2 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
3 "http://www.w3.org/TR/html4/strict.dtd">
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <link rel="stylesheet" href="llvm.css" type="text/css">
8 <title>LLVM 2.3 Release Notes</title>
12 <div class="doc_title">LLVM 2.3 Release Notes</div>
15 <li><a href="#intro">Introduction</a></li>
16 <li><a href="#whatsnew">What's New?</a></li>
17 <li><a href="GettingStarted.html">Installation Instructions</a></li>
18 <li><a href="#portability">Portability and Supported Platforms</a></li>
19 <li><a href="#knownproblems">Known Problems</a>
20 <li><a href="#additionalinfo">Additional Information</a></li>
23 <div class="doc_author">
24 <p>Written by the <a href="http://llvm.org">LLVM Team</a><p>
27 <!-- Done through Week-of-Mon-20080324.txt -->
29 <!-- *********************************************************************** -->
30 <div class="doc_section">
31 <a name="intro">Introduction</a>
33 <!-- *********************************************************************** -->
35 <div class="doc_text">
37 <p>This document contains the release notes for the LLVM compiler
38 infrastructure, release 2.3. Here we describe the status of LLVM, including
39 major improvements from the previous release and any known problems. All LLVM
40 releases may be downloaded from the <a href="http://llvm.org/releases/">LLVM
41 releases web site</a>.</p>
43 <p>For more information about LLVM, including information about the latest
44 release, please check out the <a href="http://llvm.org/">main LLVM
45 web site</a>. If you have questions or comments, the <a
46 href="http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM developer's mailing
47 list</a> is a good place to send them.</p>
49 <p>Note that if you are reading this file from a Subversion checkout or the
50 main LLVM web page, this document applies to the <i>next</i> release, not the
51 current one. To see the release notes for a specific releases, please see the
52 <a href="http://llvm.org/releases/">releases page</a>.</p>
56 <!-- *********************************************************************** -->
57 <div class="doc_section">
58 <a name="whatsnew">What's New?</a>
60 <!-- *********************************************************************** -->
62 <div class="doc_text">
64 <p>This is the fourteenth public release of the LLVM Compiler Infrastructure.
65 It includes a large number of features and refinements from LLVM 2.2.</p>
69 <!-- Unfinished features in 2.3:
75 <!--=========================================================================-->
76 <div class="doc_subsection">
77 <a name="changes">Major Changes in LLVM 2.3</a>
80 <div class="doc_text">
82 <p>LLVM 2.2 was the last LLVM release to support llvm-gcc 4.0 and llvm-upgrade.
83 llvm-gcc 4.0 has been replaced with llvm-gcc 4.2. llvm-upgrade was useful for
84 upgrading LLVM 1.9 files to LLVM 2.x syntax, but you can always use a previous
85 LLVM release to do this. One nice impact of this is that the LLVM regression
86 test suite no longer depends on llvm-upgrade, which makes it run faster.</p>
88 <p><tt>llvm2cpp</tt> tool has been folded into llc, use
89 <tt>llc -march=cpp</tt> instead of <tt>llvm2cpp</tt>.</p>
91 <p>LLVM API Changes:</p>
94 <li>Several core LLVM IR classes have migrated to use 'FOOCLASS::Create(...)'
95 instead of 'new FOOCLASS(...)' (e.g. where FOOCLASS=BasicBlock). We hope to
96 standardize on FOOCLASS::Create for all IR classes in the future, but not
97 all of them have been moved over yet.</li>
98 <li>LLVM 2.3 renames the LLVMBuilder and LLVMFoldingBuilder classes to
100 <li>MRegisterInfo was renamed to TargetRegisterInfo.</li>
101 <li>The MappedFile class is gone, please use MemoryBuffer instead.</li>
102 <li>The '-enable-eh' flag to llc has been removed. Now code should encode
103 whether it is safe to not generate unwind information for a function by
104 tagging the Function object with the 'nounwind' attribute.</li>
109 <!--=========================================================================-->
110 <div class="doc_subsection">
111 <a name="majorfeatures">Major New Features</a>
114 <div class="doc_text">
116 <p>LLVM 2.3 includes several major new capabilities:</p>
119 <li>Multiple Return Value Support.</li>
122 <li><p>LLVM 2.3 includes a complete reimplementation of the "llvmc" tool. It is
123 designed to overcome several problems with the original llvmc and to provide a
124 superset of the features of the 'gcc' driver.</p>
126 <p>The main features of llvmc2 is:</p>
129 <li>Extended handling of command line options and smart rules for
130 dispatching them to different tools.</li>
131 <li>Flexible (and extensible) rules for defining different tools.</li>
132 <li>The different intermediate steps performed by tools are represented
133 as edges in the abstract graph.</li>
134 <li>The 'language' for driver behavior definition is tablegen and thus
135 it's relatively easy to add new features.</li>
136 <li>The definition of driver is transformed into set of C++ classes, thus
137 no runtime interpretation is needed.</li>
142 <li>Reimplemented <a href="LinkTimeOptimization.html">LTO interface</a> in
146 <li>Kaleidoscope tutorial in Ocaml.</li>
153 <!--=========================================================================-->
154 <div class="doc_subsection">
155 <a name="frontends">llvm-gcc 4.2 Improvements and Clang</a>
158 <div class="doc_text">
160 <p>LLVM 2.3 fully supports the llvm-gcc 4.2 front-end.</p>
162 <p>llvm-gcc 4.2 includes numerous fixes to better support the Objective-C
163 front-end. Objective-C now works very well on Mac OS/X.</p>
165 <p>llvm-gcc 4.2 includes many other fixes which improve conformance with the
166 relevant parts of the GCC testsuite.</p>
168 <p>The <a href="http://clang.llvm.org/">clang project</a> is an effort to build
169 a set of new 'llvm native' front-end technologies for the LLVM optimizer
170 and code generator. Currently, its C and Objective-C support is maturing
171 nicely, and it has advanced source-to-source analysis and transformation
172 capabilities. If you are interested in building source-level tools for C and
173 Objective-C (and eventually C++), you should take a look. However, note that
174 clang is not an official part of the LLVM 2.3 release. If you are interested in
175 this project, please see its <a href="http://clang.llvm.org/">web site</a>.</p>
180 <!--=========================================================================-->
181 <div class="doc_subsection">
182 <a name="coreimprovements">LLVM Core Improvements</a>
185 <div class="doc_text">
186 <p>New features include:
192 Atomic operation support, Alpha, X86, X86-64, PowerPC. "__sync_synchronize",
193 "__sync_val_compare_and_swap", etc
196 <li>The C and Ocaml bindings have received additional improvements. The
197 bindings now cover pass managers, several transformation passes, iteration
198 over the LLVM IR, target data, and parameter attribute lists.</li>
203 <!--=========================================================================-->
204 <div class="doc_subsection">
205 <a name="optimizer">Optimizer Improvements</a>
208 <div class="doc_text">
210 <p>In addition to a huge array of bug fixes and minor performance tweaks, the
211 LLVM 2.3 optimizers support a few major enhancements:</p>
215 <li>Loop index set splitting on by default.<p>
216 This transformation hoists conditions from loop bodies and reduces loop's
217 iteration space to improve performance. For example, <p>
219 for (i = LB; i < UB; ++i)
226 for (i = LB; i < NUB; ++i)
231 <li>LLVM includes a new <tt>memcpy</tt> optimization pass removes
232 dead <tt>memcpy</tt> calls, unneeded copies of aggregates, and performs
233 return slot optimization. The LLVM optimizer now notices long sequences of
234 consecutive stores and merges them into <tt>memcpy</tt>'s where profitable.</li>
236 <li>Alignment detection for vector memory references and for <tt>memcpy</tt> and
237 <tt>memset</tt> is now more aggressive.</li>
239 <li>The aggressive dead code elimination (ADCE) optimization has been rewritten
240 to make it both faster and safer in the presence of code containing infinite
241 loops. Some of its prior functionality has been factored out into the loop
242 deletion pass, which <em>is</em> safe for infinite loops.</li>
244 <li>Several optimizations have been sped up, leading to faster code generation
245 with the same code quality.</li>
247 <li>The 'SimplifyLibCalls' pass, which optimizes calls to libc and libm
248 functions for C-based languages, has been rewritten to be a FunctionPass
249 instead a ModulePass. This allows it to be run more often and to be
250 included at -O1 in llvm-gcc. It was also extended to include more
251 optimizations and several corner case bugs were fixed.</li>
253 <li>LLVM now includes a simple 'Jump Threading' pass, which attempts to simplify
254 conditional branches using information about predecessor blocks, simplifying
255 the control flow graph. This pass is pretty basic at this point, but
256 catches some important cases and provides a foundation to build on.</li>
258 <li>Several corner case bugs which could lead to deleting volatile memory
259 accesses have been fixed.</li>
264 <!--=========================================================================-->
265 <div class="doc_subsection">
266 <a name="codegen">Code Generator Improvements</a>
269 <div class="doc_text">
271 <p>We put a significant amount of work into the code generator infrastructure,
272 which allows us to implement more aggressive algorithms and make it run
276 <li>MemOperand in the code generator: describe me!.</li>
278 <li>The target-independent code generator infrastructure now uses LLVM's APInt
279 class to handle integer values, which allows it to support integer types
280 larger than 64 bits. Note that support for such types is also dependent on
281 target-specific support. Use of APInt is also a step toward support for
282 non-power-of-2 integer sizes.</li>
284 <li>LLVM 2.3 includes several compile time speedups for code with large basic
285 blocks, particular in the instruction selection phase, register allocation,
286 scheduling, and tail merging/jump threading.</li>
288 <li>Several improvements which make llc's <tt>--view-sunit-dags</tt>
289 visualization of scheduling dependency graphs easier to understand.</li>
291 <li>The code generator allows targets to write patterns that generate subreg
292 references directly in .td files now.</li>
294 <li><tt>memcpy</tt> lowering in the backend is more aggressive, particularly for
295 <tt>memcpy</tt> calls introduced by the code generator when handling
296 pass-by-value structure argument copies.</li>
298 <li>Inline assembly with multiple register results now returns those results
299 directly in the appropriate registers, rather than going through memory.
300 Inline assembly that uses constraints like "ir" with immediates now use the
301 'i' form when possible instead of always loading the value in a register.
302 This saves an instruction and reduces register use.</li>
304 <li>Added support for PIC/GOT style tail calls on x86/32 and initial support
305 for tail calls on PowerPC 32 (it may also work on ppc64 but not
306 thoroughly tested).</li>
312 <!--=========================================================================-->
313 <div class="doc_subsection">
314 <a name="x86specific">X86/X86-64 Specific Improvements</a>
317 <div class="doc_text">
318 <p>New target-specific features include:
322 <li>llvm-gcc's X86-64 ABI conformance is far improved, particularly in the
323 area of passing and returning structures by value. llvm-gcc compiled code
324 now interoperates very well on X86-64 systems with other compilers.</li>
326 <li>Support for Win64 was added. This includes code generation itself, JIT
327 support and necessary changes to llvm-gcc.</li>
329 <li>The LLVM X86 backend now supports the support SSE 4.1 instruction set, and
330 the llvm-gcc 4.2 front-end supports the SSE 4.1 compiler builtins. Various
331 generic vector operations (insert/extract/shuffle) are much more efficient
332 when SSE 4.1 is enabled. The JIT automatically takes advantage of these
333 instructions, but llvm-gcc must be explicitly told to use them, e.g. with
334 <tt>-march=penryn</tt>.</li>
336 <li>The X86 backend now does a number of optimizations that aim to avoid
337 converting numbers back and forth from SSE registers to the X87 floating
340 <li>The X86 backend supports stack realignment, which is particularly useful for
341 vector code on OS's without 16-byte aligned stacks.</li>
343 <li>The X86 backend now supports the "sseregparm" options in GCC, which allow
344 functions to be tagged as passing floating point values in SSE
347 <li>Trampolines (taking the address of a nested function) now work on
350 <li><tt>__builtin_prefetch</tt> is now compiled into the appropriate prefetch
351 instructions instead of being ignored.</li>
353 <li>128-bit integers are now supported on X86-64 targets.</li>
355 <li>The register allocator can now rematerialize PIC-base computations.</li>
357 <li>The "t" and "f" inline assembly constraints for the X87 floating point stack
358 now work. However, the "u" constraint is still not fully supported.</li>
364 <!--=========================================================================-->
365 <div class="doc_subsection">
366 <a name="targetspecific">Other Target Specific Improvements</a>
369 <div class="doc_text">
370 <p>New target-specific features include:
374 <li>The LLVM C backend now supports vector code.</li>
381 <!--=========================================================================-->
382 <div class="doc_subsection">
383 <a name="otherimprovements">Other Improvements</a>
386 <div class="doc_text">
387 <p>New features include:
391 <li>LLVM now builds with GCC 4.3.</li>
392 <li>Bugpoint now supports running custom scripts (with the <tt>-run-custom</tt>
393 option) to determine how to execute the command and whether it is making
394 forward process.</li>
399 <!-- *********************************************************************** -->
400 <div class="doc_section">
401 <a name="portability">Portability and Supported Platforms</a>
403 <!-- *********************************************************************** -->
405 <div class="doc_text">
407 <p>LLVM is known to work on the following platforms:</p>
410 <li>Intel and AMD machines (IA32) running Red Hat Linux, Fedora Core and FreeBSD
411 (and probably other unix-like systems).</li>
412 <li>PowerPC and X86-based Mac OS X systems, running 10.3 and above in 32-bit and
414 <li>Intel and AMD machines running on Win32 using MinGW libraries (native).</li>
415 <li>Intel and AMD machines running on Win32 with the Cygwin libraries (limited
416 support is available for native builds with Visual C++).</li>
417 <li>Sun UltraSPARC workstations running Solaris 10.</li>
418 <li>Alpha-based machines running Debian GNU/Linux.</li>
419 <li>Itanium-based (IA64) machines running Linux and HP-UX.</li>
422 <p>The core LLVM infrastructure uses GNU autoconf to adapt itself
423 to the machine and operating system on which it is built. However, minor
424 porting may be required to get LLVM to work on new platforms. We welcome your
425 portability patches and reports of successful builds or error messages.</p>
429 <!-- *********************************************************************** -->
430 <div class="doc_section">
431 <a name="knownproblems">Known Problems</a>
433 <!-- *********************************************************************** -->
435 <div class="doc_text">
437 <p>This section contains all known problems with the LLVM system, listed by
438 component. As new problems are discovered, they will be added to these
439 sections. If you run into a problem, please check the <a
440 href="http://llvm.org/bugs/">LLVM bug database</a> and submit a bug if
441 there isn't already one.</p>
445 <!-- ======================================================================= -->
446 <div class="doc_subsection">
447 <a name="experimental">Experimental features included with this release</a>
450 <div class="doc_text">
452 <p>The following components of this LLVM release are either untested, known to
453 be broken or unreliable, or are in early development. These components should
454 not be relied on, and bugs should not be filed against them, but they may be
455 useful to some people. In particular, if you would like to work on one of these
456 components, please contact us on the <a
457 href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev list</a>.</p>
460 <li>The MSIL, IA64, Alpha, SPU, and MIPS backends are experimental.</li>
461 <li>The llc "<tt>-filetype=asm</tt>" (the default) is the only supported
462 value for this option.</li>
467 <!-- ======================================================================= -->
468 <div class="doc_subsection">
469 <a name="x86-be">Known problems with the X86 back-end</a>
472 <div class="doc_text">
475 <li>The X86 backend does not yet support
476 all <a href="http://llvm.org/PR879">inline assembly that uses the X86
477 floating point stack</a>. It supports the 'f' and 't' constraints, but not
479 <li>The X86 backend generates inefficient floating point code when configured
480 to generate code for systems that don't have SSE2.</li>
481 <li>Win64 code generation wasn't widely tested. Everything should work, but we
482 expect small issues to happen. Also, llvm-gcc cannot build mingw64 runtime
484 to <a href="http://llvm.org/PR2255">several</a>
485 <a href="http://llvm.org/PR2257">bugs</a> in FP stackifier
490 <!-- ======================================================================= -->
491 <div class="doc_subsection">
492 <a name="ppc-be">Known problems with the PowerPC back-end</a>
495 <div class="doc_text">
498 <li>The Linux PPC32/ABI support needs testing for the interpreter and static
499 compilation, and lacks support for debug information.</li>
504 <!-- ======================================================================= -->
505 <div class="doc_subsection">
506 <a name="arm-be">Known problems with the ARM back-end</a>
509 <div class="doc_text">
512 <li>Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6
513 processors, thumb programs can crash or produce wrong
514 results (<a href="http://llvm.org/PR1388">PR1388</a>).</li>
515 <li>Compilation for ARM Linux OABI (old ABI) is supported, but not fully tested.
517 <li>There is a bug in QEMU-ARM (<= 0.9.0) which causes it to incorrectly
519 programs compiled with LLVM. Please use more recent versions of QEMU.</li>
524 <!-- ======================================================================= -->
525 <div class="doc_subsection">
526 <a name="sparc-be">Known problems with the SPARC back-end</a>
529 <div class="doc_text">
532 <li>The SPARC backend only supports the 32-bit SPARC ABI (-m32), it does not
533 support the 64-bit SPARC ABI (-m64).</li>
538 <!-- ======================================================================= -->
539 <div class="doc_subsection">
540 <a name="alpha-be">Known problems with the Alpha back-end</a>
543 <div class="doc_text">
547 <li>On 21164s, some rare FP arithmetic sequences which may trap do not have the
548 appropriate nops inserted to ensure restartability.</li>
553 <!-- ======================================================================= -->
554 <div class="doc_subsection">
555 <a name="ia64-be">Known problems with the IA64 back-end</a>
558 <div class="doc_text">
562 <li>C++ programs are likely to fail on IA64, as calls to <tt>setjmp</tt> are
563 made where the argument is not 16-byte aligned, as required on IA64. (Strictly
564 speaking this is not a bug in the IA64 back-end; it will also be encountered
565 when building C++ programs using the C back-end.)</li>
567 <li>The C++ front-end does not use <a href="http://llvm.org/PR406">IA64
568 ABI compliant layout of v-tables</a>. In particular, it just stores function
569 pointers instead of function descriptors in the vtable. This bug prevents
570 mixing C++ code compiled with LLVM with C++ objects compiled by other C++
573 <li>There are a few ABI violations which will lead to problems when mixing LLVM
574 output with code built with other compilers, particularly for floating-point
577 <li>Defining vararg functions is not supported (but calling them is OK).</li>
579 <li>The Itanium backend has bitrotted somewhat.</li>
584 <!-- ======================================================================= -->
585 <div class="doc_subsection">
586 <a name="c-be">Known problems with the C back-end</a>
589 <div class="doc_text">
592 <li><a href="http://llvm.org/PR802">The C backend has only basic support for
593 inline assembly code</a>.</li>
594 <li><a href="http://llvm.org/PR1658">The C backend violates the ABI of common
595 C++ programs</a>, preventing intermixing between C++ compiled by the CBE and
596 C++ code compiled with llc or native compilers.</li>
597 <li>The C backend does not support all exception handling constructs.</li>
603 <!-- ======================================================================= -->
604 <div class="doc_subsection">
605 <a name="c-fe">Known problems with the llvm-gcc C front-end</a>
608 <!-- _______________________________________________________________________ -->
609 <div class="doc_subsubsection">Bugs</div>
611 <div class="doc_text">
613 <p>llvm-gcc does not currently support <a href="http://llvm.org/PR869">Link-Time
614 Optimization</a> on most platforms "out-of-the-box". Please inquire on the
615 llvmdev mailing list if you are interested.</p>
617 <p>The only major language feature of GCC not supported by llvm-gcc is
618 the <tt>__builtin_apply</tt> family of builtins. However, some extensions
619 are only supported on some targets. For example, trampolines are only
620 supported on some targets (these are used when you take the address of a
621 nested function).</p>
623 <p>If you run into GCC extensions which are not supported, please let us know.
628 <!-- ======================================================================= -->
629 <div class="doc_subsection">
630 <a name="c++-fe">Known problems with the llvm-gcc C++ front-end</a>
633 <div class="doc_text">
635 <p>The C++ front-end is considered to be fully
636 tested and works for a number of non-trivial programs, including LLVM
637 itself, Qt, Mozilla, etc.</p>
640 <li>Exception handling works well on the X86 and PowerPC targets, including
641 X86-64 darwin. This works when linking to a libstdc++ compiled by GCC. It is
642 supported on X86-64 linux, but that is disabled by default in this release.</li>
648 <!-- ======================================================================= -->
649 <div class="doc_subsection">
650 <a name="ada-fe">Known problems with the llvm-gcc Ada front-end</a>
653 <div class="doc_text">
654 The llvm-gcc 4.2 Ada compiler works fairly well, however this is not a mature
655 technology and problems should be expected.
657 <li>The Ada front-end currently only builds on X86-32. This is mainly due
658 to lack of trampoline support (pointers to nested functions) on other platforms,
659 however it <a href="http://llvm.org/PR2006">also fails to build on X86-64</a>
660 which does support trampolines.</li>
661 <li>The Ada front-end <a href="http://llvm.org/PR2007">fails to bootstrap</a>.
662 Workaround: configure with --disable-bootstrap.</li>
663 <li>The c380004 and <a href="http://llvm.org/PR2010">c393010</a> ACATS tests
664 fail (c380004 also fails with gcc-4.2 mainline). When built at -O3, the
665 <a href="http://llvm.org/PR2421">cxg2021</a> ACATS test also fails.</li>
666 <li>Some gcc specific Ada tests continue to crash the compiler. The testsuite
667 reports most tests as having failed even though they pass.</li>
668 <li>The -E binder option (exception backtraces)
669 <a href="http://llvm.org/PR1982">does not work</a> and will result in programs
670 crashing if an exception is raised. Workaround: do not use -E.</li>
671 <li>Only discrete types <a href="http://llvm.org/PR1981">are allowed to start
672 or finish at a non-byte offset</a> in a record. Workaround: do not pack records
673 or use representation clauses that result in a field of a non-discrete type
674 starting or finishing in the middle of a byte.</li>
675 <li>The <tt>lli</tt> interpreter <a href="http://llvm.org/PR2009">considers
676 'main' as generated by the Ada binder to be invalid</a>.
677 Workaround: hand edit the file to use pointers for <tt>argv</tt> and
678 <tt>envp</tt> rather than integers.</li>
679 <li>The <tt>-fstack-check</tt> option <a href="http://llvm.org/PR2008">is
684 <!-- *********************************************************************** -->
685 <div class="doc_section">
686 <a name="additionalinfo">Additional Information</a>
688 <!-- *********************************************************************** -->
690 <div class="doc_text">
692 <p>A wide variety of additional information is available on the <a
693 href="http://llvm.org">LLVM web page</a>, in particular in the <a
694 href="http://llvm.org/docs/">documentation</a> section. The web page also
695 contains versions of the API documentation which is up-to-date with the
696 Subversion version of the source code.
697 You can access versions of these documents specific to this release by going
698 into the "<tt>llvm/doc/</tt>" directory in the LLVM tree.</p>
700 <p>If you have any questions or comments about LLVM, please feel free to contact
701 us via the <a href="http://llvm.org/docs/#maillist"> mailing
706 <!-- *********************************************************************** -->
710 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
711 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
712 <a href="http://validator.w3.org/check/referer"><img
713 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
715 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
716 Last modified: $Date$