This document contains the release notes for the LLVM Compiler Infrastructure, release 2.4. Here we describe the status of LLVM, including major improvements from the previous release and significant known problems. All LLVM releases may be downloaded from the LLVM releases web site.

For more information about LLVM, including information about the latest release, please check out the main LLVM web site. If you have questions or comments, the LLVM Developer's Mailing List is a good place to send them.

Note that if you are reading this file from a Subversion checkout or the main LLVM web page, this document applies to the next release, not the current one. To see the release notes for a specific releases, please see the releases page.

The LLVM 2.4 distribution currently consists of code from the core LLVM repository (which roughly includes the LLVM optimizers, code generators and supporting tools) and the llvm-gcc repository. In addition to this code, the LLVM Project includes other sub-projects that are in development. The two which are the most actively developed are the Clang Project and the vmkit Project.

The Clang project is an effort to build a set of new 'LLVM native' front-end technologies for the LLVM optimizer and code generator. Clang is continuing to make major strides forward in all areas. Its C and Objective-C parsing support is very solid, and the code generation support is far enough along to build many C applications. While not yet production quality, it is progressing very nicely. In addition, C++ front-end work has started to make significant progress.

Codegen progress/state [DANIEL]

The Clang project also includes an early stage static source code analysis tool for automatically finding bugs in C and Objective-C programs. The tool performs a growing set of checks to find bugs that occur on a specific path within a program. Examples of bugs the tool finds include logic errors such as null dereferences, violations of various API rules, dead code, and potential memory leaks in Objective-C programs. Since its inception, public feedback on the tool has been extremely positive, and conservative estimates put the number of real bugs it has found in industrial-quality software on the order of thousands.

The tool also provides a simple web GUI to inspect potential bugs found by the tool. While still early in development, the GUI illustrates some of the key features of Clang: accurate source location information, which is used by the GUI to highlight specific code expressions that relate to a bug (including those that span multiple lines) and built-in knowledge of macros, which is used to perform inline expansion of macros within the GUI itself.

The set of checks performed by the static analyzer is gradually expanding, and future plans for the tool include full source-level inter-procedural analysis and deeper checks such as buffer overrun detection. There are many opportunities to extend and enhance the static analyzer, and anyone interested in working on this project is encouraged to get involved!

The vmkit project is an implementation of a JVM and a CLI Virtual Machines (Microsoft .NET is an implementation of the CLI) using the Just-In-Time compiler of LLVM.

[NICOLAS]

This release includes a huge number of bug fixes, performance tweaks and minor improvements. Some of the major improvements and new features are listed in this section.

LLVM 2.4 includes several major new capabilities:

• The most visible end-user change in LLVM 2.4 is that it includes many optimizations and changes to make -O0 compile times much faster. You should see improvements on the order of 30% or more faster than LLVM 2.3. There are many pieces to this change, described in more detail below. The speedups and new components can also be used for JIT compilers that want fast compilation as well.

• The biggest change to the LLVM IR is that Multiple Return Values (which were introduced in LLVM 2.3) have been generalized to full support for "First Class Aggregate" values in LLVM 2.4. This means that LLVM IR supports using structs and arrays as values in a function. This capability is mostly useful for front-end authors, who prefer to treat things like complex numbers, simple tuples, dope vectors, etc as Value*'s instead of as a tuple of Value*'s or as memory values.

• LLVM 2.4 also includes an initial port for the PIC16 microprocessor. This is the LLVM targer that only has support for 8 bit registers, and a number of other crazy constraints. While the port is still in early development stages, it shows some interesting things you can do with LLVM.

LLVM fully supports the llvm-gcc 4.2 front-end, which marries the GCC front-ends and driver with the LLVM optimizer and code generator. It currently includes support for the C, C++, Objective-C, Ada, and Fortran front-ends.

• LLVM 2.4 supports the full set of atomic __sync_* builtins. LLVM 2.3 only supported those used by OpenMP, but 2.4 supports them all. While llvm-gcc supports all of these builtins, note that not all targets do. X86 support them all in both 32-bit and 64-bit mode and PowerPC supports them all except for the 64-bit operations when in 32-bit mode.
• llvm-gcc now supports an -flimited-precision option, which tells the compiler that it is ok to use low-precision approximations of certain libm functions (like tan, log, etc). This allows you to get high performance if you only need (say) 14-bits of precision.
• llvm-gcc now supports a C language extension known as "Blocks . This feature is similar to nested functions and closures, but does not require stack trampolines (with most ABIs) and supports returning closures from functions that define them. Note that actually using Blocks requires a small runtime that is not included with llvm-gcc.
• llvm-gcc now supports a new -flto option. On systems that support transparent Link Time Optimization (currently Darwin systems with Xcode 3.1 and later) this allows the use of LTO with other optimization levels like -Os. Previously, LTO could only be used with -O4, which implied optimizations in -O3 that can increase code size.

New features include:

• A major change to the "Use" class landed, which shrank it by 25%. Since this is a pervasive part of the LLVM, it ended up reducing the memory use of LLVM IR in general by 15% for most programs.
• Values with no names are now pretty printed by llvm-dis more nicely. They now print as "%3 = add i32 %A, 4" instead of "add i32 %A, 4 ; <i32>:3", which makes it much easier to read.
• LLVM 2.4 includes some changes for better vector support. First, the shift operations (shl, ashr, lshr) now all support vectors and do an element-by-element shift (shifts of the whole vector can be accomplished by bitcasting the vector to <1 x i128> for example). Second, there is initial support in development for vector comparisons with the fcmp/icmp instructions. These instructions compare two vectors and return a vector of i1's for each result. Note that there is very little codegen support available for any of these IR features though.
• A new DebugInfoBuilder class is available, which makes it much easier for front-ends to create debug info descriptors, similar to the way that IRBuilder makes it easier to create LLVM IR.
• LLVM now supports "function attributes", which allows us to seperate return value attributes from function attributes. LLVM now supports attributes on a function itself, a return value, and its parameters. New supported function attributes include noinline/alwaysinline and the "opt-size" flag which says the function should be optimized for code size.
• LLVM IR now directly represents "common" linkage, instead of representing it as a form of weak linkage.

In addition to a huge array of bug fixes and minor performance tweaks, this release includes a few major enhancements and additions to the optimizers:

• The Global Value Numbering (GVN) pass now does local Partial Redundancy Elimination (PRE) to eliminate some partially redundant expressions in cases where doing so won't grow code size.
• LLVM 2.4 includes a new loop deletion pass (which removes output-free provably-finite loops) and a rewritten Aggressive Dead Code Elimination (ADCE) pass that no longer uses control dependence information. These changes speed up the optimizer and also prevents it from deleting output-free infinite loops.
• The new AddReadAttrs pass works out which functions are read-only or read-none (these correspond to 'pure' and 'const' in C) and marks them with the appropriate attribute.
• LLVM 2.4 now includes a new SparsePropagation framework, which makes it trivial to build lattice-based dataflow solvers that operate over LLVM IR. Using this interface means that you just define objects to represent your lattice values and the transfer functions that operate on them. It handles the mechanics of worklist processing, liveness tracking, handling PHI nodes, etc.
• The Loop Strength Reduction and induction variable optimization passes have several improvements to avoid inserting MAX expressions, to optimize simple floating point induction variables and to analyze trip counts of more loops.
• Various helper functions (ComputeMaskedBits, ComputeNumSignBits, etc) were pulled out of the Instruction Combining pass and put into a new ValueTracking.h header, where they can be reused by other passes.
• The tail duplication pass has been removed from the standard optimizer sequence used by llvm-gcc. This pass still exists, but the benefits it once provided are now achieved by other passes.

We put a significant amount of work into the code generator infrastructure, which allows us to implement more aggressive algorithms and make it run faster:

• The target-independent code generator supports (and the X86 backend currently implements) a new interface for "fast" instruction selection. This interface is optimized to produce code as quickly as possible, sacrificing code quality to do it. This is used by default at -O0 or when using "llc -fast" on X86. It is straight-forward to add support for other targets if faster -O0 compilation is desired.
• In addition to the new 'fast' instruction selection path, many existing pieces of the code generator have been optimized in significant ways. SelectionDAG's are now pool allocated and use better algorithms in many places, the ".s" file printers now use raw_ostream to emit text much faster, etc. The end result of these improvements is that the compiler also takes substantially less time to generate code that is just as good (and often better) than before.
• Each target has been split to separate the .s file printing logic from the rest of the target. This enables JIT compilers that don't link in the (somewhat large) code and data tables used for printing a .s file.
• The code generator now includes a "stack slot coloring" pass, which packs together individual spilled values into common stack slots. This reduces the size of stack frames with many spills, which tends to increase L1 cache effectiveness.
• Various pieces of the register allocator (e.g. the coalescer and two-address operation elimination pass) now know how to rematerialize trivial operations to avoid copies and include several other optimizations.
• The graphs produced by the llc -view-*-dags options are now significantly prettier and easier to read.
• LLVM 2.4 includes a new register allocator based on Partitioned Boolean Quadratic Programming (PBQP). This register allocator is still in development, but is very simple and clean.

New target-specific features include:

• Exception handling is supported by default on Linux/x86-64.
• Position Independent Code (PIC) is now supported on Linux/x86-64.
• @llvm.frameaddress now supports getting the frame address of stack frames > 0 on x86/x86-64.
• MIPS floating point support? [BRUNO]
• The PowerPC backend now supports trampolines.

New features include:

• llvmc2 (the generic compiler driver) gained plugin support. It is now easier to experiment with llvmc2 and build your own tools based on it.
• LLVM 2.4 includes a number of new generic algorithms and data structures, include a scoped hash table, 'immutable' data structures, and a raw_ostream class. The raw_ostream class and format allow for efficient file output, and various pieces of LLVM have switched over to use it. The eventual goal is to eliminate std::ostream in favor of it.
• Recycler + pool allocation stuff? [DAN]

If you're already an LLVM user or developer with out-of-tree changes based on LLVM 2.3, this section lists some "gotchas" that you may run into upgrading from the previous release.

• The LLVM IR generated by llvm-gcc no longer names all instructions. This makes it run faster, but may be more confusing to some people. If you prefer to have names, the 'opt -instnamer' pass will add names to all instructions.
• The LoadVN and GCSE passes have been removed from the tree. They are obsolete and have been replaced with the GVN and MemoryDependence passes.

In addition, many APIs have changed in this release. Some of the major LLVM API changes are:

• Attributes changes [DEVANG]
• The DbgStopPointInst methods getDirectory and getFileName now return Value* instead of strings. These can be converted to strings using llvm::GetConstantStringInfo defined via "llvm/Analysis/ValueTracking.h".
• The APIs to create various instructions have changed from lower case "create" methods to upper case "Create" methods (e.g. BinaryOperator::create). LLVM 2.4 includes both cases, but the lower case ones are removed in mainline, please migrate.
• Various header files like "llvm/ADT/iterator" were given a .h suffix. Change your code to #include "llvm/ADT/iterator.h" instead.
• In the code generator, many MachineOperand predicates were renamed to be shorter (e.g. isFrameIndex() -> isFI()), SDOperand was renamed to SDValue (and the "Val" member was changed to be the getNode() accessor), and the MVT::ValueType enum has been replaced with an "MVT" struct.

LLVM is known to work on the following platforms:

• Intel and AMD machines (IA32) running Red Hat Linux, Fedora Core and FreeBSD (and probably other unix-like systems).
• PowerPC and X86-based Mac OS X systems, running 10.3 and above in 32-bit and 64-bit modes.
• Intel and AMD machines running on Win32 using MinGW libraries (native).
• Intel and AMD machines running on Win32 with the Cygwin libraries (limited support is available for native builds with Visual C++).
• Sun UltraSPARC workstations running Solaris 10.
• Alpha-based machines running Debian GNU/Linux.
• Itanium-based (IA64) machines running Linux and HP-UX.

The core LLVM infrastructure uses GNU autoconf to adapt itself to the machine and operating system on which it is built. However, minor porting may be required to get LLVM to work on new platforms. We welcome your portability patches and reports of successful builds or error messages.

This section contains all known problems with the LLVM system, listed by component. As new problems are discovered, they will be added to these sections. If you run into a problem, please check the LLVM bug database and submit a bug if there isn't already one.

The following components of this LLVM release are either untested, known to be broken or unreliable, or are in early development. These components should not be relied on, and bugs should not be filed against them, but they may be useful to some people. In particular, if you would like to work on one of these components, please contact us on the LLVMdev list.

• The MSIL, IA64, Alpha, SPU, and MIPS backends are experimental.
• The llc "-filetype=asm" (the default) is the only supported value for this option.

• The X86 backend does not yet support all inline assembly that uses the X86 floating point stack. It supports the 'f' and 't' constraints, but not 'u'.
• The X86 backend generates inefficient floating point code when configured to generate code for systems that don't have SSE2.
• Win64 code generation wasn't widely tested. Everything should work, but we expect small issues to happen. Also, llvm-gcc cannot build mingw64 runtime currently due to several bugs due to lack of support for the 'u' inline assembly constraint and X87 floating point inline assembly.
• The X86-64 backend does not yet support the LLVM IR instruction va_arg. Currently, the llvm-gcc front-end supports variadic argument constructs on X86-64 by lowering them manually.

• The Linux PPC32/ABI support needs testing for the interpreter and static compilation, and lacks support for debug information.

• Thumb mode works only on ARMv6 or higher processors. On sub-ARMv6 processors, thumb programs can crash or produce wrong results (PR1388).
• Compilation for ARM Linux OABI (old ABI) is supported, but not fully tested.
• There is a bug in QEMU-ARM (<= 0.9.0) which causes it to incorrectly execute programs compiled with LLVM. Please use more recent versions of QEMU.

• The SPARC backend only supports the 32-bit SPARC ABI (-m32), it does not support the 64-bit SPARC ABI (-m64).

• On 21164s, some rare FP arithmetic sequences which may trap do not have the appropriate nops inserted to ensure restartability.

• The Itanium backend is highly experimental, and has a number of known issues. We are looking for a maintainer for the Itanium backend. If you are interested, please contact the llvmdev mailing list.

• The C backend has only basic support for inline assembly code.
• The C backend violates the ABI of common C++ programs, preventing intermixing between C++ compiled by the CBE and C++ code compiled with llc or native compilers.
• The C backend does not support all exception handling constructs.

llvm-gcc does not currently support Link-Time Optimization on most platforms "out-of-the-box". Please inquire on the llvmdev mailing list if you are interested.

The only major language feature of GCC not supported by llvm-gcc is the __builtin_apply family of builtins. However, some extensions are only supported on some targets. For example, trampolines are only supported on some targets (these are used when you take the address of a nested function).

If you run into GCC extensions which are not supported, please let us know.

The C++ front-end is considered to be fully tested and works for a number of non-trivial programs, including LLVM itself, Qt, Mozilla, etc.

• Exception handling works well on the X86 and PowerPC targets. Currently only linux and darwin targets are supported (both 32 and 64 bit).

The llvm-gcc 4.2 Ada compiler works fairly well, however this is not a mature technology and problems should be expected.

• The Ada front-end currently only builds on X86-32. This is mainly due to lack of trampoline support (pointers to nested functions) on other platforms, however it also fails to build on X86-64 which does support trampolines.
• The Ada front-end fails to bootstrap. Workaround: configure with --disable-bootstrap.
• The c380004, c393010 and cxg2021 ACATS tests fail (c380004 also fails with gcc-4.2 mainline).
• Some gcc specific Ada tests continue to crash the compiler.
• The -E binder option (exception backtraces) does not work and will result in programs crashing if an exception is raised. Workaround: do not use -E.
• Only discrete types are allowed to start or finish at a non-byte offset in a record. Workaround: do not pack records or use representation clauses that result in a field of a non-discrete type starting or finishing in the middle of a byte.
• The lli interpreter considers 'main' as generated by the Ada binder to be invalid. Workaround: hand edit the file to use pointers for argv and envp rather than integers.
• The -fstack-check option is ignored.

A wide variety of additional information is available on the LLVM web page, in particular in the documentation section. The web page also contains versions of the API documentation which is up-to-date with the Subversion version of the source code. You can access versions of these documents specific to this release by going into the "llvm/doc/" directory in the LLVM tree.

If you have any questions or comments about LLVM, please feel free to contact us via the mailing lists.