<div class="doc_text">
-<p>LLVM 2.1 brings two new beta C front-ends. First, Duncan, Anton and Devang
-has started syncing up llvm-gcc with GCC 4.2, yielding "llvm-gcc 4.2" (creative,
-huh?). llvm-gcc 4.2 has the promise to bring much better FORTRAN and Ada
-support to LLVM as well as features like atomic builtins, OpenMP, and many other
-things. Check it out!</p>
+<p>LLVM 2.1 brings two new beta C front-ends. First, a new version of llvm-gcc
+based on GCC 4.2, innovatively called "llvm-gcc-4.2". This promises to bring
+FORTRAN and Ada support to LLVM as well as features like atomic builtins and
+OpenMP. None of these actually work yet, but don't let that stop you checking
+it out!</p>
<p>Second, LLVM now includes its own native C and Objective-C front-end (C++ is
in progress, but is not very far along) code named "<a
<div class="doc_text">
-<p>Some of the most noticable improvements this release have been in the
-optimizer, speeding it up and making it more aggressive</p>
+<p>Some of the most noticable feature improvements this release have been in the
+optimizer, speeding it up and making it more aggressive. For example:</p>
<ul>
-<li>Owen DSE and MemDep analysis</li>
-<li>Owen GVN</li>
-<li>Owen GVN-PRE, not in llvm-gcc</li>
+<li>Owen Anderson wrote the new MemoryDependenceAnalysis pass, which provides
+ a lazy, caching layer on top of <a
+ href="AliasAnalysis.html">AliasAnalysis</a>. He then used it to rewrite
+ DeadStoreElimination which resulted in significantly better compile time in
+ common cases, </li>
+<li>Owen implemented the new GVN pass, which is also based on
+ MemoryDependenceAnalysis. This pass replaces GCSE/LoadVN in the standard
+ set of passes, providing more aggressive optimization at a some-what
+ improved compile-time cost.</li>
+<li>Owen implemented GVN-PRE, a partial redundancy elimination algorithm that
+ shares some details with the new GVN pass. It is still in need of compile
+ time tuning, and is not turned on by default.</li>
<li>Devang merged ETForest and DomTree into a single easier to use data
-structure.</li>
+ structure. This makes it more obvious which datastructure to choose
+ (because there is only one) and makes the compiler more memory and time
+ efficient (less stuff to keep up-to-date).</li>
<li>Nick Lewycky improved loop trip count analysis to handle many more common
-cases.</li>
+ cases.</li>
</ul>
<div class="doc_text">
+<p>One of the main focuses of this release was performance tuning and bug
+ fixing. In addition to these, several new major changes occurred:</p>
+
<ul>
-<li>Dale finished up the Tail Merging optimization in the code generator,
-enabling it by default. This produces smaller code that is also faster in some
-cases.</li>
+<li>Dale finished up the Tail Merging optimization in the code generator, and
+ enabled it by default. This produces smaller code that is also faster in
+ some cases.</li>
+
+<li>Christopher Lamb implemented support for virtual register sub-registers,
+ which can be used to better model many forms of subregisters. As an example
+ use, he modified the X86 backend to use this to model truncates and
+ extends more accurately (leading to better code).</li>
<li>Dan Gohman changed the way we represent vectors before legalization,
-significantly simplifying the SelectionDAG representation for these and making
-the code generator faster for vector code.</li>
+ significantly simplifying the SelectionDAG representation for these and
+ making the code generator faster for vector code.</li>
-<li>Evan remat rewrite (coalesced intervals + folding of remat'd loads) and
-live intervals improvements.</li>
+<li>Evan contributed a new target independent if-converter. While it is
+ target independent, so far only the ARM backend uses it.</li>
-<li>Dan Gohman contributed support for better alignment and volatility handling
-in the code generator, and significantly enhanced alignment analysis for SSE
-load/store instructions.</li>
+<li>Evan rewrote the way the register allocator handles rematerialization,
+ allowing it to be much more effective on two-address targets like X86,
+ and taught it to fold loads away when possible (also a big win on X86).</li>
-<li>Christopher Lamb virtual register sub-register support, better truncates and
-extends on X86.</li>
+<li>Dan Gohman contributed support for better alignment and volatility handling
+ in the code generator, and significantly enhanced alignment analysis for SSE
+ load/store instructions. With his changes, an insufficiently-aligned SSE
+ load instruction turns into <tt>movups</tt>, for example.</li>
<li>Duraid Madina contributed a new "bigblock" register allocator, and Roman
-Levenstein contributed several big improvements. BigBlock is optimized for code
-that uses very large basic blocks. It is slightly slower than the "local"
-allocator, but produces much better code.</li>
+ Levenstein contributed several big improvements. BigBlock is optimized for
+ code that uses very large basic blocks. It is slightly slower than the
+ "local" allocator, but produces much better code.</li>
<li>David Greene refactored the register allocator to split coalescing out from
-allocation, making coalescers pluggable.</li>
+ allocation, making coalescers pluggable.</li>
</ul>
</p>
<ul>
-<li>Bruno Cardoso Lopes contributed initial MIPS support.</li>
-<li>Bill Wendling added SSSE3 support.</li>
-<li>New Target independent if converter, ARM uses it so far</li>
+<li>Bruno Cardoso Lopes contributed initial MIPS support. It is sufficient to
+ run many small programs, but is still incomplete and is not yet
+ fully performant.</li>
+
+<li>Bill Wendling added SSSE3 support to the X86 backend.</li>
+
<li>Nicholas Geoffray contributed improved linux/ppc ABI and JIT support.</li>
+
<li>Dale Johannesen rewrote handling of 32-bit float values in the X86 backend
-when using the floating point stack, fixing several nasty bugs.</li>
-<li>Dan contributed rematerialization support for the X86 backend.</li>
+ when using the floating point stack, fixing several nasty bugs.</li>
+
+<li>Dan contributed rematerialization support for the X86 backend, in addition
+ to several X86-specific micro optimizations.</li>
</ul>
</div>
</p>
<ul>
-<li>Duncan and Anton exception handling in llvm-gcc 4.0/4.2</li>
+<li>Duncan and Anton made significant progress chasing down a number of problems
+ with C++ Zero-Cost exception handling in llvm-gcc 4.0 and 4.2. It is now at
+ the point where it "just works" on linux/X86-32 and has partial support on
+ other targets.</li>
-<li>Devang and Duncan: Bitfields, pragma pack</li>
+<li>Devang and Duncan fixed a huge number of bugs relating to bitfields, pragma
+ pack, and variable sized fields in structures.</li>
-<li>Tanya implemented support for __attribute__((noinline)) in llvm-gcc, and
-added support for generic variable annotations which are propagated into the
-LLVM IR, e.g. "<tt>int X __attribute__((annotate("myproperty")));</tt>".</li>
+<li>Tanya implemented support for <tt>__attribute__((noinline))</tt> in
+ llvm-gcc, and added support for generic variable annotations which are
+ propagated into the LLVM IR, e.g.
+ "<tt>int X __attribute__((annotate("myproperty")));</tt>".</li>
<li>Sheng Zhou and Christopher Lamb implemented alias analysis support for
-'restrict' arguments to functions.</li>
+"restrict" pointer arguments to functions.</li>
-<li>Duncan contributed support for trampolines (pointers to nested functions),
-currently only supported on x86 target.</li>
+<li>Duncan contributed support for trampolines (taking the address of a nested
+ function). Currently this is only supported on the X86-32 target.</li>
-<li> Lauro Ramos Venancio contributed support to encode alignment info in
-load and store instructions.</li>
+<li>Lauro Ramos Venancio contributed support to encode alignment info in
+ load and store instructions, the foundation for other alignment-related
+ work.</li>
</ul>
</div>
</p>
<ul>
-<li>Neil Booth APFloat, foundation for long double support that will be wrapped
-up in 2.2. Dale contributed most of long double support, will be enabled in
-2.2.</li>
+<li>Neil Booth contributed a new "APFloat" class, which ensures that floating
+ point representation and constant folding is not dependent on the host
+ architecture that builds the application. This support is the foundation
+ for "long double" support that will be wrapped up in LLVM 2.2.</li>
+
+<li>Based on the APFloat class, Dale redesigned the internals of the ConstantFP
+ class and has been working on extending the core and optimizer components to
+ support various target-specific 'long double's. We expect this work to be
+ completed in LLVM 2.2.</li>
-<li>LLVM now provides an LLVMBuilder class which makes it significantly easier
-to create LLVM IR instructions.</li>
+<li>LLVM now provides an LLVMBuilder class, which makes it significantly easier
+ to create LLVM IR instructions.</li>
<li>Reid contributed support for intrinsics that take arbitrary integer typed
-arguments, Dan Gohman and Chandler extended it to support FP and vectors.</li>
-</li>
-
+ arguments. Dan Gohman and Chandler extended it to support arbitrary
+ floating point arguments and vectors.</li>
</ul>
</div>
</p>
<ul>
-<li>BrainF frontend by Sterling Stein.</li>
-
-<li>David Green contributed a new --enable-expensive-checks configure option
-which enables STL checking, and fixed several bugs exposed by it.</li>
-
-</li>
+<li>Sterling Stein contributed a new BrainF frontend, located in llvm/examples.
+ This shows a some of the more modern APIs for building a front-end, and
+ demonstrates JIT compiler support.</li>
+<li>David Green contributed a new <tt>--enable-expensive-checks</tt> configure
+ option which enables STL checking, and fixed several bugs exposed by
+ it.</li>
</ul>
</div>
components, please contact us on the <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVMdev list</a>.</p>
<ul>
-<li>The <tt>-cee</tt> pass is known to be buggy, and may be removed in in a
+<li>The <tt>-cee</tt> pass is known to be buggy, and may be removed in a
future release.</li>
-<li>C++ EH support is disabled for this release.</li>
<li>The MSIL backend is experimental.</li>
<li>The IA64 code generator is experimental.</li>
-<li>The Alpha JIT is experimental.</li>
+<li>The Alpha backend is experimental.</li>
<li>"<tt>-filetype=asm</tt>" (the default) is the only supported value for the
<tt>-filetype</tt> llc option.</li>
</ul>
<ul>
<li>The X86 backend does not yet support <a href="http://llvm.org/PR879">inline
assembly that uses the X86 floating point stack</a>.</li>
+<li>The X86 backend occasionally has <a href="http://llvm.org/PR1649">alignment
+ problems</a> on operating systems that don't require 16-byte stack alignment
+ (including most non-darwin OS's like linux).</li>
</ul>
</div>
<ul>
<li>Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6
-processors, thumb program can crash or produces wrong
+processors, thumb programs can crash or produce wrong
results (<a href="http://llvm.org/PR1388">PR1388</a>).</li>
<li>Compilation for ARM Linux OABI (old ABI) is supported, but not fully tested.
</li>
<ul>
<li><a href="http://llvm.org/PR802">The C backend does not support inline
assembly code</a>.</li>
+<li><a href="http://llvm.org/PR1126">The C backend does not support vectors
+ yet</a>.</li>
+<li><a href="http://llvm.org/PR1658">The C backend violates the ABI of common
+ C++ programs</a>, preventing intermixing between C++ compiled by the CBE and
+ C++ code compiled with LLC or native compilers.</li>
</ul>
</div>
<li><p>llvm-gcc <b>partially</b> supports these GCC extensions:</p>
<ol>
- <li><a href="http://gcc.gnu.org/onlinedocs/gcc/Nested-Functions.html#Nested%20Functions">Nested Functions</a>: As in Algol and Pascal, lexical scoping of functions.<br>
- Nested functions are supported, but llvm-gcc does not support non-local
- gotos or taking the address of a nested function.</li>
+ <li><a href="http://gcc.gnu.org/onlinedocs/gcc/Nested-Functions.html#Nested%20Functions">Nested Functions</a>:
+
+ As in Algol and Pascal, lexical scoping of functions.
+ Nested functions are supported, but llvm-gcc does not support
+ taking the address of a nested function (except on the X86-32 target)
+ or non-local gotos.</li>
<li><a href="http://gcc.gnu.org/onlinedocs/gcc/Function-Attributes.html#Function%20Attributes">Function Attributes</a>:
<b>Supported:</b> <tt>alias</tt>, <tt>always_inline</tt>, <tt>cdecl</tt>,
<tt>constructor</tt>, <tt>destructor</tt>,
<tt>deprecated</tt>, <tt>fastcall</tt>, <tt>format</tt>,
- <tt>format_arg</tt>, <tt>non_null</tt>, <tt>noreturn</tt>, <tt>regparm</tt>
+ <tt>format_arg</tt>, <tt>non_null</tt>, <tt>noinline</tt>, <tt>noreturn</tt>, <tt>regparm</tt>
<tt>section</tt>, <tt>stdcall</tt>, <tt>unused</tt>, <tt>used</tt>,
<tt>visibility</tt>, <tt>warn_unused_result</tt>, <tt>weak</tt><br>
- <b>Ignored:</b> <tt>noinline</tt>, <tt>pure</tt>, <tt>const</tt>, <tt>nothrow</tt>,
+ <b>Ignored:</b> <tt>pure</tt>, <tt>const</tt>, <tt>nothrow</tt>,
<tt>malloc</tt>, <tt>no_instrument_function</tt></li>
</ol>
</li>
itself, Qt, Mozilla, etc.</p>
<ul>
-<li>llvm-gcc4 only has partial support for <a href="http://llvm.org/PR870">C++
-Exception Handling</a>, and it is not enabled by default.</li>
+<li>Exception handling only works well on the linux/X86-32 target.
+In some cases, illegally throwing an exception does not result
+in a call to terminate.</li>
<!-- NO EH Support!