm of assembly, disassembly, object file format handling,
657 and a number of other related areas that CPU instruction-set level tools work
660 <p>The MC subproject has made great leaps in LLVM 2.8. For example, support for
661 directly writing .o files from LLC (and clang) now works reliably for
662 darwin/x86[-64] (including inline assembly support) and the integrated
663 assembler is turned on by default in Clang for these targets. This provides
664 improved compile times among other things.</p>
667 <li>The entire compiler has converted over to using the MCStreamer assembler API
668 instead of writing out a .s file textually.</li>
669 <li>The "assembler parser" is far more mature than in 2.7, supporting a full
670 complement of directives, now supports assembler macros, etc.</li>
671 <li>The "assembler backend" has been completed, including support for relaxation
672 relocation processing and all the other things that an assembler does.</li>
673 <li>The MachO file format support is now fully functional and works.</li>
674 <li>The MC disassembler now fully supports ARM and Thumb. ARM assembler support
675 is still in early development though.</li>
676 <li>The X86 MC assembler now supports the X86 AES and AVX instruction set.</li>
677 <li>Work on ELF and COFF object files and ARM target support is well underway,
678 but isn't useful yet in LLVM 2.8. Please contact the llvmdev mailing list
679 if you're interested in this.</li>
682 <p>For more information, please see the <a
683 href="http://blog.llvm.org/2010/04/intro-to-llvm-mc-project.html">Intro to the
684 LLVM MC Project Blog Post</a>.
690 <!--=========================================================================-->
691 <div class="doc_subsection">
692 <a name="codegen">Target Independent Code Generator Improvements</a>
695 <div class="doc_text">
697 <p>We have put a significant amount of work into the code generator
698 infrastructure, which allows us to implement more aggressive algorithms and make
702 <li>The clang/gcc -momit-leaf-frame-pointer argument is now supported.</li>
703 <li>The clang/gcc -ffunction-sections and -fdata-sections arguments are now
704 supported on ELF targets (like GCC).</li>
705 <li>The MachineCSE pass is now tuned and on by default. It eliminates common
706 subexpressions that are exposed when lowering to machine instructions.</li>
707 <li>The "local" register allocator was replaced by a new "fast" register
708 allocator. This new allocator (which is often used at -O0) is substantially
709 faster and produces better code than the old local register allocator.</li>
710 <li>A new LLC "-regalloc=default" option is available, which automatically
711 chooses a register allocator based on the -O optimization level.</li>
712 <li>The common code generator code was modified to promote illegal argument and
713 return value vectors to wider ones when possible instead of scalarizing
714 them. For example, <3 x float> will now pass in one SSE register
715 instead of 3 on X86. This generates substantially better code since the
716 rest of the code generator was already expecting this.</li>
717 <li>The code generator uses a new "COPY" machine instruction. This speeds up
718 the code generator and eliminates the need for targets to implement the
719 isMoveInstr hook. Also, the copyRegToReg hook was renamed to copyPhysReg
721 <li>The code generator now has a "LocalStackSlotPass", which optimizes stack
722 slot access for targets (like ARM) that have limited stack displacement
724 <li>A new "PeepholeOptimizer" is available, which eliminates sign and zero
725 extends, and optimizes away compare instructions when the condition result
726 is available from a previous instruction.</li>
727 <li>Atomic operations now get legalized into simpler atomic operations if not
728 natively supported, easy the implementation burden on targets.</li>
729 <li>The bottom-up pre-allocation scheduler is now register pressure aware,
730 allowing it to avoid overscheduling in high pressure situations while still
731 aggressively scheduling when registers are available.</li>
732 <li>A new instruction-level-parallelism pre-allocation scheduler is available,
733 which is also register pressure aware. This scheduler has shown substantial
734 wins on X86-64 and is on by default.</li>
735 <li>The tblgen type inference algorithm was rewritten to be more consistent and
736 diagnose more target bugs. If you have an out-of-tree backend, you may
737 find that it finds bugs in your target description. This support also
738 allows limited support for writing patterns for instructions that return
739 multiple results (e.g. a virtual register and a flag result). The
740 'parallel' modifier in tblgen was removed, you should use the new support
741 for multiple results instead.</li>
742 <li>A new (experimental) "-rendermf" pass is available which renders a
743 MachineFunction into HTML, showing live ranges and other useful
745 <li>The new SubRegIndex tablegen class allows subregisters to be indexed
746 symbolically instead of numerically. If your target uses subregisters you
747 will need to adapt to use SubRegIndex when you upgrade to 2.8.</li>
750 <li>The -fast-isel instruction selection path (used at -O0 on X86) was rewritten
751 to work bottom-up on basic blocks instead of top down. This makes it
752 slightly faster (because the MachineDCE pass is not needed any longer) and
753 allows it to generate better code in some cases.</li>
758 <!--=========================================================================-->
759 <div class="doc_subsection">
760 <a name="x86">X86-32 and X86-64 Target Improvements</a>
763 <div class="doc_text">
764 <p>New features and major changes in the X86 target include:
768 <li>The X86 backend now supports holding X87 floating point stack values
769 in registers across basic blocks, dramatically improving performance of code
770 that uses long double, and when targetting CPUs that don't support SSE.</li>
772 <li>The X86 backend now uses a SSEDomainFix pass to optimize SSE operations. On
773 Nehalem ("Core i7") and newer CPUs there is a 2 cycle latency penalty on
774 using a register in a different domain than where it was defined. This pass
775 optimizes away these stalls.</li>
777 <li>The X86 backend now promote 16-bit integer operations to 32-bits when
778 possible. This avoids 0x66 prefixes, which are slow on some
779 microarchitectures and bloat the code on all of them.</li>
781 <li>The X86 backend now supports the Microsoft "thiscall" calling convention,
782 and a <a href="LangRef.html#callingconv">calling convention</a> to support
783 <a href="#GHC">ghc</a>.</li>
785 <li>The X86 backend supports a new "llvm.x86.int" intrinsic, which maps onto
786 the X86 "int $42" and "int3" instructions.</li>
788 <li>At the IR level, the <2 x float> datatype is now promoted and passed
789 around as a <4 x float> instead of being passed and returns as an MMX
790 vector. If you have a frontend that uses this, please pass and return a
791 <2 x i32> instead (using bitcasts).</li>
793 <li>When printing .s files in verbose assembly mode (the default for clang -S),
794 the X86 backend now decodes X86 shuffle instructions and prints human
795 readable comments after the most inscrutible of them, e.g.:
798 insertps $113, %xmm3, %xmm0 <i># xmm0 = zero,xmm0[1,2],xmm3[1]</i>
799 unpcklps %xmm1, %xmm0 <i># xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]</i>
800 pshufd $1, %xmm1, %xmm1 <i># xmm1 = xmm1[1,0,0,0]</i>
808 <!--=========================================================================-->
809 <div class="doc_subsection">
810 <a name="ARM">ARM Target Improvements</a>
813 <div class="doc_text">
814 <p>New features of the ARM target include:
818 <li>The ARM backend now optimizes tail calls into jumps.</li>
819 <li>Scheduling is improved through the new list-hybrid scheduler as well
820 as through better modeling of structural hazards.</li>
821 <li><a href="LangRef.html#int_fp16">Half float</a> instructions are now
823 <li>NEON support has been improved to model instructions which operate onto
824 multiple consequtive registers more aggressively. This avoids lots of
825 extraneous register copies.</li>
826 <li>The ARM backend now uses a new "ARMGlobalMerge" pass, which merges several
827 global variables into one, saving extra address computation (all the global
828 variables can be accessed via same base address) and potentially reducing
829 register pressure.</li>
831 <li>The ARM has received many minor improvements and tweaks which lead to
832 substantially better performance in a wide range of different scenarios.</li>
834 <li>The ARM NEON intrinsics have been substantially reworked to reduce
835 redundancy and improve code generation. Some of the major changes are:
838 All of the NEON load and store intrinsics (llvm.arm.neon.vld* and
839 llvm.arm.neon.vst*) take an extra parameter to specify the alignment in bytes
840 of the memory being accessed.
843 The llvm.arm.neon.vaba intrinsic (vector absolute difference and
844 accumulate) has been removed. This operation is now represented using
845 the llvm.arm.neon.vabd intrinsic (vector absolute difference) followed by a
849 The llvm.arm.neon.vabdl and llvm.arm.neon.vabal intrinsics (lengthening
850 vector absolute difference with and without accumlation) have been removed.
851 They are represented using the llvm.arm.neon.vabd intrinsic (vector absolute
852 difference) followed by a vector zero-extend operation, and for vabal,
856 The llvm.arm.neon.vmovn intrinsic has been removed. Calls of this intrinsic
857 are now replaced by vector truncate operations.
860 The llvm.arm.neon.vmovls and llvm.arm.neon.vmovlu intrinsics have been
861 removed. They are now represented as vector sign-extend (vmovls) and
862 zero-extend (vmovlu) operations.
865 The llvm.arm.neon.vaddl*, llvm.arm.neon.vaddw*, llvm.arm.neon.vsubl*, and
866 llvm.arm.neon.vsubw* intrinsics (lengthening vector add and subtract) have
867 been removed. They are replaced by vector add and vector subtract operations
868 where one (vaddw, vsubw) or both (vaddl, vsubl) of the operands are either
869 sign-extended or zero-extended.
872 The llvm.arm.neon.vmulls, llvm.arm.neon.vmullu, llvm.arm.neon.vmlal*, and
873 llvm.arm.neon.vmlsl* intrinsics (lengthening vector multiply with and without
874 accumulation and subtraction) have been removed. These operations are now
875 represented as vector multiplications where the operands are either
876 sign-extended or zero-extended, followed by a vector add for vmlal or a
877 vector subtract for vmlsl. Note that the polynomial vector multiply
878 intrinsic, llvm.arm.neon.vmullp, remains unchanged.
887 <!--=========================================================================-->
888 <div class="doc_subsection">
889 <a name="changes">Major Changes and Removed Features</a>
892 <div class="doc_text">
894 <p>If you're already an LLVM user or developer with out-of-tree changes based
895 on LLVM 2.7, this section lists some "gotchas" that you may run into upgrading
896 from the previous release.</p>
899 <li>The build configuration machinery changed the output directory names. It
900 wasn't clear to many people that "Release-Asserts" build was a release build
901 without asserts. To make this more clear, "Release" does not include
902 assertions and "Release+Asserts" does (likewise, "Debug" and
903 "Debug+Asserts").</li>
904 <li>The MSIL Backend was removed, it was unsupported and broken.</li>
905 <li>The ABCD, SSI, and SCCVN passes were removed. These were not fully
906 functional and their behavior has been or will be subsumed by the
907 LazyValueInfo pass.</li>
908 <li>The LLVM IR 'Union' feature was removed. While this is a desirable feature
909 for LLVM IR to support, the existing implementation was half baked and
910 barely useful. We'd really like anyone interested to resurrect the work and
911 finish it for a future release.</li>
912 <li>If you're used to reading .ll files, you'll probably notice that .ll file
913 dumps don't produce #uses comments anymore. To get them, run a .bc file
914 through "llvm-dis --show-annotations".</li>
915 <li>Target triples are now stored in a normalized form, and all inputs from
916 humans are expected to be normalized by Triple::normalize before being
917 stored in a module triple or passed to another library.</li>
922 <p>In addition, many APIs have changed in this release. Some of the major LLVM
925 <li>LLVM 2.8 changes the internal order of operands in <a
926 href="http://llvm.org/doxygen/classllvm_1_1InvokeInst.html"><tt>InvokeInst</tt></a>
927 and <a href="http://llvm.org/doxygen/classllvm_1_1CallInst.html"><tt>CallInst</tt></a>.
928 To be portable across releases, please use the <tt>CallSite</tt> class and the
929 high-level accessors, such as <tt>getCalledValue</tt> and
930 <tt>setUnwindDest</tt>.
933 You can no longer pass use_iterators directly to cast<> (and similar),
934 because these routines tend to perform costly dereference operations more
935 than once. You have to dereference the iterators yourself and pass them in.
938 llvm.memcpy.*, llvm.memset.*, llvm.memmove.* intrinsics take an extra
939 parameter now ("i1 isVolatile"), totaling 5 parameters, and the pointer
940 operands are now address-space qualified.
941 If you were creating these intrinsic calls and prototypes yourself (as opposed
942 to using Intrinsic::getDeclaration), you can use
943 UpgradeIntrinsicFunction/UpgradeIntrinsicCall to be portable accross releases.
946 SetCurrentDebugLocation takes a DebugLoc now instead of a MDNode.
947 Change your code to use
948 SetCurrentDebugLocation(DebugLoc::getFromDILocation(...)).
951 The <tt>RegisterPass</tt> and <tt>RegisterAnalysisGroup</tt> templates are
952 considered deprecated, but continue to function in LLVM 2.8. Clients are
953 strongly advised to use the upcoming <tt>INITIALIZE_PASS()</tt> and
954 <tt>INITIALIZE_AG_PASS()</tt> macros instead.
957 The constructor for the Triple class no longer tries to understand odd triple
958 specifications. Frontends should ensure that they only pass valid triples to
959 LLVM. The Triple::normalize utility method has been added to help front-ends
960 deal with funky triples.
964 Some APIs got renamed:
966 <li>llvm_report_error -> report_fatal_error</li>
967 <li>llvm_install_error_handler -> install_fatal_error_handler</li>
968 <li>llvm::DwarfExceptionHandling -> llvm::JITExceptionHandling</li>
969 <li>VISIBILITY_HIDDEN -> LLVM_LIBRARY_VISIBILITY</li>
978 <!-- *********************************************************************** -->
979 <div class="doc_section">
980 <a name="knownproblems">Known Problems</a>
982 <!-- *********************************************************************** -->
984 <div class="doc_text">
986 <p>This section contains significant known problems with the LLVM system,
987 listed by component. If you run into a problem, please check the <a
988 href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
989 there isn't already one.</p>
993 <!-- ======================================================================= -->
994 <div class="doc_subsection">
995 <a name="experimental">Experimental features included with this release</a>
998 <div class="doc_text">
1000 <p>The following components of this LLVM release are either untested, known to
1001 be broken or unreliable, or are in early development. These components should
1002 not be relied on, and bugs should not be filed against them, but they may be
1003 useful to some people. In particular, if you would like to work on one of these
1004 components, please contact us on the <a
1005 href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev list</a>.</p>
1008 <li>The Alpha, Blackfin, CellSPU, MicroBlaze, MSP430, MIPS, PIC16, SystemZ
1009 and XCore backends are experimental.</li>
1010 <li><tt>llc</tt> "<tt>-filetype=obj</tt>" is experimental on all targets
1011 other than darwin-i386 and darwin-x86_64.</li>
1016 <!-- ======================================================================= -->
1017 <div class="doc_subsection">
1018 <a name="x86-be">Known problems with the X86 back-end</a>
1021 <div class="doc_text">
1024 <li>The X86 backend does not yet support
1025 all <a href="http://llvm.org/PR879">inline assembly that uses the X86
1026 floating point stack</a>. It supports the 'f' and 't' constraints, but not
1028 <li>Win64 code generation wasn't widely tested. Everything should work, but we
1029 expect small issues to happen. Also, llvm-gcc cannot build the mingw64
1030 runtime currently due to lack of support for the 'u' inline assembly
1031 constraint and for X87 floating point inline assembly.</li>
1032 <li>The X86-64 backend does not yet support the LLVM IR instruction
1033 <tt>va_arg</tt>. Currently, front-ends support variadic
1034 argument constructs on X86-64 by lowering them manually.</li>
1039 <!-- ======================================================================= -->
1040 <div class="doc_subsection">
1041 <a name="ppc-be">Known problems with the PowerPC back-end</a>
1044 <div class="doc_text">
1047 <li>The Linux PPC32/ABI support needs testing for the interpreter and static
1048 compilation, and lacks support for debug information.</li>
1053 <!-- ======================================================================= -->
1054 <div class="doc_subsection">
1055 <a name="arm-be">Known problems with the ARM back-end</a>
1058 <div class="doc_text">
1061 <li>Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6
1062 processors, thumb programs can crash or produce wrong
1063 results (<a href="http://llvm.org/PR1388">PR1388</a>).</li>
1064 <li>Compilation for ARM Linux OABI (old ABI) is supported but not fully tested.
1070 <!-- ======================================================================= -->
1071 <div class="doc_subsection">
1072 <a name="sparc-be">Known problems with the SPARC back-end</a>
1075 <div class="doc_text">
1078 <li>The SPARC backend only supports the 32-bit SPARC ABI (-m32); it does not
1079 support the 64-bit SPARC ABI (-m64).</li>
1084 <!-- ======================================================================= -->
1085 <div class="doc_subsection">
1086 <a name="mips-be">Known problems with the MIPS back-end</a>
1089 <div class="doc_text">
1092 <li>64-bit MIPS targets are not supported yet.</li>
1097 <!-- ======================================================================= -->
1098 <div class="doc_subsection">
1099 <a name="alpha-be">Known problems with the Alpha back-end</a>
1102 <div class="doc_text">
1106 <li>On 21164s, some rare FP arithmetic sequences which may trap do not have the
1107 appropriate nops inserted to ensure restartability.</li>
1112 <!-- ======================================================================= -->
1113 <div class="doc_subsection">
1114 <a name="c-be">Known problems with the C back-end</a>
1117 <div class="doc_text">
1119 <p>The C backend has numerous problems and is not being actively maintained.
1120 Depending on it for anything serious is not advised.</p>
1123 <li><a href="http://llvm.org/PR802">The C backend has only basic support for
1124 inline assembly code</a>.</li>
1125 <li><a href="http://llvm.org/PR1658">The C backend violates the ABI of common
1126 C++ programs</a>, preventing intermixing between C++ compiled by the CBE and
1127 C++ code compiled with <tt>llc</tt> or native compilers.</li>
1128 <li>The C backend does not support all exception handling constructs.</li>
1129 <li>The C backend does not support arbitrary precision integers.</li>
1135 <!-- ======================================================================= -->
1136 <div class="doc_subsection">
1137 <a name="llvm-gcc">Known problems with the llvm-gcc front-end</a>
1140 <div class="doc_text">
1142 <p>llvm-gcc is generally very stable for the C family of languages. The only
1143 major language feature of GCC not supported by llvm-gcc is the
1144 <tt>__builtin_apply</tt> family of builtins. However, some extensions
1145 are only supported on some targets. For example, trampolines are only
1146 supported on some targets (these are used when you take the address of a
1147 nested function).</p>
1149 <p>Fortran support generally works, but there are still several unresolved bugs
1150 in <a href="http://llvm.org/bugs/">Bugzilla</a>. Please see the
1151 tools/gfortran component for details. Note that llvm-gcc is missing major
1152 Fortran performance work in the frontend and library that went into GCC after
1153 4.2. If you are interested in Fortran, we recommend that you consider using
1154 <a href="#dragonegg">dragonegg</a> instead.</p>
1156 <p>The llvm-gcc 4.2 Ada compiler has basic functionality. However, this is not a
1157 mature technology, and problems should be expected. For example:</p>
1159 <li>The Ada front-end currently only builds on X86-32. This is mainly due
1160 to lack of trampoline support (pointers to nested functions) on other platforms.
1161 However, it <a href="http://llvm.org/PR2006">also fails to build on X86-64</a>
1162 which does support trampolines.</li>
1163 <li>The Ada front-end <a href="http://llvm.org/PR2007">fails to bootstrap</a>.
1164 This is due to lack of LLVM support for <tt>setjmp</tt>/<tt>longjmp</tt> style
1165 exception handling, which is used internally by the compiler.
1166 Workaround: configure with <tt>--disable-bootstrap</tt>.</li>
1167 <li>The c380004, <a href="http://llvm.org/PR2010">c393010</a>
1168 and <a href="http://llvm.org/PR2421">cxg2021</a> ACATS tests fail
1169 (c380004 also fails with gcc-4.2 mainline).
1170 If the compiler is built with checks disabled then <a href="http://llvm.org/PR2010">c393010</a>
1171 causes the compiler to go into an infinite loop, using up all system memory.</li>
1172 <li>Some GCC specific Ada tests continue to crash the compiler.</li>
1173 <li>The <tt>-E</tt> binder option (exception backtraces)
1174 <a href="http://llvm.org/PR1982">does not work</a> and will result in programs
1175 crashing if an exception is raised. Workaround: do not use <tt>-E</tt>.</li>
1176 <li>Only discrete types <a href="http://llvm.org/PR1981">are allowed to start
1177 or finish at a non-byte offset</a> in a record. Workaround: do not pack records
1178 or use representation clauses that result in a field of a non-discrete type
1179 starting or finishing in the middle of a byte.</li>
1180 <li>The <tt>lli</tt> interpreter <a href="http://llvm.org/PR2009">considers
1181 'main' as generated by the Ada binder to be invalid</a>.
1182 Workaround: hand edit the file to use pointers for <tt>argv</tt> and
1183 <tt>envp</tt> rather than integers.</li>
1184 <li>The <tt>-fstack-check</tt> option <a href="http://llvm.org/PR2008">is
1189 <!-- *********************************************************************** -->
1190 <div class="doc_section">
1191 <a name="additionalinfo">Additional Information</a>
1193 <!-- *********************************************************************** -->
1195 <div class="doc_text">
1197 <p>A wide variety of additional information is available on the <a
1198 href="http://llvm.org">LLVM web page</a>, in particular in the <a
1199 href="http://llvm.org/docs/">documentation</a> section. The web page also
1200 contains versions of the API documentation which is up-to-date with the
1201 Subversion version of the source code.
1202 You can access versions of these documents specific to this release by going
1203 into the "<tt>llvm/doc/</tt>" directory in the LLVM tree.</p>
1205 <p>If you have any questions or comments about LLVM, please feel free to contact
1206 us via the <a href="http://llvm.org/docs/#maillist"> mailing
1211 <!-- *********************************************************************** -->
1215 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1216 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1217 <a href="http://validator.w3.org/check/referer"><img
1218 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1220 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
1221 Last modified: $Date$